AI agents in finance are really powerful when they have the right context, controls, and operating environment.

I (OnlyCFO) have been working closely with Brex for over 2 years, and what they’ve built for agentic finance (AF) is incredible. With Brex, agents handle all the tedious stuff, like expense reports, policy enforcement, and month-end close, so your team doesn’t have to. Check out Brex AF!

Check out Brex AF

“Tokenmaxxing” is Dead

AI token spend is quickly becoming one of the largest vendor expenses, only behind salaries and maybe facilities.

So one of the highest ROI activities a CFO can do right now is to evaluate how the company can be more efficient with AI token spend. Not necessarily to cut AI usage, but to be more efficient with it.

I have talked to several engineers, CTOs, and people much smarter than me to develop this list on where to save on AI spend. I am biased but I think every CFO (and other leaders) should understand the ways to reduce AI model costs below.

1. Right Model for the Task

To understand the importance of the “right” AI model, you need to know the cost differences in these models.

If we use Claude as an example (the others are similar), the “smartest” AI model is 5x+ more expensive than the cheapest model. That means companies could be paying up to 5x more than necessary for many tasks.

The meme below summarizes the problem well…

There are LOTS of cases where the latest model is overkill for the task. If you tell Claude to use Opus 4.8 to rename files or update formatting in a board deck, it will.

This is where “model routing” comes in.

Model Routing = For each AI request, it decides which model should handle the task.

Model routing can be as simple as setting up some basic rules that define “IF this task type THEN use this AI model”. Or it can get as complicated as using an actual AI model to determine what AI model should handle the task.

Awesome! So how do I start model routing?

There are two different ways to use an AI model

1. Directly in App (Claude, ChatGPT, etc) - this is how most internal teams are using AI models.

2. Via API Calls - through another platform that sits between the user and the AI model. If you are selling an AI product, then it’s via API calls to the AI models.

Model routing basically only works if done via API calls. But many people (including me) wrongly assumed model routing wasn’t available for internal use, which is where a lot of AI spend occurs. That is not correct.

Company-Level Model Router: At a company level, you can add an “AI gateway” that sits between the AI models and the user. Instead of directly using Claude/ChatGPT, the AI gateway interface is where AI chats happen and it uses API calls. The AI gateway can then run your model router. You just need to build or buy an AI gateway…

Verticalized Model Router: Having a good model router is obviously important because you don’t want hard tasks going to a cheap model and screwing things up. So some model routers are built for types of people (like software developers). Maybe someday we’ll have model routers for every department...

Model Routing Savings Example:

Let’s say you’re spending $100K/month on Claude:

You currently use Opus 4.8 (most expensive model) for everything, but if the company only used the model it actually needed for the task then it could look something like the below:

Opus is rarely needed for every task. So if users shifted to using Sonnet and Haiku when appropriate, the company could save nearly 50% ($576K per year)…

Model routing is getting quite good so the risk of routing to an incapable model can be pretty low if you set it up correctly.

2. Prompt caching:

I am not going to get deep into this because the important thing is for you to simply ask engineering “Are we prompt caching on repetitive tasks?”

If you’re sending the same large context on every API call (a system prompt, a reference document, a knowledge base), then you’re paying full input price every single time. Prompt caching stores that content so repeat calls cost ~90% less on the cached portion.

This is where data visibility becomes so important. If you can track token spend per API call then you can see what is burning a ton of tokens and often the recurring token-heavy tasks can be reduced a lot by caching.

3. AI Cost Visibility & Reporting

AI costs will soon be your largest vendor expense (if not already). As such, it’s important to be able to drill down into the costs and understand the drivers….

A couple of the big unlocks that come from AI spend reporting:

• AI spend by employee by AI model.

  • Are employees using the most expensive model for simple tasks?

  • Help with forecasting AI spend by department

  • Are you allocating AI costs properly? AI spend needs to go to the right departments so department leaders are incentivized to control it.

• AI spend per API call

  • What products/features are driving the highest AI spend?

  • How much does the above vary by customer?

  • How should we forecast gross margins?

4. Batch API

If a task doesn’t need a real-time response, you can use the AI model’s “batch API” and save ~50% on token costs. A lot of stuff doesn’t need to be real-time so figure out where batching is fine because it can save significant money.

We did this exact thing with our ASC 606 agent. Our agent was reviewing every single contract as it was closed-won in SFDC. It reads our revenue policy, compares it to the contract terms, reviews our list of concerns, and then it documents the treatment and notifies us of any issues. We didn’t need this real-time. It was processing lots of contracts every day. So we batched them at the end of each day. And we stacked cost savings with prompt caching (#2 above) as well. We saved over 50% on the costs for this somewhat expensive AI task. And it took <1 hour to do.

5. AI Admin Controls

• Set the default to the middle-tier model (e.g. Sonnet for Claude). If you are not using an internal model router, then at least do this. 90%+ of users will just use the default model. While they usually won’t be using the most expensive model, they will also almost never downgrade to the cheapest model. Just doing this will save you up to 40% (price difference between Opus and Sonnet) across most teams.

• User Spend Limits: Put controls on users so they don’t create something that just blows up AI usage. Uber just limited developers to $1,500/month in token spend after they blew through their full-year AI budget in 4 months…Set a limit, even if it’s high, so nothing completely blows up. Then you can start to lower the spend limit later if you want.

6. Employee Education

Make sure employees understand this stuff. Most employees just don’t know how to be efficient with AI spend yet. Some examples (in addition to everything above):

• New topic = new chat. Each time you prompt in the same chat window, it re-reads the entire chat history. This gets expensive quick (especially if you are on a premium model). I have seen people just default to the same chat window every time they go to AI…

• Be upfront with everything you need in one prompt. Back-and-forth exchanges compound quickly. Every reply re-reads the full history. Give Claude the full context and ask for exactly what you want in one shot.

• Update the AI model in your automations, agents, skills, etc. A lot of scheduled tasks are structured and repetitive (often fine for a cheaper model). Users can define which AI model to use in agents/skills/automations. Test out a lower tier model and see if it works fine. This is like model routing but it’s manual since there is no automatic solution.

• Output token discipline. Output (AI responses) costs 5x input (user prompts). Asking for detailed responses, not capping responses, letting agents narrate every step will all multiply the expensive side of the bill. Telling models to be concise and setting hard output limits is close to free and underrated.

Final Thoughts

Every CIO and CFO I have talked to recently said the same thing about “tokenmaxxing”:

Tokenmaxxing served its purpose. We wanted to push AI adoption as fast as possible and get employees using the tools that would get us ahead.

Now it’s time for us to drive efficiency in that spend and ensure we are creating real ROI.

It’s time to be efficient with AI. Not slow people down necessarily. But be efficient. Start being efficient now before you have to start making harder cuts (like people or setting hard token limits).

The goal is not to cut AI usage, but to be more efficient with it and drive real ROI.

Footnotes:

• Check out Magpie to see the kind of AI visibility and reporting Brex is building and for a deeper look into the complexities of AI cost analysis.

• I cover a lot more AI stuff under my new newsletter (CFOpilot). Subscribe

• Subscribe and share this post with your finance teams. They need to understand this stuff.

Subscribe now

AI agents in finance are really powerful when they have the right context, controls, and operating environment.

I (OnlyCFO) have been working closely with Brex for over 2 years, and what they’ve built for agentic finance (AF) is incredible. With Brex, agents handle all the tedious stuff, like expense reports, policy enforcement, and month-end close, so your team doesn’t have to. Check out Brex AF!

Check out Brex AF

What is “ARR” in 2026?

There is a lot of deception/fraud in “ARR” numbers being reported today. While ARR isn’t governed by formal accounting standards like revenue, there are common practices that should be followed. If not followed, then you are either trying to deceive investors or you don’t understand basic financial stuff. Both are bad…

Let’s start with what “ARR” stands for.

Traditional ARR = Annual Recurring Revenue

2026 ARR = Annual Run-Rate Revenue

With AI, new pricing models, contractual changes, etc. the first “R” in ARR feels a little misleading so I think most companies should think about it as a run-rate amount, but everyone still simplifies it to “ARR”.

This post outlines my views on ARR in an AI world and how you should approach defining it at your company.

There are three primary questions that need to be answered:

1. What counts as ARR?

2. How do you report ARR?

3. How do you calculate ARR?

1. What Counts as ARR

There should be a strong correlation between the following:

1. Current “ARR”

2. GAAP revenue, billings, and cash collections over the next 12 months

A greater divergence or risk in that correlation increases the odds that you need either 1) an ARR policy change or 2) separate disclosure of that piece of ARR.

*GAAP (or generally accepted accounting principles) contains the rules that govern how revenue must be recognized (and also lots of other accounting rules)

How ARR Should Convert to Billings, Revenue and Cash

Let’s look at a single customer that closed an annual deal on Jan 1st for $1,200. In January there is $1,200 of ARR for that customer. If we billed upfront then we would bill them once for $1,200 and then maybe collect the cash the subsequent month. And then we should recognize $100 per month for the next 12 months.

ARR → Billings → Cash → Revenue

The expectation is that ARR is the baseline of what GAAP revenue should be over the next 12 months. And that should be billed/cash collected for that same amount.

How Being Aggressive on ARR Backfires

• Lose investors’ or prospective investors’ trust, obviously.

• Metrics all look much better than they are. Maybe good for tricking investors for fundraising, but bad for building long-term trust (which you need). Plus once “ARR” growth slows down (and it will), the metrics are going to look much worse than they should since you pulled ARR forward into earlier periods.

• Hiring and spending on inflated metrics. Related to the above. A lot of companies make spending decisions based on ARR-related metrics (new ARR, NRR, GTM efficiency metrics, etc). If those are wrong, then bad decisions will be made.

• If ARR is aggressive then churn will be higher as well. If you are aggressive on what gets counted as ARR then you must realize that the only way off the ARR waterfall is churn… High churn numbers are going to make your business look like it’s not durable (AI product is fake or your product is being commoditized by the AI models).

Spotting ARR Lies

If I want to quickly check for ARR deception, then I will compare their ARR number to a simple GAAP revenue estimate that I perform based on ARR.

It won’t be exact, but if the difference is too large and it isn’t clear why, then that will be a major red flag for me.

There are two primary components of my estimate:

1. Baseline GAAP Revenue: the expected revenue from starting ARR. This is what revenue theoretically should be if there were no changes (no new business, expansion, or churn)

2. New Revenue: revenue derived from new deals

The example below illustrates how I would estimate revenue from ARR and why it may be a red flag:

If a company provided me the above ARR and revenue numbers and gave me no further context, I would be very skeptical.

There are a lot of reasons that my simple forecast might be materially off, but most of them mean that there are lies/fraud/aggressiveness in ARR definitions.

Linearity/Seasonality

This is a valid reason to have a difference, especially if the company is growing crazy fast relative to its base ARR.

Revenue is recognized over time so if 90% of deals are closed in the last quarter, last month in the quarter, or last day in the month then ARR will diverge from a simple ARR forecast. But you could put some basic assumptions in here as well if you have the detail.

Counting Bad Deals

Every company has some bad deals that should have never been booked, but these can get large if the company is super aggressive. And they will show up in the gap between ARR and revenue.

If a customer invoice (“billings”) isn’t paid, then it shows up in one of two ways on the company’s financials

1. Reversal of revenue (e.g. CSM says “don’t pay we made mistakes”)

2. Bad debt expense is recorded (e.g. customer goes out of business)

There are accounting rules that supposedly determine which one should occur based on the scenario, but it can certainly be grey.

So if the company has a lot of #1, then there is going to be a large gap between ARR and revenue. And large revenue reversals likely mean a lot of ARR is fake.

But also…even if ARR lines up with revenue you may have a problem if bad debt expense is massive. It just means the company is burying the problem in expenses because they want revenue to be higher (since investors care more about revenue than some G&A expense)

Cash don’t lie so if ARR isn’t converting to cash then that’s an ARR definition problem.

Gap between ARR Date and Rev Rec Start Date

My general rule is if the time between contract closing and revenue recognition start date is <14 days then it can be ARR on contract close. If >14 days then wait to count the ARR until the revenue recognition start date (which is often the start date of the contract).

I have seen companies start counting ARR immediately when they had a 6-month delay in contract start date because the company wasn’t ready to move forward yet.

That contract will get $0 in revenue, billings, and cash for 6 months…

Ramping Deals

Just count this ARR when the ramp starts. It’s contracted upsells. Why count that upfront before the upsell and related costs kick in? It screws up your metrics.

A lot of companies just average it into ARR (TCV/contracted # of months) but that would recognize too much ARR upfront and not enough at the end of the contract.

Contracted ARR (CARR) vs ARR

This is the “fraud” that many people refer to in regard to ARR at hot new AI companies. They are reporting a very inflated number.

Contracted ARR = future contracted value

If you have a deal that is $50K in year 1 and $150K in year 2, then contracted ARR would be $150K. That is a 3x reality distortion from whatever an investor assumes “ARR” is. And that $100K extra ARR in year 1 won’t convert to revenue or cash.

I think “contracted ARR” can be an interesting data point, but do not mix it with ARR…disclose it completely separately if you want.

“Experimental ARR”

An existing product that has 1+ year contracts is very different than a new product with new technology that is charged month-to-month or on a pure usage basis.

A lot of new AI products have a high churn rate.

Maybe this can get included in the “ARR” bucket, but you should at least report it separately in your ARR waterfall to your board if the characteristics are materially different than the rest of ARR (see reporting section below).

Usage Overages

I only include overages in ARR if I am confident the overages will reoccur for the foreseeable future. Spikes in usage happen and then customers figure out how to control them.

An AI consultant told Axios that one customer accidentally spent $500M in a single month after they forgot to put usage limits in place. That would be…6 billion dollars that Anthropic could have counted as “ARR” 🤣. If $500M is accurate (big if), then certainly the entire thing wasn’t “accidental”. Likely some portion was a min commitment or had a trended growth over the past few months. And then the rest was a one-time spike. That spike certainly won’t reoccur so hopefully Anthropic didn’t book $6B in ARR related to that customer…

Trials/POC

Not ARR. Please don’t count free trials as ARR based on some estimate or contractual amount that the customer has to opt in to at the end of the trial period.

What about paid trials?

This can be somewhat grey in my opinion…If fully paid and you have a history that 90%+ convert then maybe I would be OK with it as long as the ARR amount tied to it is separately disclosed.

Because guess what? You have to be comfortable with the other side of the double-edged sword. Are you willing to add it to ARR and then recognize a high amount of churn soon after?

One-Time Stuff

Not ARR. Duh. Surprising how often people try to sneak some of this stuff into ARR though…

Other Red Flags That Your ARR Definition is Broken

• Lots of debookings. If you have a lot of debookings (i.e. you reduce net new ARR in the current period for prior period booked), then you probably have an ARR definition problem.

• High churn <6 months after deals close. This usually just means the sales leader bullies the CS leader to count it as churn so it doesn’t impact his/her team’s numbers. If lots of stuff is being marked as churn, expected to churn, and/or is not paying then you got a problem.

• Previously reported ARR numbers constantly change. Have seen several times where companies go back and adjust prior ARR numbers and expect me to not notice. They don’t want to count it as churn or debooking in the current period and impact growth rates so they adjust prior period numbers to make it look like the ARR was never recognized.

2. How to Report ARR

How you report ARR is important for decision making and preventing misunderstanding of what ARR is.

My Big Rule: Always define ARR wherever it’s presented. Changes should be infrequent.

In traditional SaaS, we could usually report ARR as one number because the characteristics of all revenue were mostly the same - highly predictable, durable revenue.

But ARR ain’t what it used to be….

• Pricing models are much more complicated: lots of flavors of usage-based pricing (pure usage, hybrid, etc), dealing with overages, etc

• Contractual terms create more variability - pure usage-based billing, min commitments, etc

• Gross margins across products (particularly AI) can vary widely. An 80% SaaS gross margin product obviously has very different unit economics than a 40% AI product.

• High retention is often no longer expected. And this is reinforced by shorter contract durations. Old software was 1+ year contracts (often longer) with fixed pricing… This made using ARR as a starting baseline for next year’s billing/revenue easy.

This increased complexity requires more detail in your ARR waterfall. I have never heard a board member or investor complain about too much ARR detail.

Example: Mostly Seat/Platform Fixed Pricing with Some Usage

You should break out usage-based ARR from contractually fixed so it’s clear how much unpredictability exists in ARR.

The table below breaks out usage-based increases and decreases. You could then further break that out by committed usage vs pure usage-based, by product if gross margins vary wildly, etc.

3. How to Calculate ARR

If you read this far hoping I would tell you exactly how to calculate ARR, you are going to be disappointed.

There are lots of different ARR calculation methodologies that companies may adopt. And often companies will have different calculations depending on the product and how it is priced and contracted. Below are a few common ways.

• MRR * 12

• Last month’s GAAP revenue * 12

• Trailing [X] month average MRR * 12

• TCV/contract length x 12

• Variation of one of the above less [some portion of variable revenue not expected to reoccur]

• Annualized contractual commitments (for usage-based)

I don’t think any of these are necessarily wrong and if you follow the principles I laid out in the previous two steps then you will likely land on a decent ARR calculation.

TLDR: Please Don’t Lie on ARR

If you learned nothing else, then at least do these two things:

1. Clearly define ARR every time you report it

2. Be consistent in your definition

I know there is a lot of pressure to show growth right now, but cheating with your numbers is not the way to do it. Be honest and transparent.

If you want to show the ARR potential with trials, future expected ARR with ramping deals, or something like that….great! Make a separate slide that shows those numbers. But make sure you’re clear with what you are presenting.

Footnotes:

• See how Agentic Finance can transform your finance department (today’s sponsor is Brex).

• Subscribe and share the OnlyCFO newsletter with your teams

Subscribe now

AI agents in finance are really powerful when they have the right context, controls, and operating environment.

I (OnlyCFO) have been working closely with Brex for over 2 years, and what they’ve built for agentic finance (AF) is incredible. With Brex, agents handle all the tedious stuff, like expense reports, policy enforcement, and month-end close, so your team doesn’t have to. Check out Brex AF!

Check out Brex AF

Entrata Files to IPO

Entrata filed their S-1 this past week and folks were surprised that they hadn’t heard of this hot new AI company before. Oh wait, Entrata is property management software…

There has been A LOT of excitement around the hot AI IPOs (SpaceX, OpenAI, Anthropic, etc), but most of us didn’t have Entrata on our list of companies to lead out the massive backlog of software IPOs. We have been constantly told over the last year that software is dead…

But Entrata seems alive and well! Not incredible metrics (particularly compared to all the AI stuff we constantly hear about), but really strong metrics compared to most software companies at their scale.

I know that a lot of software companies will be anxiously watching to see if Entrata will open the IPO floodgates for the hundreds of other “pretty good” private companies hoping to IPO in the near future.

Let’s dig into Entrata…the financials and story are really interesting.

1. Entrata’s Financial & Metrics

2. Estimating Entrata’s Valuation

3. Entrata’s Unique Fundraising History

How Does Entrata’s Financials Stack Up?

Entrata is a great business. They scaled without real venture capital money so they have always been efficient. So like most bootstrapped businesses, their growth rates have been more modest.

*Yellow bars are Entrata

Revenue

Revenue Scale

Revenue scale is important context. I see too many people say we are growing just as fast as X company but they are 10% the revenue scale…

Revenue Growth Rate

Entrata has a good growth rate (21st out of 73 companies). The market has definitely shifted to being more growth-focused again.

The general sentiment is that if growth isn’t strong, durable, and maybe even accelerating, then you are an AI loser. And AI losers have a terminal value of near $0.

Revenue Endurance

Revenue endurance measures how persistent revenue growth rates year-over-year. It is calculated as follows:

Entrata may not have had rocket ship levels of growth, but it has consistently just kept growing revenue at a good rate for a really long time. Entrata has a better growth endurance rate than almost any public software company over the past 5 years…

Profitability

Gross Margins

Entrata has bottom-quartile gross margins. Since they represent the ceiling of profitability, gross margins is a key metric.

While Entrata has lower gross margins, the improvement over the past couple of years has been incredible. From 54% GAAP gross margins in Q1’24 to 63% in Q1’26….

Increasing gross margins that much is hard to do. I believe a lot of it is a higher mix of customers with higher gross margin products, but also from continued product additions.

GAAP Profits

According to Alex Clayton at Meritech, Entrata is the largest software IPO to be GAAP profitable while growing 20%+. Not to diminish that record, but a lot of that is because most software IPOs were growing faster so they were trading profitability for growth (at least that was the idea…)

Regardless, Entrata is very efficient. Especially after considering their lower than average gross margins. Their OpEx is very efficient.

Free Cash Flow Margins

They have to be FCF positive for a long time given their [mostly] bootstrapped history. Entrata’s good (not great) FCF margins.

Rule of 40

The median software company has a Rule of 40 Score of 38. Entrata is just slightly better than that at 40.

Retention

Entrata also has great customer retention metrics:

• GRR: 99% in 2024 and 97% in 2025

• NRR: 117% for 2024 and 2025

Customers just don’t leave Entrata…And that is exactly what every software investor is trying to assess right now. “How durable will revenue growth be over the next 5 to 10 years?”

What Will Entrata’s Valuation Be?

Entrata’s last fundraising in May 2025 valued them at $4.3B. They also did a tender offer a couple of months later at the same price where employees and former employees sold $37M worth of shares.

It’s worth noting the public cloud stocks are down on average by 8% and their closest comp (AppFolio) is down 23% over the same period.

There isn’t going to be irrational exuberance around the Entrata IPO given all investors’ attention being on AI and other IPOs, but MANY folks will be paying close attention to it as a signal for the broader software IPO market.

Below is how I would expect Entrata to price at IPO.

Bootstrapping, Magic Mushrooms, CEO Forced Exit, & PE Takeover

Entrata (property management software) was founded in Utah 23 years ago when vertical software was a lot less sexy than it is today. Being in Utah and in vertical software at the time meant they didn’t follow the traditional VC path.

Two Early Investments (2005-2006)

Despite almost all reporting saying they bootstrapped for 18 years, Entrata did raise a couple of small early rounds (not institutional money though) in the early days.

In 2005 Entrata raised $125K at a $1M valuation. And then not too much longer after that they raised another small round at a $2M valuation (amount unknown but probably around $200K-300K)

Between these two rounds, Entrata was diluted by ~25% (my best guess)

Bateman recently said that the $125K investor/s made a 1,100x return on their investment (probably exited in the next round below). Not bad!

Founder Takes Magic Mushrooms (2018)

Bateman went to some magic mushroom retreat and decided he was done running Entrata. Bateman wanted to keep his equity but was done with the day-to-day. So Entrata hired a new CEO to run things.

1st Institutional Capital (2021):

PE Firm, Silver Lake, led a $507M round in 2021 (the largest private round in Utah history). The valuation was never disclosed, but I have a guess…

The Silver Lake investment was not all primary capital. There was likely a large amount of founder and early investor liquidity. Much of the details were never disclosed, but Bateman has said that he owned ~60% after the Silver Lake investment, which likely means a lot of the $507M bought out those early investors.

Based on some triangulation of the numbers above, I would guess the valuation was ~$1.5B.

CEO Forced to Sell 100% of His Holding Due to Unhinged Email (Early 2022)

Bateman sent a conspiracy-filled, racist email about COVID-19 to the most prominent figures in Utah.

Bateman was kicked off the board and they forced him to sell 100% of his holdings, which was led by Silver Lake. If Silver Lake acquired Bateman’s ~60% shares at the previously estimated round price of ~$1.5B, then he would have walked away with ~$0.9B, but maybe Silver Lake got to buy his shares at a bit of a discount given the circumstances….

After the sales of Bateman’s shares, Silver Lake became the majority shareholder of Entrata.

Blackstone Invests $200M at $4.3B (May 2025)

In May 2025, Blackstone invested $200M at a $4.3B valuation. A big increase in valuation from a few years earlier. The average public cloud company’s stock price fell by ~30% over that same period of time.

Debt and Dividend Recap

Entrata had $200M of debt on the books by 2024, but then in 2025 they refinanced and took on $400M of total debt. They used almost all of that to pay a dividend of $356M….

Entrata got good debt terms on it though.

They only have to pay $1M per quarter with a bullet payment at the end and a rate of 3% + SOFR.

Will Entrata Have A Successful IPO?

I really hope so because maybe that means Entrata will start to open up the IPO window for the hundreds of others waiting for their turn.

The timing is pretty good from both a market perspective (public software is up nearly 40% from the lows last month) and Entrata continues to execute strongly.

While not some sexy AI company, Entrata is a solid business. The big question is what valuation will investors give them when it seems like the only thing folks are excited about are all the potential hot AI IPOs (OpenAI, Anthropic, SpaceX, etc). And investors remain nervous about software, despite the recent bounce in stock prices.

Entrata seems like a great company with a lot going for it. We all just need to be able to answer one question: “How durable is revenue over the next few years?”

Congrats to the Entrata team! I am rooting for you.

Footnotes:

• See how Agentic Finance can transform your finance department (today’s sponsor is Brex).

• I wrote about how to significantly cut your ChatGPT/Claude spend on my new AI-focused newsletter. Check it out and subscribe

• Subscribe and share the OnlyCFO newsletter with your teams

Subscribe now

*nothing in this post should be considered investment, legal, or tax advice

I made plenty of mistakes when expanding internationally. I’ve learned a lot over the years, so I wrote a guide on some of the most important finance considerations:

• General guidelines for how to hire (contractor, EOR, set up an entity)

• Offshore cost benchmarks by country

• Additional employer burden (don’t forget about these!)

I partnered with Deel to pull real cost benchmarks (the kind of data I wish I had when doing this). Download my playbook for expanding internationally.

OnlyCFO's Global Hiring Playbook

A CFO’s Worst Nightmare: “Tokenmaxxing”

Anthropic is awesome. I am going to probably use $300 million of Anthropic this year at Salesforce. Coding. Everything’s going to be cheaper to make. — Marc Benioff (CEO of Salesforce)

AI tokens are becoming the biggest vendor expense on the income statement (if it isn’t already). From basically $0 spend to the largest in just 2 years...If token costs aren’t rapidly becoming your largest vendor expense, then that probably means you are behind in AI adoption.

Everyone is “tokenmaxxing” right now. And everyone seems to be encouraging it. Well, everyone except for maybe the CFO 🤣.

Tokenmaxxing: maximizing AI token consumption as a signal for AI adoption.

The primary goal, which everyone can agree with, is to make sure folks (across departments) are really adopting AI. AI adoption is critical to every tech company’s survival in 2026.

But this AI adoption is creating a very large expense that companies need to make sure is treated properly on their income statement. And leaders need to understand how it impacts the financials, their budgets, and the company’s long-term unit economics…

Share

AI Costs on the P&L

Cost of Revenue & Gross Margins

When AI is embedded in your product, then that portion of the costs is included in COGS. It’s an incremental expense line in COGS.

The COGS lines 1-5 are the same, but now we have a large new one (#6).

1. Professional services

2. Support team

3. DevOps

4. Customer Success Management (diversity in practice)

5. Infrastructure/hosting costs (AWS, Azure, etc)

6. AI token/inference costs

   1. API spend with AI models (Claude, ChatGPT, etc)

   2. Model hosting when you self-host (not calling an API but using AWS, Azure, etc GPU power)

   3. Vector database costs

   4. Orchestration and observability

This new (and large) expense line in COGS is creating much lower gross margins and that requires a shift in how leaders think about long-term unit economics.

While AI gross margins are up a lot in the last couple of years (from 41% in 2024 to 52% expected in 2026), they are still significantly lower than typical software gross margins.

The median public software gross margin is 77%. This is largely because true AI revenue still isn’t a significant portion of their total revenue yet. Gross margins will be materially lower for most of these companies in 12-18 months…

Because of the vastly different economic profile of AI products, it’s important to be able to track the revenue and COGS of all products separately.

I have talked to a lot of companies that have been blending AI token costs in their “hosting” P&L expense line and also can’t segment gross margins by product. Do NOT do this. The extra data is critical for making decisions.

Create a new line just for the AI costs running through COGS. You, the broader exec team, and the board will want to pay close attention to COGS from AI so tell your accounting team to create a separate line for it.

Below are a few new metrics that every company should be tracking:

• AI COGS as a % of AI revenue. How much lower are AI gross margins?

• Gross margins by individual AI product. Do some AI products have worse gross margins?

• AI COGS relative to total revenue. How much is AI degrading overall gross margins? What is the trend?

• Correlation of AI COGS with revenue. Do you have a broken pricing model?

R&D AI Costs

Your “builders” (engineering, product, and design) are typically the people at the company who are burning the most tokens/spend on AI. And they should be spending the most given what they should be building.

The breakdown below shows the costs required to develop an AI product. Costs are pretty consistent across product development stages.

According to a survey performed by Scale (that Jason Lemkin discussed on SaaStr), the average developer spent $1.3K/month (or $16K per year). And many folks think this is too low…

But you can’t have more token costs and not have something else impacted. Either the higher spend has to directly translate into more revenue. Or…it means you have to cut headcount costs.

You have to figure out the ROI math on when the additional token cost per employee is less effective than the headcount that must be cut as a result of the higher token spend.

S&M AI Costs

There are a lot of AI use-cases in sales and marketing, but most of them are not that expensive in tokens from what I have seen.

One S&M AI cost that can be large (and often misclassified) is the incremental POC/trial costs for AI products. The costs associated with these should go to the sales/marketing department. I have seen many companies not track these well, and they end up erroneously in COGS.

These POCs/trials are increasingly common with AI products as customers want to see the value before they commit. Companies need to make sure the costs are not spinning out of control as they add more AI stuff.

G&A AI Costs

Beware of the AI cost dumping ground into G&A. G&A frequently becomes a cost dumping ground (even before AI):

1. Ignorance or bad accountants who don’t know that some of those costs don’t belong in G&A

2. Intentionally pushed to G&A so it makes all the important financial metrics look better

3. Some diversity in practice where some costs go

But company-wide enterprise AI licenses (Claude, ChatGPT, etc) do NOT all belong in G&A…In fact, the amount of AI-related costs sitting in G&A should be smaller than the other buckets (COGS, R&D, and S&M).

Yes, my G&A teams are doing some cool stuff with AI (check out CFOpilot to see what my teams and I are building with AI). But we should not be nearly as token-heavy as what R&D (or usually even S&M) should be spending.

Allocating AI Costs

Performing allocations of certain shared costs is nothing new for most companies. Although many do it wrong (especially private companies)

Example: IT Department Allocation

All company-wide software that is used by everyone (Slack, Zoom, Gmail, etc) is put into the IT department and then at the end of each month, it’s allocated across all the departments based on headcount.

Every department absorbs some of the costs since they all use it. So these costs end up in COGS (allocation to the support team), R&D (allocation to engineers), etc.

Your AI bill should NOT be allocated this way.

Headcount was a good proxy for how seat-based costs should be allocated across the P&L, but that breaks down for consumption-based pricing vendors. Especially when the usage can vary so much across departments and when it is a massive expense (like token spend).

More exact allocations are needed for token costs. The only good approach now is to do direct allocations based on specific department usage. Finance teams need to segment the different types of AI costs by department. Anything else will result in materially incorrect financials and unit economics.

Final Thoughts

Token costs are going to be the largest vendor expense in nearly every leader’s budget by the end of 2026, so it’s important to get the accounting and department budgeting right.

Trade-offs and changes will have to be made. You can’t just add a massive line of expense across the P&L and expect nothing else to change.

But companies can’t make the right decisions if the financial metrics are screwed up because AI costs aren’t properly allocated across the P&L.

Footnotes:

• Download my playbook for expanding internationally (when to use an EOR, set up an entity, compensation benchmarks, and more).

• Subscribe and share the OnlyCFO newsletter with your teams

Subscribe now

📚 Other Interesting Things:

Learn the Numbers

So many leaders and founders today don’t understand the numbers. And that will be a problem as they try to scale…

Forward my newsletter to your teams so they can get smarter on the numbers :)

Anthropic’s Revenue

The impact on all of our P&Ls from “tokenmaxxing” can be clearly seen in the unprecedented revenue growth rates of ChatGPT and Claude. We may never see something like this again…

Deel’s Financial Path

Deel is a company that continues to surprise me with their financial metrics.

But growing as fast (and efficiently) as they have isn’t luck. And they can’t slow down now. Every company is facing an “accelerate or die” moment. Deel is choosing acceleration…

Revenue Growth

• $0 → $100M ARR in 20 months: A record growth rate at the time (before the AI growth madness). “I am sure their ARR definitions are aggressive, right?”

• $100M → $1B ARR in 6 years: One of the fastest companies to hit $1B ARR. “Aren’t they just a payroll company? They can’t be growing this fast!”

And the “SaaSpocalypse” hasn’t slowed Deel down….

Despite the insane growth rates you might see among some private hot AI companies, 50% growth at $1.5B ARR is incredible. At least top-decile.

They are growing faster than every public software company (except for Palantir).

Profitability

After doubting their revenue growth, my next instinct was to assume they are burning cash like there is no tomorrow. “A payroll company probably has terrible margins, right?”

Last I heard, a few months ago, they have ~17% EBITDA margins…which is also best-in-class, especially considering their high growth rate…

Deel’s “Accelerate or Die” Moment

I recently sat down with Deel’s new CFO, Joe Kauffman, to discuss how Deel is accelerating in the age of AI and the importance of capital allocation to succeed today. Deel brought Joe in to be the leader that is able to take Deel public (he has the right experience).

Every single company is facing an “accelerate or die” moment in 2026. Below is how Joe and Deel plan to accelerate and become an AI beneficiary.

#1 Metric: High, Durable Growth

Leading up to an IPO, Deel needs to continue to prove why it’s worthy of the public markets. That is much harder today than a few years ago. Stellar financial metrics alone aren’t enough anymore. Every investor is questioning the terminal value of all technology companies.

If terminal value (TV) is trending to zero, then investors won’t want to touch it. And valuations will tank.

We are seeing this with many public names in the low growth bucket. There is a 5x difference between low growth and high growth company valuations…

CFOs should be thinking, at a minimum, about how to make their businesses AI-defensible. Think about how the core business may be susceptible to disruption, not just what’s right in front of you, but further down the road.

This focus on AI defensibility can be especially hard for public company CFOs, as it may require disrupting themselves in order to re-architect the business to be more defensible. — Joe Kauffman

Revenue growth continues to be the most important financial metric, but there is a lot more emphasis on durable revenue growth in 2026. “Can you keep strong growth rates for a long period of time?”

Whether you want to be acquired or IPO, nothing matters more than proving your revenue is durable in order to have a good valuation.

You Must “Accelerate”

AI can certainly help companies cut a lot of costs. And many CFOs will celebrate and brag about how much money they are saving by slashing software, headcount, etc.

But, if revenue growth falls off a cliff, do those savings matter? No…

When I refer to “accelerate” I am mostly talking about product. And the revenue will follow. The companies that don’t accelerate product in 2026 are going to get run over by the competition. Churn is going to skyrocket and acquisition costs are going to increase.

The most difficult challenge may be finding incremental businesses that are indexed to AI. That way, every time there’s a model improvement or other AI tailwind, you win too.— Joe Kauffman

Deel certainly seems to be accelerating product…And they are doing it in a way that benefits from their existing advantages (payroll system of record).

• Released Akai (last week) - interconnected agents that learn your team’s workflows so they can automate as much as possible.

• Acquired Sastrify (2 weeks ago) - software management and procurement

• Release of token monitoring tool (2 weeks ago) - ties token usage from LLMs to employees to help understand ROI and employee performance

• Deel app for ChatGPT (3 weeks ago) - Deel is now directly in ChatGPT so users can use AI to search and report directly from Deel

And there is probably even more cool stuff Deel is releasing, but the above are the things I am excited to try out.

Besides just their sheer speed of building, they are creating the right stuff that builds upon their moat/advantage - they are becoming the “operating system for work” (people and agents).

Getting Efficient to “Accelerate”

The second step is thinking about productivity gains from AI, and considering how to take those savings and reinvest them in products that drive top-line growth. — Joe Kauffman

Deel has acquired 14 companies since it started 7 years ago. And it continues to release products at a rapid clip. Deel (like everyone else) needs to use AI to become more efficient so they can reinvest some of those savings in the mission of “accelerating”.

Companies need to rebuild for efficiency. Some rebuilding should have likely happened regardless of all the AI progress (high-growth companies build bloat which also slows them down). AI is another strong reason to rethink how everything is done today.

Deel's new Ghostbuster role is one of my favorite org ideas of 2026.

Deel describes the problem as follows:

Over time, multiple layers accumulate, some become outdated, some become friction, and eventually people start optimizing things that probably shouldn’t exist in the first place. — Joe Kauffman

Every single person reading the above that has worked at a growing company knows that this waste can become VERY significant. And it slows everything down. You can’t be slow in 2026.

It’s a perfect time for this type of role because not only can they find regular waste, but they can find ways to rebuild things from an AI-first perspective. $10M in savings as their goal seems aggressive, but likely doable. The role has a 50x ROI based on the salary target if that goal is met.

Companies need to throw away conventional wisdom around org design and span of control because they are less relevant in an AI-first world. Historical headcount benchmarks are pretty meaningless in 2026.

Getting to “Real” Profitability

More and more investors will be looking at GAAP operating income and GAAP net income. CFOs would do well to look at the fully loaded cost per headcount (and fully loaded cost per agent) when considering ROI decisions for their human capital investments. — Joe Kauffman

Investors want to see that you can get to real profitability. And fast. As terminal values get called into question, investors are questioning mid to long-term profitability.

Not “adjusted EBITDA”….but real profitability.

Final Thoughts

Speed in capital allocation is really a consequence of the position you’ve built. The companies that will win are the ones that have built the muscle to move quickly and smartly. — Joe Kauffman

If you are competing with a company like Deel and you lack speed and/or direction, then you are going to get steamrolled.

Build with urgency. Build upon your strengths. Build things that will benefit from and become more valuable as the AI foundational models become more powerful (which they will).

Footnotes:

• Check out Deel (today’s sponsor) and all the cool stuff they are building

• Check out my new newsletter (cfopilot) for all the AI stuff in finance/accounting

• Join 37K other subscribers reading the OnlyCFO newsletter

Subscribe now

For finance leaders, managing international payroll and compliance requires expertise. Our ‘Guide to EOR for Startups’ offers a detailed look at efficient global payroll management & compliance. This guide will help you:

• Manage cross-border payroll easily

• Navigate tax implications across jurisdictions

• Reduce financial risks associated with global growth

Take control of your financial strategy today.

EOR Guide

Rebuilding Finance with AI

AI has dramatically changed how my team operates and we are still early (lots more work to do)…

I built an AI agent that turned a 2-hour task for my accounting team into 5 minutes. And we used this same AI flow across many similar processes (which cut a full day off of the month-end close).

I just launched a separate newsletter (CFOpilot) to write about everything I am learning and building with AI.

*The below post was shared at CFOpilot a few days ago. I got a ton of feedback on how valuable it was, so I wanted to share it here as well.*

My 6-Step Process to Building an Accounting AI Agent

I built our first AI agent to tackle accounting reconciliations (focused on prepaids initially).

Here is the process I ended up with…

An “agent” is essentially the worker that navigates all of these systems and steps to complete a task.

Step 1: The “Skill” Itself

Skill: A self-contained workflow an agent runs. A system prompt that defines what to do. MCP connectors to the systems it needs. A trigger that kicks it off. A config that drives the behavior. You build it once and it runs forever.

Every skill has a SKILL.md file. The skill defines the inputs, workflows, rules, output, edge cases, etc. My prepaid skill is ~200 lines.

You’re not writing a prompt every time you run it. You’re writing it once. And then updating it if/when something changes.

Step 2: The Config File

My config (YAML file) contains:

• Prepaid GL account codes

• Vendor → coding map

• Materiality threshold

• Partial-month convention

• Routing: who gets pinged, in which channel

When something changes, you edit the config (not the skill).

This is one thing I was doing wrong before. I was dumping everything into a prompt (or a skill). Every time something changed, I would have to update the skill and re-do it.

Having a separate config file with the stuff that might change breaks things up so it’s easier to maintain. It’s like why you don’t put your entire model and all the components in one giant tab or make Excel model formulas huge.

Putting everything in one skill file also makes it so you can’t reuse your skill in other recons, entities, etc because it contains the exact detail for one prepaid recon process. To make it scalable/reusable, you separate out those details into the config file. Then you can reuse the skill and just update the config file.

Step 3: MCP Connector to the ERP

MCP (Model Context Protocol): A standard way for an AI agent to plug into your software systems (Deel, NetSuite, Google Drive, etc) so it can read data, do work, and take actions in those systems on your behalf.

A surprisingly large number of companies I talk to still use AI like a simple chatbot.

The biggest unlock comes from giving AI visibility to everything…Only then can AI agents perform meaningful work end-to-end. The primary way to do that is via MCP connectors.

If AI is not connected to all your systems, then it lacks context, so it will require a lot more manual human effort for each handoff and coordination. AI needs to “see” all the context and have the ability to take actions (even if read-only) to be really useful.

My prepaid agent queries my ERP GL directly via the MCP. No CSV exports/imports.

The skill I created reads the config file to know which prepaid accounts to pull. The agent calls the ERP connector and pulls current period activity for those GL accounts.

If you don’t have an MCP connector to your ERP, that’s the first thing to fix.

Step 4: MCP Connector to the Shared Drive

The workpapers don’t live in your ERP. They live in GDrive, Box, SharePoint, etc.

The skill needs to:

• Read last month’s workpaper

• Write the new workpaper to a new file

• Enable the AI agent to create and edit the new prepaid workpaper

An MCP connector for the shared drive handles this.

Step 5: Slack Is the Trigger and the UI

Slack is the UI for my AI agent.

Slash command: I kick off the recon by typing “/prepaid April”. Or I just chat in Claude “I want to kick off the prepaid recon for April” and it will find it.

The agent will find the prior workpaper (location is defined in the config file), copy that workpaper to the current month-end close folder. It will then use that as the starting point for adding new prepaids, adding a new month of amortization, etc. The agent is trained just like a human would be…it brings over Excel formulas just like the prior month so the amortization calculates correctly.

When the agent finishes, it posts a digest with the workpaper, the JE CSV, and a tag for the preparer.

Below is an AI-generated Slack example of what I see (done with AI so you don’t see my actual Slack…)

If the AI agent flags something I need to fix, then I’ll do one of two things:

1. Open up the workpaper and fix the errors/flags (e.g., add the Cloudflare term from the example above). Then I add a ✅ to the Slack thread, which signals to the agent that I am done so it re-reads the files to see if I fixed the flags. It updates the files and posts a follow-up message for the reviewer.

2. If the issue will be repeating, then I will update the skill or config files to automatically fix it going forward.

Step 6: The Human Part

The skill doesn’t auto-post the JE. We still have a human preparer and reviewer.

• Human preparer: Person managing the agent and fixing the flags/errors that come up. They also do a review of the agent’s work.

• Human reviewer: This is the same process that the reviewer has always performed, and is an important control in accounting.

Essentially, we have turned both of these humans into reviewers. The preparer does the first pass and makes sure the agent did everything properly and then the normal reviewer acts as the final control and sign-off.

But we have even changed the reviewer’s process as well…more on that to come in future CFOpilot articles.

In Summary…

What previously took lots of manual steps (copy workpaper to new location, open up ERP, download detail, update workpaper, create JE upload file, etc) is now 90%+ completed by our AI agent.

From ~2 hours of human work to ~5 minutes.

It did take longer the first time I ran the reconciliation, but then the team took it and kept iterating and improving the skill and the config to automate more and more. Eventually, you get some major time savings.

And then repeat that across your other recons and you will shave off days from your month-end close…Smaller accounting teams, faster close, fewer errors, etc

Footnotes:

• Download this EOR guide from Deel. Every company looking to expand internationally should read this.

• If you are in finance/accounting, then check out my new AI newsletter (CFOpilot).

Subscribe to CFOpilot

A lot can go wrong when expanding internationally. I’ve made many of those mistakes. So learn from Deel—what they’ve seen doing this thousands of times and in hundreds of different countries.

• How should you hire? Contractor, through an EOR, or set up an entity

• What are the risks? Compliance, financial, etc

• What countries should you hire in? Each country has different rules and risks

Check out Deel’s International Hiring Guide

Download the Global Hiring Guide

Insane Demand for Anthropic Stock

Anthropic is on the most insane valuation runs in history. And I doubt there will be anything like it anytime soon…

Anthropic went from $120 billion valuation to an implied $1.4 trillion in ~7 months. The valuation is up 40% in the last 24 days 🤯. With annualized revenue climbing from $100M to $45B in <3 years…

Everyone has FOMO about it. Every investor wants exposure to Anthropic, but there are very few sellers. An expected IPO and insane revenue numbers has sent the valuation parabolic.

Where there is incredible demand, supply will be created. But that supply may be fraud or contractually illegal…

Anthropic’s Warning

Imagine investing in Anthropic over a year ago and seeing your investment increase by 50x+. Your $100K investment (min investment required by the SPV that you invested in) is now worth $5M+ in just a few years. Generational money, right?

Now imagine the investment is deemed voided and you are told you are just going to get your original investment of $100K back (if even that). Ouch!

Well, this is what many folks who thought they were Anthropic investors are wondering right now…

Anthropic sent a blunt warning to anyone trying to get equity access before their IPO. OpenAI did something similar as well this week.

Anthropic even took the warning a step further and publicly put 8 firms on a shame list. Anthropic said that if you bought from these firms, your shares are not valid…

Anthropic secondary investors are nervous, but especially those who got exposure through these firms.

A lot of people on social media said they were shocked and “how could Anthropic do this?”

What I am shocked about is the number of people who are shocked about this. This is very very standard amongst private companies... Clearly a lot of folks haven’t read their equity agreements.

Nontransferability and the “ROFR”

Nontransferability Clause

Investors and employees that are granted equity (stock options or RSUs) of a private company have a “Nontransferability” clause. Below is a very standard clause in a company equity plan:

No Option share shall be assignable or transferable (including arrangements that transfer any of the economic consequences of ownership of Option). Any transfers in violation of the foregoing shall be void.

Not voidable….but void. The distinction matters.

Void vs. voidable: A voidable contract can be enforced until one party elects to cancel it. A void transaction never legally exists at all. You can’t ratify it retroactively, or enforce it in court. If Anthropic’s board didn’t approve the transfer, you simply do not own shares, regardless of what paperwork you signed or what you paid.

And here is what Anthropic said about secondaries:

Any sale or transfer of Anthropic stock, or any interest in Anthropic stock, that has not been approved by our Board of Directors is void and will not be recognized on our books and records.

The ROFR

Every private company has a ROFR (right of first refusal) provision in their equity plans.

1. Sellers must give written notice to the company with:

   1. real offer from an investor to purchase

   2. potential buyer’s name and information (to make sure it’s legit)

   3. details of the offer (# of shares, price, terms, etc)

2. Company typically has 30-60 days after receipt of offer notice to elect to purchase the shares at the same price and terms.

3. If the company (or nominees) doesn’t elect to purchase then seller can sell the shares to the buyer at the stated price or higher (but not lower). They usually have a set period for the transaction to close (like 60 days) before they have to go through step 1 and step 2 again.

*Note - Anthropic’s message implies that sale of their stock must actually be approved by the board, which is much more strict than the typical process of a company nominee just letting the ROFR period lapse.

Anthropic Specifically Banned SPVs

What is a Special Purpose Vehicle (SPV)?

An SPV is a temporary investment entity. It is usually formed as an LLC and it is created for one specific deal, so a bunch of investors can pool money together and show up on a startup’s cap table as one line item instead of 50.

SPVs aren’t necessarily bad. There are many single layer SPVs, that work well for both the company and all the investors.

But when there are multiple SPV layers (and SPV buys into another SPV which buys into another SPV)….that’s when it often starts looking like a scam. Fees on top of fees on top of fees.

Anthropic stated that they have specifically banned SPVs....

We do not permit special purpose vehicles (SPVs) to acquire Anthropic stock and any transfer of shares to an SPV are void under our transfer restrictions.

Even if someone did acquire shares legitimately, transferring them into an SPV is itself a transfer requiring board approval. The chain of approvals required makes most SPV structures non-starters. Companies generally hate them (at least CFOs and legal does) because it creates a compliance and cap table management headache.

Most equity agreements have mechanisms to block SPVs. Again, not all SPVs are bad but it’s important for companies to be in control what SPVs they accept and when.

SPVs Were Used in the FTX Bankruptcy Sales

One confusing part that Anthropic doesn’t seem to acknowledge is that apparently a lot of SPVs bought Anthropic stock via the FTX bankruptcy.

When FTX was going through bankruptcy, it sold $1.3B worth of Anthropic stock. Most of this was sold to SPVs who got the appropriate ROFR waivers…

The list of SPV buyers is publicly disclosed as part of FTX court records from the bankruptcy. Here are some of the Anthropic’s largest SPVs that should be legit since they apparently had Anthropic board approval.

The problem with these SPVs came immediately after though:

• FTX sells shares via court-approved process to SPVs→ legitimate

• Many of those SPV buyers then sold LP interests downstream to retail/family offices → void under Anthropic’s policies

• Those second layer SPV buyers then sold LP interests again (3rd layer) → void and likely very insane fees.

Why Do Companies Have These Rights?

Companies have transfer restrictions and ROFRs for a few reasons:

1. Keep the cap table clean. Less people to manage and deal with in a potential future exit. Companies don’t want unknown investors, competitors, etc. Messy cap tables delay IPOs and M&A…

2. Compliance nightmares. Section 12(g) creates strong incentives to limit secondaries and SPVs because tripping certain thresholds creates A LOT more compliance and unwanted disclosures. Every founder/finance leader should know these rules.

3. Prevent 409a issues. If pre-IPO shares are freely traded in the market then the 409a price will likely have to take those into account. Then the 409a price for new employees (and additional grants) is a lot higher. So these employees get screwed because there is less upside for them because of the secondaries.

4. Prevent fraudsters or major grifters being associated with them. These contractual rights and the blunt warning hopefully will deter potential investors from getting scammed. No company wants scams associated with them. It looks bad and it often creates a distraction (and often legal work).

5. Take advantage of undervalued shares. They think the price the seller is getting is cheap so the company (or their investors) want to take advantage. Note - this is a potential side benefit but the prior reasons are the most important.

What Actually Happens to Investors’ “Exposure” in Anthropic ?

The IPO will expose the pure frauds versus those that have stepped into a more grey area. Most of the grifters are operating in the grey, but there 100% is fraud out there in many deals.

1. Fraud (no contractual exposure): Seller doesn’t actually have contractual exposure (regardless if technically voided because of rules) or their exposure doesn’t cover what they promised. When Anthropic IPOs, the seller can’t deliver the massive paper return that was promised because there are no actual underlying shares/exposure. So the buyer sues. Securities fraud, wire fraud, etc. Send these guys to prison!

2. Grey Area (has exposure but broke contractual restrictions): There are a couple of practical options here:

   1. It is just quietly settled post IPO once seller gets the cash. Anthropic doesn’t know about it or doesn’t want to try to intervene. This is what usually happens.

   2. Seller (with actual shares) gets greedy. They find out that the agreement they entered into with a sketchy SPV manager is technically void and think they hit the lottery. “Sorry I guess our transaction is void so I’ll give you your investment back, but I keep the 100x return” over the past couple of years.

3. Legit secondary: Congrats! You are rich and don’t have to be nervous.

Note - While many grifters have multiple SPV layers and insane fees, they aren’t outright fraud. Just deeper into the grey area. However, many of big grifters are probably the same folks who would make misrepresentations and commit fraud.

Final Thoughts

Anthropic and OpenAI’s blunt warnings are mostly a CYA before an IPO.

These companies (and several other hot private companies) have seen a lot of these shadow secondaries/transfers. Not outright fraud but deep into the grey and void under their equity agreements.

Anthropic has to put out these messages to clearly condemn these transactions so it doesn’t blow up a bunch of compliance stuff, create a ton of legal work, delay an IPO, etc.

Anthropic is not going to spend legal resources on chasing these transactions down. As long as they can claim they are legally void and doesn’t impact its cap table, there is no real incentive for Anthropic. It’s great to see bad actors called out though.

Until the IPO, many secondary investors in Anthropic and OpenAI are going to be sweating a bit about whether they actually get the cash. The vast majority will settle without Anthropic, but the IPO will surface the fraud…

Footnotes:

• Check out this International Hiring Guide from Deel if you are looking at expanding internationally. It’s an excellent resource.

• If you are in finance/accounting then check out my new newsletter (CFOpilot).

CFOPilot

I Built an AI Agent for Accounting Recons

👋 I am sharing the cool AI stuff that I am building. AI for finance/accounting folks (and others who just want to learn AI generally). Follow along and subscribe 👇…

Read more

a month ago · 19 likes · OnlyCFO

*nothing contained in this post is legal, tax, or investment advice

Subscribe now

I’ve seen companies roll out new global comp strategies that drove employees to quit. It felt unfair, was poorly rolled out, or was just a bad comp strategy. It’s hard to get right.

• What pay model to use? Anchor city-based, single global pay scale, localized, etc.

• How do we do equity comp across jurisdictions?

• Compliance considerations? Local HR laws, mandatory bonuses, statutory benefits, overtime, etc.

Check out Deel’s guide for a breakdown of key decisions when creating your global comp strategy.

Download the Guide

A Buyer’s Market

It is a software/AI tool buyer’s market in 2026. This has become very clear to me based on some recent vendor renewal conversations. Most companies (and their sales reps) are panicking…

I have been adopting Michael Scott’s negotiating tactics for my renewals. It’s working. And it’s scary how much leverage I have in 2026 with a lot of my vendors.

This post is about what I’m thinking about as a leader at a tech company and how that impacts my approach to vendor management and procurement in 2026.

AI Gives Buyers Power

Before I get into some of my vendor negotiation tactics, it’s important to understand why I have so much leverage today.

The Spend Buckets:

Spend falls into one (or sometimes multiple) of the four buckets below. I have ordered them based on their general importance. The first three rankings may move slightly based on the company’s phase and objectives, but #4 is ALWAYS last.

1. Increases revenue

2. Mission-critical (cybersecurity, ERP, etc)

3. Improves efficiencies

4. Candy (or “nice-to-have”) tool that doesn’t really fit into the above groups

“Candy” is not the bucket you want to be in as a company. You are the first spend bucket to get cut, pricing power is weak, and there are usually lots of options. I frequently talk to vendors who believe that they are in the top 3 when they actually are candy.

But the bigger problem that a lot of software/AI companies are facing in 2026 is that the nice-to-have (candy) bucket is rapidly expanding.

AI Pushes Vendors Into “Nice-to-Have”

AI is pushing a lot of vendors into the “nice-to-have” spend bucket.

• Build vs buy: the solution that the tool itself solves might be necessary, but can the AI base models solve it directly? Or is it easy now for companies to build internally something that solves 80% of what they need?

• “Nice-to-have” add-on features: There are lots of tools where 50%+ of the price relates to add-on features. The core product may sit in buckets 1-3 (and you really need it), but you no longer need all the add-ons. Folks are going to see a lot of downgrades in 2026.

• Vendors are going more multi-product. I am a strong proponent of going more multi-product faster. This is basically required in the age of AI because you can move so much faster and it makes you stickier. But it also means many of my vendors are starting to have more overlap. Some are offering 70%+ of the same functionality for free that I am getting from another vendor…

• AI tokenmaxxing is eating budget. The AI models now sit at the very top of the need-to-have list (before anything else), which means if you have limited budget, then everything else gets pushed down in priority. Tokenmaxxing is where employee productivity is measured by number of AI tokens consumed. The result? Costs are going up, up, up, and away.

2026 is the Year of Churn

I wrote about this a few weeks ago, but churn is going way up in 2026 (especially in H2). And more specifically, downgrades. Because of the things I mentioned above, customers are downgrading features, they are getting bigger price discounts, usage is decreasing on many modules, etc.

This will show up in your retention metrics like GRR and NRR, but you will want to drill into what’s happening. Internally, I am breaking down the churn number a lot more in 2026.

I want the information below (and by major product as well).

Buyers Need to Ask More Questions

Below are some questions I am asking for software purchase requests. The level of detail I dive into depends on the size of the contract and importance of the tool.

Tool Needs

• What bucket does the tool fall into and can you quantify it? This is important for me to understand its priority against our current company objectives.

  1. Increases revenue

  2. Mission-critical

  3. Improves efficiencies

  4. Candy (nice-to-have)

• New vendors: Why do we need [X] tool now? We were fine before.

• Renewals:

  • Have AI and related process changes made this tool not needed or just a nice-to-have now?

  • Has our business changed in a way that this tool isn’t needed right now? Example: I know a lot of finance teams that bought expensive tools in preparation for an IPO that is no longer coming (at least anytime soon). Once purchased, these tools can feel very critical, but often aren’t and they don’t add real value if you aren’t actually close to an IPO.

Budget Considerations

• Is the spend in the budget? If not, how will we make up the difference? You’ve got two choices…1) people or 2) software/AI.

• Could we build all or some of this internally? Some features are not only cheaper to build internally but more customizable so they work even better for you.

• How much are the alternatives? AI is quickly closing the product gap with many competitors, so I expect more pricing concentration (fewer major outliers) than before.

• What tools do we already have with overlapping functionality? As I mentioned earlier, this is accelerating in the age of AI. One of my vendors just released a module, for free, that is 80% of the functionality that I was paying another vendor $50K/year to do. This is going to become more common.

Agent/AI Capabilities

• Do you have an MCP connector and are you easy for agents to work with? Everyone needs to be asking these questions in 2026. Agents will be running a significant amount of your processes (if they aren’t already) so you need a tool that plays nice with them.

• Ask the seller, “How is your AI stuff different from what I can do with Claude/ChatGPT? If nothing else, it will make them nervous and result in a bigger discount. But I also want to actually know. If the sales rep fails to articulate a good answer then maybe the base models are fine to use instead.

Pricing

• What are the implementation costs?

• Will we need to upgrade plans soon to enterprise or some other tier?

• I want a detailed analysis on how pricing works and modeling for what our planned usage will look like.

  • I start with distrust on the total estimated tool cost for usage-based pricing, so you’d better earn my trust with a thoughtful analysis.

  • What does the cost look like this year and how does it scale next year?

• If we have significantly underutilized the tool under the first contract, then I am pushing hard come renewal. So teams need to make sure the company is getting maximum value out of the tool at least a few months before renewal, otherwise I am going to try to slash.

Just another thought on pricing below from the CTO of HubSpot and Jason.

What Are CFOs Doing?

Maxing Discounts

5-Step Renewal Playbook

Step 1: Send cancellation notice (>3 months before renewal). Say you are evaluating options and don’t want to get stuck in an auto-renewal.

Step 2: Wait for the vendor to come back to you closer to your renewal with pricing, which will already be discounted more than usual given the cancellation notice.

Step 3: Stay silent. Wait for them to follow up.

Step 4: Tell them you like the tool, but it’s not critical, other current vendors have similar functionality, and/or the new one… “We are looking at doing some stuff internally with AI instead.”

Step 5 (optional): If a real threat, tell them you are churning after all because the price isn’t worth it given the other options.

I talk to a lot of vendors and a lot of companies selling stuff. All of them are trying to hold the discounting line right now, but if pushed…there is definitely willingness to go lower.

You don’t have to be that aggressive, but you should definitely be pushing hard on pricing on all vendor contracts.

Pricing Benchmarks

Pricing benchmarks are important, but…don’t let them anchor you on the discount you might be able to get. Like I said, 2026 is a buyer’s market. Pricing benchmarks are lagging and they are an average/median.

I like to have them, but I also know sales reps are discounting a lot more than that in 2026…

Multi-Year Contracts:

Company Policy: No multi-year contracts

I, of course, will do some multi-year contracts (AWS, Azure, CRM, etc), but basically the only ones we will do are ones that I am involved with negotiating anyway so the company policy can still be “we don’t do them.”

And more folks are following this advice. Signing multi-year deals on most software in 2026 feels irresponsible. Who knows what tools/needs will look like in 2 years…

The larger discount with multi-year contracts is not worth the risk.

Auto-renewal:

I always try to negotiate these out, and if the vendor refuses, then I will just send an opt-out notice the day after the contract is signed. Don’t get stuck in an auto-renew.

In the past, vendors would often allow you to get out if you were past the auto-renew date, but I have heard more stories recently of vendors holding customers to it. Desperate times…

Everyone Wants a POC

Most AI vendors are offering POCs. If you are not, then I am very unlikely to consider you. I want a real POC that proves value before I commit to a one-year deal.

Final Thoughts

On the buyer side, continually re-evaluate your needs. Yes, you can save a lot of money with vendors right now, but that is only part of it. Companies need to make sure they have the right tools for their team at the right time. And needs are changing rapidly. If you are at a company and everyone doesn’t have access to some AI model, then you are bringing a keyboard to a gunfight.

On the seller side, figure out how you stay out of the nice-to-have bucket as much as possible. The amount of churn, downgrades, and concessions made on contracts will indicate if you are in that category.

You know who isn’t a nice-to-have in 2026? Anthropic. And it is evident from how their sales calls go :). Unless your deal is massive, they may pick up the phone to chat but not much beyond that.

Footnotes:

• Check out this compensation guide for managing a global company. Deel has some of the best data and resources out there.

• Email me at onlycfo@onlycfo.io to say hi, provide feedback, or ask questions.

Subscribe now

📚 This Week’s Interesting Things:

More Layoffs

Cloudflare responded with the following on social media about why they have so many job openings after a massive layoff.

This isn’t about cutting headcount. It’s about shifting what roles we need. More people building and selling product. Less providing back office functions to support them.

Everyone is a builder in 2026.

Agents are Eating Tokens

Token budgets are about to get craaazy…

A lot can go wrong when expanding internationally. I’ve made many of those mistakes. So learn from Deel—what they’ve seen doing this thousands of times and in hundreds of different countries.

• How should you hire? Contractor, through an EOR, or set up an entity

• What are the risks? Compliance, financial, etc

• What countries should you hire in? Each country has different rules and risks

Check out Deel’s International Hiring Guide

Download the Global Hiring Guide

Becoming “AI-Pilled” is Existential

I have been spending lots of time figuring out what it means to become “AI pilled” and how I can help my teams/company become AI super users. It’s my #1 priority. It’s existential.

AI-Pilled: Realizing AI will change everything. You reimagine your role, current processes/workflows, and what is possible. Your family and non-work friends hate you because AI is all you talk about.

The AI threat is existential in two related ways:

1. Your role/career

2. Your company

They are related. Companies that are going to succeed in an AI world are the ones that push their teams on AI and set high expectations. Employees win because they become AI experts (which is good for their career). Companies win because their employees are AI experts, which drives innovation. We are in an “accelerate or die” moment.

Side Note: Just because your company is “AI-pilled” doesn’t mean your job is safe. AI may still disrupt your role or employee layoffs may be necessary, but your chances of success are greater either at your company or finding a new role if your company is “AI pilled”.

AI Lies on LinkedIn

You are not going to become AI-pilled by consuming LinkedIn garbage like “Steal These 5 AI Prompts” or “The Prompt Framework For an AI-First CFO.”

If you take nothing else from this post, remember this: The best way to become an AI expert is to just start building with AI.

Not all AI content is useless, but 90%+ of what I have seen isn’t helpful. I will occasionally post interesting stuff I am building with AI, but the objective is not for you to “steal my prompts”. It is so folks can see what is possible.

Just go build something. We are all builders now.

The Levels of AI Adoption

AI feels like magic. Even when you are doing some of the most basic stuff. For that reason, I talk to many CFOs who incorrectly believe their company/teams are AI experts because everything that they are doing feels like magic. The reality is that most of them are just scratching the surface.

There have been quite a few good articles on the stages of AI adoption (see posts by Ann Miura-Ko and Geoff Charles, which inspired some of my framework below).

Level 1 (Noob): The Prompt Guy

Everyone starts here. You have played with AI chat interfaces personally, but now you are trying to use it for work. AI can’t complete anything end to end because nothing is really connected. Everything is individual prompts.

• Reconcile this account

• Summarize this contract and look for rev rec issues

• Review this board deck

It certainly feels magical and you save some time, but nothing has significantly changed.

Finding the noobs? Daily AI usage may be high, but token usage will be very low because they aren’t building real workflow/process changes.

Level 2 (Casual): The “Magic” Feeling

The finance team has some shared context and maybe a few AI tools wired into workflows. AI starts to become part of how the team operates.

• Shared prompt library for variance commentary

• Monthly close checklist with AI-drafted flux explanations

• Board deck narrative auto-drafted from the actuals

Individuals are using AI (and it makes them meaningfully more efficient), but nothing has really changed at the company. Headcount is similar. Each user creates things independently so efficiencies are limited and are tightly tied to people (if someone leaves, it breaks).

Level 3 (Pro): Systems Are Talking to Each Other

Agents work across systems by themselves. Real work is starting to happen with zero or minimal human intervention. This is the stage where folks start talking about layoffs due to AI because roles or entire teams no longer feel needed.

• Agents pull from CRM, billing, and rev rec schedules to flag revenue risks before quarter close

• Spend agent auto-categorizes, flags policy violations, drafts the exception memo

• Shared skills/agents that capture how finance does things

• Non-finance people self-serve 80%+ of their finance questions

Level 4 (Cracked): Company Processes and Org Designs Change

Processes and teams have changed to adapt to AI’s capabilities. All core systems are connected to AI and agents are set up to perform work across systems automatically. Headcount shrinks (at least relative to scale) and org structure changes.

• Procurement, AP, and vendor management collapsed into one person with an agent stack. It used to be a team of five

• Forecasting model auto-updates each cycle and flags variances before FP&A sees them. Company goes from 3 analysts to 1.

• Contract review agent has learned from the last 200 redlines and handles 80% of standard MSAs without legal review. Legal team doesn’t make any more hires as the company triples revenue.

Level 5 (Creative Mode): Infinite Possibilities

In many games, there is a “creative mode”. Unlimited resources, you can’t die, normal game rules don’t apply, etc.

AI is continually advancing. Each new model unlocks new possibilities that weren’t conceivable just a few weeks earlier. The learning and building won’t stop anytime soon. Companies need people who are excited by “creative mode”.

In a couple of months, there might be a better way to do things with AI that you just barely finished. Folks have to be willing to destroy the old process and recreate again (better).

There could be single-person accounting, FP&A, or other departments at larger scale companies for those that reach level 5.

Some Tips on Becoming AI-Pilled

• Create the right culture. Companies need builders. Not process followers. That is a culture shift for many companies/departments (particularly at more scaled companies). Leaders need to be building too. Having leaders be one of the first to build something cool with AI lights a fire underneath everyone else.

  • Make AI part of the hiring criteria. “What have you actually built?” “How would you do [X] with AI?”

  • Fire folks who aren’t adaptable. People unwilling to learn and create need to go. Things are moving too fast to keep them.

• Host department AI hackathons. This gets the juices flowing on what’s possible. We did this for my org and almost immediately the AI projects started pouring in. Folks felt the AI magic and its possibilities. Claude usage spiked immediately afterwards as everyone was experimenting.

• Incentivize AI adoption. Company rewards, spotlights, leaderboards, etc. If you have the right people, this will get everyone excited about AI and they will progress a lot faster to higher levels.

• Give EVERYONE AI tools. Have seen too many CFOs limit budgets here (small token budgets or limits on who gets licenses) because the costs weren’t in the 2026 plan. This is a bad idea. You can’t let spending spin out of control, obviously, but the level of productivity gains far outweighs the cost. Find the budget. Maybe that means cuts elsewhere, but everyone should have an AI tool.

• Don’t centralize AI work. “Our IT/AI folks understand AI better than anyone, so it’s more efficient for them to build and maintain the AI projects.” You will move too slowly if you are doing this. The successful companies will make everyone AI power users with the help of a few AI specialists.

• Create an AI Czar/team. Not to just build stuff for folks. But to train, make sure they have the tools, and help guide/suggest next steps as folks move through the AI adoption levels. Their job is to guide everyone through the AI adoption levels and help take projects to the next level as needed. (needs to be hosted, has security considerations, etc).

In Summary…

Go and build stuff with AI. Experiment and see what’s possible.

The continual learning and required adaptability scares a lot of people (and companies). But it’s going to be needed to succeed.

Join companies that are going to push you. And invest in companies that are AI-pilled. A lot of people/companies claim to be AI-pilled when they are just barely scratching the surface.

Footnotes:

• Check out this International Hiring Guide from Deel if you are looking at expanding internationally. It’s an excellent resource.

• Want to sponsor? Email onlycfo@onlycfo.io

Subscribe now

*nothing contained in this post is investment advice

I’ve seen companies roll out new global comp strategies that drove employees to quit. It felt unfair, was poorly rolled out, or was just a bad comp strategy. It’s hard to get right.

• What pay model to use? Anchor city-based, single global payscale, localized, etc

• How do we balance retaining top talent while also becoming more efficient?

• How do we do equity comp across jurisdictions?

• Compliance considerations? Local HR laws, mandatory bonuses, statutory benefits, overtime, etc

Check out Deel’s guide for a breakdown of key decisions when creating your global comp strategy.

Download the Guide

Private Equity is in Trouble…

We are going to witness a spectacular collapse of many large PE-backed companies over the next couple of years. And the entire tech industry is going to suffer (not just PE-backed companies).

There are two primary stakeholders in a PE leveraged buyout (LBO):

1. Private credit lenders (Blackstone, KKR, Golub, Apollo, etc)

2. Private equity firms (Thoma Bravo, Vista, etc)

Lenders are always first in line to get paid and then anything remaining (if any) goes to the PE investors (and any employees with equity).

Many PE-backed software companies have debt coming due over the next couple of years. When that happens, they (and other companies that took debt) have the following options:

• Repay loan: lol…these companies don’t have the cash for that

• Refinance: unlikely for most given the conditions today

• “Amend & extend” (A&E): kick the can down the road with existing lenders

• Sell the company: not happening right now at valuations that make sense for most of the deals that were done during ZIRP.

• Lender takes the keys: PE equity gets wiped and lenders’ debt is at least partially swapped for equity

“Amend & extend” is the classic playbook, but at a certain point it becomes borderline irresponsible for lenders to keep doing it. Blackstone (largest holder of Medallia debt) refused to “amend & extend” in late 2025 so the PIK toggle expired. And this pushed Thoma Bravo to hand over the keys…

Lenders Take The Keys to Medallia

Medallia is the largest software PE-backed deal to fail to the point where lenders take the keys. But, unfortunately, more of this is coming…

Medallia was taken private at a $6.4B valuation in 2021 (peak of software valuations) by Thoma Bravo at a 9x NTM revenue multiple.

• Lenders: provided $1.8B of debt at acquisition

• Equity: Thoma Bravo and co-investors put in $5B

Thoma Bravo’s $5B stake is now worth zero. Gone. They just handed the keys to the lenders and are walking away. The lenders took control of the company and swapped debt for Medallia equity.

No lender wants to “take the keys” to a company. It’s a last resort. They don’t want to manage companies, but at a certain point it may be the best option left.

The Private Credit Warning Signs

There has been a lot of discussion about private credit over the last couple of months, driven by their large exposure to the deteriorating software industry, which has created an investor redemption problem (large number of investors requesting to withdraw their money) while software loans continue to get marked down.

Once deemed the perfect LBO candidate, software debt is clearly struggling a lot more than every other industry (it’s a casualty of AI).

But folks are realizing that not all software is created equal. Certain software segments are more durable in the age of AI than others. The least impacted? Vertical software...there are riches in the niches!

How are these loans valued?

There is no real market price discovery in private credit because the loans are privately held and not frequently traded. 3rd-party valuation firms are used to determine loan prices and then that is approved by the investment committee each quarter. Obviously lenders don’t want to adjust prices downward until absolutely necessary. Often these loans will look totally fine until they just explode (see Medallia’s loan marks below).

Because each lender has a different 3rd party doing the valuations, the marks on the same loan can be quite a bit different across lenders.

Are loans’ fair values publicly disclosed?

A lot of PE-backed private company debt is held by lenders that are required by the SEC to report details of all their investments, which includes details about the interest rates and current fair value of the assets.

Often at least some of a company’s debt is held by lenders that are not required to publicly disclose their holdings so the public filings won’t tell you the exact amount of total company debt.

If your company has debt you can check to see if it is held by any lenders that are required to publicly disclose it. Just go to EDGAR’s full text search and search your company’s name.

What do the marks mean?

How many cents on the dollar are going to get repaid?

Below is a rough guideline of how I think about these GAAP marks. Remember, if the loans are materially impaired then the PE firm’s holding is likely zero since they only get paid after the debt gets paid.

• >$0.98 (excellent status): This is just the original issue discount that gets accreted back to income for GAAP purposes.

• $0.95 - $0.98 (early stress): Valuation model applies larger discount rate because the asset is riskier. Still expect full principal but pricing in higher risk now.

• $0.85 - $0.95 (restructure likely): Now there is a material chance of principal loss.

• $0.75 - $0.85 (restructuring soon): Priced now as a distressed asset that won’t cover principal.

• <$0.75 (deep distress): “Taking the keys” becomes likely and “non-accrual” status gets triggered. Non-accrual means the lender must stop recognizing interest income because it is unlikely to be collected.

Note - just because a loan hasn’t been impaired does not mean the company isn’t in trouble. Medallia has clearly been a loser for Thoma Bravo for several years (while the loan fair value was still at par). Lenders just thought they could at least get their money back.

Medallia’s Loan Prices

Medallia debt was in “excellent” condition (basically par) for 3 years and then sharply fell to distressed in just a few quarters…

What Happened to Medallia?

There are two ways lenders get their money back:

1. Company profits: uses FCF to pay interest and principal

2. Exit or fundraising event: proceeds pay debt down first

Drowning in Debt

Medallia had one of the lowest FCF margins relative to their growth when they were acquired (bottom decile Rule of 40 score). Thoma Bravo assumed strong growth rates would continue while they turned the company to 25%+ FCF margins.

The LBO loan was underwritten based on Medallia’s ARR (not traditional cash flow metrics). Medallia was burning cash when acquired so the first year of the loan was 100% PIK. Instead of paying interest in cash, PIK allows the company to just add the interest to the loan balance.

Two things with Medallia’s debt made their problem much worse.

1. LIBOR/SOFR: The cost of variable debt increased significantly. The variable portion rose from ~0.75% in 2021 (good old ZIRP days) to a peak of 5.4% in late 2023 and is now 3.7%. That’s a material increase.

2. PIK: Medallia went from not having to pay any cash interest to eventually having to pay 100%. 100% PIK for 1 year → 50% PIK for 1 year → ~35% PIK for 2 years → 0% PIK

Below is what Medallia’s debt waterfall likely looks like based on the original $1.8B. The PIK over the last several years took the balance from $1.8B to $2.2B.

Medallia reportedly has ~$2.8B of debt today. The increase likely came from additional loan draws to fund acquisitions that Medallia performed since Thoma Bravo took it private.

Medallia Debt and Interest: Annual interest on the loan went from ~$135M to ~$300M as the loan balance grew from $1.8B to ~$3B and interest rates increased. On top of that, interest payments slowly went to all cash.

Medallia Profitability: They are reportedly only generating $200M/year in EBITDA.

$200M EBITDA with $300M of interest is broken…

Overpaid, Sinking Valuations, and No Exits

Back in the glory days of 2021, a 9x revenue multiple seemed relatively cheap. The overall median when Medallia was acquired was 36x (!!) and the median for mid-growth (where Medallia sat) was 15x.

PE usually gets liquidity the same way as everyone else: M&A (PE or strategic) or IPO. The typical playbook is PE buys at a good price, they cut costs to drive profitability, and hopefully maintain decent growth rates. Then they can sell in 3-5 years for a higher multiple. And that playbook worked for a long time.

• IPO: not happening for anyone other than high-growth companies.

• M&A: Mostly frozen for larger software deals.

Most folks don’t realize how important PE has been for software company exits…If PE deals slow down then it’s going to impact the entire sector. Multiples continue to fall.

FCF multiples for software have never been lower…

So if we look at Medallia and assume $200M FCF and apply a reasonable FCF multiple based on the market today, you get an enterprise value that in a best-case scenario will barely cover the current loan balance.

Orlando Bravo recently said the following in a CNBC interview about the Medallia acquisition price:

When we bought it, we way overestimated or extrapolated the very high rate of growth of that company into the future. We made a mistake. And that cost us to pay too much.

Are Leveraged Companies Cooked?

Many software companies that took on debt (particularly PE-backed) will follow the fate of Medallia over the next couple of years.

Their debt will create a death spiral that is nearly impossible to escape. Revenue growth has fallen off a cliff, pricing pressure is rising, and AI is disrupting their business. If you are running the business for cash (which they have to in order to pay increasing interest rates) then it will be really hard to innovate in the age of AI.

If you don’t innovate with AI, then your terminal value is definitely going to zero and eventually your profit margins will too. And you are worth less than your debt.

Byron (one of the best software VCs) agrees…

Final Thoughts

The overall theme of 2026: accelerate or die.

But it is MUCH harder to accelerate/innovate when you have a heavy debt burden. This is why most PE-backed companies are in trouble.

For CFOs at companies with debt…talk to your lenders sooner rather than later. It’s going to get harder over the next two years to extend or refinance.

With wishes of “amend and extends” for all,

OnlyCFO 🫡

Footnotes:

• Check out this compensation guide for managing a global company. Deel has some of the best data and resources out there.

• Want to sponsor? Email onlycfo@onlycfo.io

Subscribe now

For finance leaders, managing international payroll and compliance requires expertise. Our ‘Guide to EOR for Startups’ offers a detailed look at efficient global payroll management & compliance. This guide will help you:

• Manage cross-border payroll easily

• Navigate tax implications across jurisdictions

• Reduce financial risks associated with global growth

Download the EOR Guide

10 Reasons Why I’m Churning

Churn is going to get really ugly in 2026. It’s going to break unit economics and put many companies out of business. But…it will also be an opportunity for other vendors to capture more new business.

Some reasons are the same, but AI has created new reasons to churn vendors in 2026. If you own your department’s budget, then you should be thinking about these as well.

Below are 10 reasons that are top of mind (in no particular order):

1. Agent-Readiness Is the New Integration Story

Vendors used to brag about integration counts (and that matters). Now I want to know if your product can be driven by an agent.

• Do you have an MCP connector?

• Are your APIs clean enough that an agent can actually use them, or is everything locked behind a UI?

• Are you the best for AI agents to work with?

Customers will flock to the vendors that are the best for AI and AI agents. Vendors must expose clean APIs/MCP/agent-readable surfaces in 2026. And leaders should be obsessed with doing that (like Pedro).

2. Vendor Consolidation Is Growing

AI is enabling (and forcing) companies to go multi-product much sooner and with a lot more products. But this also means I will be consolidating even more vendors because there will be more overlap.

Recently, one of my vendors added a tool that did ~60% of what I was paying another one of my vendors ~$80K/year for. The 60% wasn’t perfect, but it was good enough. And the remaining 40% gap either wasn’t that big of a deal or we could leverage AI to fill the gap.

We churned, saved $80K, and had minimal impact on our processes (some was actually a bit better because we built a custom tool for those specific features).

3. Trading Nice-to-Haves for Tokens

“Nice-to-haves” have always been at risk of being cut when belts tighten, but two additional things are happening today:

1. AI has pushed more tools into 'nice-to-have' territory.

2. Departments are being tasked with budget tradeoffs. If you want a bigger token budget (most do), then something else needs to be cut.

Headcount is obviously the biggest lever, but as more things move into “nice-to-have”, those become a bigger risk.

4. Stupid Pricing Models

Pricing models have become a key strategic decision with AI. Seat-based pricing was so easy in comparison. It’s much harder to get right today with all the flavors of consumption-based pricing.

I hate nothing more than feeling like I am getting a bad deal. So when pricing doesn’t align with how I perceive value, I feel like I am getting screwed.

Companies are scrambling to update pricing for an AI world (where seats become meaningless for most). And they should. But a few pieces of advice:

• Get cross-functional involvement and test it with customers. And continue to iterate. You can’t set it and forget it anymore.

• Be generous with existing customers. Pricing model changes often feel like it’s a bad deal for them (especially CFOs).

• Don’t sneakily add a price increase with a pricing model change. This happened to me recently and that kicked off a churn conversation.

5. Weak Support

This is increasingly becoming a differentiator. The AI support bots are cool (but they are table stakes).

I want real support (technical support, “forward deployed engineers”, CSM, or whatever you want to call it) that can help me get implemented quickly and be there as needs change, especially as I adopt your AI stuff.

Last year I churned from a vendor that refused to get on a call with me to discuss massive overages. They wanted me to sign an SOW and pay for the privilege of talking with them about why I didn’t understand our bills. That is a problem in itself…

6. Weak Brand Erodes Trust in AI Roadmap

Every (good) tech company has goals in 2026 to materially leverage AI across the company so everyone wants to be on the best tools to leverage AI.

Branding works on me whether I like to admit it or not. A strong AI brand from another vendor opens up the churn conversation. “Am I on the right tool? My vendor can’t do that much with AI.” I want my teams using the best AI stuff.

7. Building Internally

I don’t want to build software internally. No one does (at least they shouldn’t). But it’s increasingly becoming a viable option for many areas.

But even more important than a full internal build replacement (which I think will continue to be rare) is the internal build of add-on features. We still want the core tool, but the individual features can be replaced with some light internal vibe coding or AI magic.

I have already done this with a few features. Not only did we save money, but we were able to customize those features more to our needs.

8. AI Lowers Switching Costs

In the past, I found myself frequently saying the following as it relates to our tech stack:

If I could snap my fingers and be on [X] vendor, then I would. But I am too busy to implement right now, it’s not worth the change risk, we have other priorities, etc.

AI and the AI-native companies have lowered this barrier. This means downgrades will increase for most vendors…

9. The AI Experimental Budget is Ending

We created an AI slush fund for our teams to experiment with AI. But that era is ending. Now we are pushing departments to narrow the list and choose.

You can’t have Claude and ChatGPT enterprise plans. Choose one.

Or the finance department has a lot of overlapping AI forecasting tools. You must consolidate. We want teams to start measuring the value they are getting. If value is weak, then it’s time to cut.

10. I Have Had Time to Prepare for the Churn

AI (and process changes) has changed what I need in my tech stack.

This started happening last year, but it was too soon to churn, so I got stuck renewing for another year. I have a few of those contracts ending this year.

90%+ of the software churn from AI hasn’t hit yet because of this. Time was needed to update processes, evaluate alternatives, build something internally, etc. And folks may have had to sign another one-year contract to bridge the gap.

More churn is coming…they just haven’t had the chance yet.

Churn Breaks Current Unit Economics

If churn is increasing (which it is), then a lot of company unit economics are going to break.

A CAC payback period of ~18+ months is no longer an option for most companies. That only worked in the old SaaS world when we assumed 5+ year average customer retention.

• Goal #1: Figure out how to win in 2026 (with AI)

• Goal #2: Lower your CAC payback period

The hottest AI companies can still fundraise to fund their high CAC paybacks, but most of us can’t do that anymore. And even the fundraising for the hottest AI companies will eventually dry up if CAC stays high.

Higher churn requires a lower CAC payback period for the math to work.

Below are some public and private company CAC benchmarks. They are both too high for the age of AI.

Public companies: median 20-month S&M payback period

Private companies: might have a slightly lower median, but still high. Below are the top-quartile payback periods of a top-decile firm (ICONIQ)

Footnotes:

• Check out the EOR Guide. It’s a must-read if you have international employees or are thinking about expanding more internationally.

• Want to sponsor? Email onlycfo@onlycfo.io

Subscribe now

📚 This Week’s Interesting Things:

Big 4 is getting leaner…

Deloitte is cutting a lot of their generous benefits (parental leave from 16 to 8 weeks, reduced PTO, pension plan frozen, etc). There will be more companies that follow…

And KPMG just fired 10% of their US audit partners.

Slow growth is a valuation death sentence

As I discussed in my last post, investors are assuming slow growth means AI is destroying their business, so it has zero terminal value now. And the valuation multiples reflect that…

Everyone Trying to Get Anthropic Stock

The seller is a banker and really wants some Anthropic stock…This used to be a joke, but now homes in the Bay Area are being priced with pre-IPO equity.

For finance leaders, managing international payroll and compliance requires expertise. Our ‘Guide to EOR for Startups’ offers a detailed look at efficient global payroll management & compliance.
This guide will help you:

• Manage cross-border payroll easily

• Navigate tax implications across jurisdictions

• Reduce financial risks associated with global growth

Take control of your financial strategy today.

Download the EOR Guide

ServiceNow’s Worst Day…Ever

Three days ago, ServiceNow experienced the largest single-day stock price crash in its history (-18%). And it took the entire software industry down with it (software ETFs were down ~6%).

ServiceNow’s Q1 results must have been disastrous to send the stock down ~18%, right?

Nope. They were actually pretty good with a “beat and raise” quarter.

• Quarterly Revenue: $3.770B (22% YoY growth) vs $3.745B consensus (+0.7% beat)

• Adjusted EPS: $0.97 vs. $0.96–$0.97 consensus (in line to slight beat)

• Q2 Revenue Guidance: $3.818B guidance vs $3.748B consensus (1.8% beat)

• AI Momentum: Raised 2026 AI-related revenue target to $1.5B (from $1B).

In more ordinary times, you’d think the stock might be up (or at least flat) with those results. Certainly not down 18%.

But that is not the environment software is in today. ServiceNow’s results were just “pretty good” and the story wasn’t super clean…

Software Investors in 2026: Shoot First, Ask Questions Later

The stock market is not always rational (especially in the short term). While ServiceNow was down nearly 18% on Thursday, some investors realized on Friday that maybe this was a buying opportunity and sent the stock back up 6.5%.

Every software investor has their finger on the sell button and is just looking for a reason to press it. Any reason…

On Thursday, there were 83M ServiceNow shares traded, which has only happened a couple of times prior in the history of the company.

There are two related things that got ServiceNow shot by investors:

1. “Pretty good” doesn’t cut it in 2026

2. There was noise in the financials/story

Anything other than clean, stellar performance is interpreted as “AI is killing the business”. Software investors are looking for reasons to say you are going to get wiped out by AI. And if you look for something hard enough, you will find stuff that supports the “AI just wiped out [X] segment of software” thesis.

And ServiceNow’s results and story definitely gave the thesis room to grow:

• Is AI destroying their pricing power? Q2 non-GAAP operating margin guided at ~26.5% (well below Street expectations of ~30–31%)

• Is there really just a timing issue or is growth just slowing? Management said there was a ~75 bp headwind to subscription revenue growth from delayed on-prem deals in the Middle East due to the conflict in Iran.

• Are they truly winning with AI or mostly just cannibalizing their legacy revenue? Q2 revenue guide was disappointing given the Q1 beat. If they are a major AI beneficiary then why is overall growth not stronger?

• Is ServiceNow buying growth via M&A because organic is quickly weakening? ServiceNow’s $7.8B acquisition of cybersecurity company Armis closed in Q1 and had a noisy impact on results and forecasts.

Bill McDermott, ServiceNow CEO, mentioned their “beat and raise” a couple of times on the earnings call. Below is how I heard it in my head, lol.

C’mon guys…We just beat and raised. Why are you giving us such a hard time!

In the Q4 earnings call Bill actually said the below in response to ServiceNow supposedly dispelling organic growth fears:

I noticed that we lost about $10 billion in market cap on that because of the worry. So now the worry is gone, you can give us back the market cap. — Bill McDermott

Moral of the story?

By default, software investors want to sell. They are looking for a reason. They are practically begging for it. And it’s causing all software names to take a hit.

Not all software or their AI story is created equal, though. For instance, the ones that win in the long term may see earlier short-term pain as they aggressively rebuild for the AI world.

The broad-based fear selling might create some buying opportunities…

Is ServiceNow Cheap Yet?

As a multiple of EBITDA, ServiceNow has never been cheaper. By a long shot. Part of that is simply being a more mature business, but they are still one of the fastest growing public software companies AND they are the 3rd largest in revenue scale.

ServiceNow’s FCF multiple is the 6th lowest FCF/EV multiple for companies growing revenue >15%.

And ServiceNow has one of the highest RPOs of any public software company. They have about two years of current run-rate revenue locked in under contracts.

But the bajillion-dollar question is:

What companies get disrupted by AI versus what companies become more valuable because of AI?

Problem is that 90%+ of legacy software probably sits in the first bucket. And the 10% may temporarily look like losers with cracks in their story as well as they rebuild to become AI winners.

The AI Story: Nothing Else Matters

If a company doesn’t have a clean AI story (with financials to back it), then investors will shoot first.

But in order to create a long-term AI story, companies may need to walk in front of a firing squad. They may take some flesh wounds along the way, but it might be necessary for them. And it could present a potential buying opportunity for investors if the company can become an AI winner.

But many stocks will become “cheap” for a reason. The terminal value goes to zero. And if that is true, it’s hard to be cheap enough for investors to not lose money.

Some folks like to say the following about building companies: “It’s a marathon, not a sprint”. But that is likely bad advice today…if you are not sprinting toward an AI story then the rest of the race may not matter.

Footnotes:

• Check out the EOR Guide. It’s a must-read if you have international employees or are thinking about expanding more internationally.

• Want to sponsor? Email onlycfo@onlycfo.io

Subscribe now

📚 This Week’s Interesting Things:

Insane how fast sales rep sentiment has fallen at ServiceNow…sales reps are the closest to the customers and what they are saying/doing so it’s usually good to listen to them.

I feel this one deeply…those additional token costs got to come from somewhere. And it’s headcount.

Global payroll costs extend far beyond salaries. This guide breaks down international payroll expenses, compliance costs, and hidden fees so you can forecast accurately, reduce risk, and improve cost efficiency as you scale. Built for finance leaders managing multi-country operations.

Guide to International Payroll Costs

Pricing Pressure Will Crush You

AI is lowering the ceiling, squeezing the middle, and pulling out the floor. This is one of the biggest threats to company valuations. And it has completely changed how I am approaching new vendor contracts and renewals.

I am the signer on all our material vendor contracts so I am not only seeing this threat with our customers, but I AM the threat on the vendor side…

The ROI companies get from Claude (and other AI tools) is creating a new pricing anchor that everyone is subconsciously referencing. The effect? It feels like I am overpaying for basically every single other tool.

I hate feeling like I am overpaying…

There are three pricing reference points I think about:

1. Ceiling: Perceived value by the customer

2. Middle: Vendor alternatives

3. Floor: Building internally

The Ceiling is Falling

AI feels a lot like magic today. The perceived value is through the roof because these AI tools deliver much stronger outcomes than what was possible just a few months ago.

And because of this, vendors can price aggressively.

With our AI tool you can 10x productivity. Just think of all the headcount savings you will achieve by leveraging our tool. Based on our discussions, we believe it can save you two FTEs. At an average cost of $140K per FTE ($280K headcount savings), you will be saving $130K because our tool is only $150K!!

And it’s true. The labor market is huge…AI has barely started to make a dent in the theoretical TAM of knowledge work.

The Carlyle Group released some data this week on where companies are finding budget for AI. And guess where it is coming from? Most of the budget offset is coming from internal headcount. AI isn’t trying to replace your SaaS tools. It is trying to replace your people.

CFOs are increasingly finding the budget through hiring freezes (no backfills) and layoffs.

But eventually the AI magic wears off. Not that AI isn’t amazing. Of course it is. But our expectations are also increasing really fast. Human replacement cost is the current anchor price, but that won’t last long.

As mentioned in my tweet above, my new anchor price/ROI is Claude. And as AI adoption in the corporate environment grows, everyone’s expectations will also rise. Many employees have really just started to understand what’s possible with AI over the last few months, and that may be bad news for many vendors…

While the savings from two fewer employees may seem amazing today, tomorrow it will just be the expectation that more work can get done with a lot less.

Our willingness to pay near the human replacement cost will diminish (the ceiling falls) as we start to ask more about alternatives and the possibility of building some things internally.

The Middle is Tightening

AI productivity gains mean:

1. More competition. Anyone can build software today (and many are). More competition = more pricing pressure.

2. Existing vendors are going multi-product a lot faster with a lot more products. A multi-product company can price more aggressively because the incremental product is much cheaper to build, sell, and maintain.

Renewals have often been pretty automatic. Send the renewal to me and as long as you don’t increase prices too much then I will sign it. But I am increasingly pushing an evaluation at renewal for the larger vendors. The price difference for essentially the same functionality has shocked us at least a few times already.

The Floor is Disappearing

Historically, I have viewed the pricing floor as vendor alternatives, but building internally is increasingly becoming a real threat. No one really wants to build internally, but the threat has a lot more teeth today.

Everyone is doing it to some degree. It may not be a full vendor replacement (although that is definitely happening), but building smaller features internally is becoming increasingly common. A lot of downgrades are coming…

Example: I downgraded a large vendor's AI feature

During a recent low 6-figure vendor renewal, I negotiated a 45% decrease in spend. We added their AI features last year, but I churned them on the renewal. It was incredibly valuable when we bought it. Saved a bunch of time.

But…since then my company fully rolled out Claude. And we quickly discovered that we could do 95% of what their AI tool could do for us with some Claude integrations. And the Claude token costs were ~15% of what the vendor was charging us. Plus it was more customizable.

Easy decision to churn that feature.

In Summary…

This post is a little bit doom and gloom, but the very best companies will still maintain real pricing power. However, the gap between the best companies and everyone else will continue to widen. Pretty good doesn’t cut it anymore. Expectations are too high.

My recurring message stands: accelerate or die. Seriously. If you don’t, the pricing pressure will squeeze the life out of you.

Footnotes:

• Check out these offshoring costs by country (both salary and employer burden) so you can decide where to focus global expansion.

• Want to sponsor? Email onlycfo@onlycfo.io

Subscribe now

📚 This Week’s Interesting Things:

Contract Term Length:

This is from me, but it’s related to this post so I am going to share it :)

I have expressed a lot of opinions about multi-year contracts in the past, so you won’t be surprised to hear that I still hate them. At least for 99% of vendors. AI just makes multi-year contracts even less sensible.

“Anthropic, please don’t enter our category 🙏”

Alex summed up exactly how every single software/AI CEO feels in 2026.

Claude Comes for Business Intelligence Companies

Even if it’s not perfect, if it does ~80% of the job (and at a fraction of the price) then it is probably good enough for most companies. And pricing pressure increases…

Payroll is entering a new era with AI-driven compliance, real-time visibility, and higher expectations from the board.

Deel’s Payroll Strategy Toolkit gives finance leaders a step-by-step roadmap to modernize payroll, reduce manual risk, and build a scalable strategy with confidence.

Payroll Strategy Toolkit

Snowflake CFO Retires a Billionaire

Michael Scarpelli joined Snowflake as CFO shortly before their IPO and had stock options worth ~$1.5 billion (at least on paper) just a couple of months after they went public.

It’s rare that CFOs join the “Three Comma Club” (billionaire), but CFOs and other executives are definitely not underpaid…

In this post:

• Breakdown of all the public information on exec compensation

• CFO compensation data across all public software companies

Filings and Proxy Statements

If you are a named officer (or board member) of a public company, you kiss privacy goodbye. We get to see a lot of your finances (at least with respect to a significant part of your net worth).

Everyone Gets Your Offer Letter

When a company goes public, they must file the offer letters of the named officers (at least a total of 5). As such, we have Scarpelli’s original offer letter…

• Base salary: $300K

• Target bonus: $300K

• Equity (stock options): 1.25% of fully diluted with early exercise rights

• Right to invest in Series F: Right to purchase up to 0.25% of fully diluted shares

The cash comp was certainly on the low side given Snowflake’s scale and growth. Equity at 1.25% was on the high side for a CFO joining at the scale Snowflake was at. Not crazy, but high side. The options had a grant date fair value of $20.2M.

The right to invest in the next round is interesting. This isn’t unheard of, but isn’t included for most private company CFOs. More common when near an IPO like in this case.

The offer letter provides helpful context to the next piece of compensation information we get when a company is public…

💡 Check out Deel’s payroll guide to learn more about compensation strategies. They also have great compensation benchmarks.

Form 3 - What You Got?

When going public or when someone gets promoted into a named officer position they must file form 3. It lists the type of equity, the amount, and nature of indirect relationships.

Preferred Stock: We know Scarpelli had the right to buy Series F shares based on his offer letter and now we know he did invest (he also invested in the Series G). He spent ~$15M of his own money investing alongside those two rounds.

Common Stock: Scarpelli has an early exercise provision (common for all execs) so he probably early exercised to get those shares.

Stock Options: This is the equity compensation he got that we saw in his offer letter. Looks like he got another ~330K stock options between his signed offer letter and his initial grant.

Form 4 - What You Doing?

Officers and directors also must report all their equity activity (buys, sells, exercises, etc) on Form 4.

Typically officers sell stock in one of two ways:

1. Net-share withholding: Public companies typically issue RSUs so there is a tax event each time RSUs vest. It’s common to sell (or company withholds) enough RSUs to cover the tax liability. These are automatic and don’t signal anything about an officer’s belief about the company.

2. 10b5-1 planned sales (scheduled sales): Almost all officer stock sales go through this so they have a solid defense against insider trading. These scheduled sales are predetermined (at least 90 days after adoption).

Scarpelli had both of these because he also got PRSUs after they went public and those vested over time after they met the performance conditions. Scarpelli wasn’t shy about selling Snowflake stock….he has sold $600M worth of stock since the IPO.

And he did pretty well timing the top with his largest block. He sold the biggest block of shares ($240M worth) near the peak of Snowflake’s stock price (the golden era of SaaS).

Proxy Statement - How Much We Paying You?

Few people actually read proxy statements, but there is some interesting information in them…particularly around officer and director compensation.

Scarpelli announced he was retiring last year (following Slootman again). Scarpelli was Slootman’s CFO at ServiceNow and Data Domain before joining him at Snowflake. Just a reminder that if you are a good leader, then your bosses will bring you to the best companies.

Base salary: $500K

Target bonus: $500K

Snowflake’s bonus plan is paid out quarterly based on revenue. It had a minimum target of 85% (i.e. <85% got zero) and then paid out linearly through 100%.

A total of 110% bonus pool over-achievement was possible if they met the targets of three “gate” metrics:

• Non-GAAP gross margin

• Non-GAAP operating margin

• Revenue growth of a specific product

New equity awards: $29M

Equity comp is obviously where the money is at…

100% of Scarpelli’s equity comp was in the form of PRSUs (had to hit certain targets to vest). This is somewhat unusual. Most have a combination of regular service-based RSUs and PRSUs, which is what the rest of the officers at Snowflake got.

His target PRSU grant was based on a minimum achievement of 80%, otherwise, 0% would vest (they hit 89% in FY25).

PRSU targets:

• 50% revenue

• 25% FCF

• 25% non-GAAP operating margin

The grant date value of the award (assuming 100% achievement) was $29M. This is based on the stock price at the time of grant of $163 so obviously, if the stock price increases/decreases, his realized value will change. Also, he could overachieve and get up to 120% of the target PRSU. So his grant could be worth up to $36M.

He vests in the PRSUs as follows: 33% upon achievement and then quarterly over the next 2 years.

Still Vesting

The big money potential is often made by joining a pre-IPO company so you can get a large grant and a much smaller valuation before the IPO (like Scarpelli did). But executives can make decent money after going public as well…

A closely tracked number is the market value of unvested awards. How much is the executive getting paid to stick around?

Most of the below is vesting over the next 2 years.

How Much Equity Does Scarpelli Have Today?

Scarpelli officially stepped down as CFO in September of last year. At the time he left, he had $518M net value in Snowflake stock (vested stock options and all of his common stock).

While we don’t know if/when he sold that, since after stepping down as CFO he doesn’t have to report it, based on the last known amount…Scarpelli made over $1B at Snowflake. He actually realized $600M (by selling) and had $518M worth remaining. Also, Scarpelli continues to vest in his previously outstanding awards because he entered into a 12-month advisor contract after stepping down as CFO in September. So he will make another ~$20M as a very part-time advisor. Not bad…

Seeking the Tres Comma Club

Usually you only join the Tres Comma Club by being a founder, but some lucky executives can also become billionaires. Scarpelli certainly did well. There might be a couple of other CFOs that realized more (like the Nvidia CFO), but relative to company size and length of time, I am not sure anyone can beat Scarpelli…

Below is a list of all public software company CFO compensation. Couple of notes:

• I worked with Claude to get all the details. Took several iterations but there could still be some mistakes

• Equity comp is based on grant date FV, averaged over 3 years

Footnotes:

• Great read: Payroll Strategy Toolkit. My friends (and today’s sponsor :) at Deel wrote the playbook for how payroll should work in 2026.

• Want to Sponsor? I am opening up the rest of the year now so email me at onlycfo@onlycfo.io to have your wildest marketing dreams come true.

Subscribe now

*nothing should be considered investment, legal, tax, or any other kind of advice

Global payroll costs extend far beyond salaries. This guide breaks down international payroll expenses, compliance costs, and hidden fees so you can forecast accurately, reduce risk, and improve cost efficiency as you scale. Built for finance leaders managing multi-country operations.

Guide to International Payroll Costs

The Hidden Costs of Stock-Based Comp

With the collapse of software valuations, stock-based compensation (SBC) expense is being debated again. But these discussions usually lack real substance so they aren’t helpful.

The question is not “Does SBC expense matter?” Of course SBC matters. All else being equal, lower SBC expense is obviously better. But the “How much should we have?” and “How to measure it?" are much harder questions to answer.

Bill Gurley (VC at Benchmark) kicked off the discussions last week with an article where he argued that software folks are finally waking up to the real cost of stock-based compensation. Here is his final paragraph:

The current environment is revealing a cost that was always there, hidden in plain sight behind an add-back. The companies that navigate this well will be the ones that start doing the honest accounting — treating SBC as what it always was: compensation, denominated in cash, paid for by shareholders.

We knew that high revenue growth in the software industry was hiding a lot of sins of inefficiency (both cash expenses and non-cash SBC sins). But the promise of extremely high future profitability and the large revenue multiples that came with it were justifications for those inefficiencies.

Since software efficiency metrics and shorthand valuation metrics ignore SBC, the problem only becomes painfully obvious when revenue growth slows.

As they say, “when the tide goes out”…

While I agree with most of Bill’s article, our conclusions on what companies should do likely differ slightly.

I want to address 4 questions in this post:

1. Is the company’s SBC expense too high?

2. Is focusing on profitability a trap in 2026?

3. How should you measure SBC’s impact?

4. Will SBC drive more layoffs?

1. Is Your SBC Expense Too High?

Where I agree with Bill is that SBC expense is basically just like a cash expense. It’s a real cost that has negative impacts on shareholder returns. Essentially all of our metrics and shorthand valuation models completely ignore SBC. That created an environment of ignoring this real cost.

The ultimate long-term valuation of a company is determined by its FCF/share potential. Cash expenses decrease “FCF” while equity compensation (SBC) increases the “share” part. Both are bad for the metric.

High-growth companies:

If growth is strong enough and terminal values are high, then relatively high SBC expenses don’t really matter.

Yes. All else being equal, lower SBC is better. But if you are growing 100% YoY and SBC is kinda high at 5%….does anyone really care?…No.

You are growing the pie a lot faster than you are diluting it, so the tradeoff is acceptable.

Low-growth companies:

Most public software companies (and LOTS of privates) sit in this bucket today. And that’s why SBC is getting a lot more attention.

Revenue growth has slowed and terminal values have been decimated (thanks Claude :), which has crushed valuation multiples.

When this happens, investors value profitability a lot more (particularly near-term profitability). And folks start caring about SBC a lot more because near-term FCF/share drives the valuation.

The expectation has been that when a company’s growth rate slows and the future is uncertain (terminal value falls), then you cut expenses and focus on near-term profitability. This is how you maximize shareholder value given the new investor focus on near-term profitability (including the impact of SBC).

This is the private equity playbook - acquire a struggling business with decent retention (slowing growth, weak profitability, and/or out of control dilution) and prioritize extreme profitability.

However, I think this playbook is going to get a lot of companies killed in 2026 (see next section).

2. The Profitability Trap

The software companies that overly focus on FCF (and cutting SBC) over innovation in 2026 are going to get killed.

Byron is one of the smartest software VCs and he agrees with me…accelerate or die.

In the short-term I can see this playbook appearing to work. Investors may reward them with higher valuations because FCF/share increased while growth is little impacted.

But it’s not the next 6-12 months that matter…

It’s 12+ months out after long enterprise contracts end and customers have time to move off of tools embedded in their processes. It will quickly become a death spiral of churn, slow/negative growth, and weakening FCF margins.

Cutting cash expenses (and/or SBC expense) is not going to save your company in the Age of AI. We are in a “accelerate or die” technology shift.

Am I saying “growth at all costs” is back? That’s a dangerous thing to say, but…kinda. Doesn’t mean you should blow all your money, but if you don’t innovate then I don’t think you will be around in 18ish months…

We will see many companies that took the profits over innovation path that generate 35%+ FCF margins within 12 months. And then we will see them get absolutely crushed and that FCF/share will quickly sink to zero.

Be efficient if you can, but you must innovate and grow.

3. How To Measure SBC’s Impact

There are two ways to look at the impact of SBC:

SBC Expense (GAAP):

This is what most people look at. It’s the accounting expense that gets recorded on the income statement. But accounting rules can be misleading…

There are lots of complex rules, but essentially the accountants come up with a fair value on the grant date and then that same amount is expensed over the vesting period (regardless of what happens to the stock price).

The most common types:

1. Stock options: The default for private companies. To determine the fair value companies perform a Black-Scholes model and then that is the price that gets expensed over the vesting period.

2. RSUs: Public companies usually give RSUs and the fair value that gets expensed is typically just the stock price on the grant date.

Example: $1,000 RSU granted at $50/share with 4-year vesting = $12,500 expense per year over 4 years. It doesn’t matter if the stock price drops to $10 after 6 months. The expense is the same.

So is that $12,500 GAAP expense too high?

Without further information, I have no idea. If the stock trades at a 100x revenue multiple (when it probably should be 10x) then that is probably an awesome deal. Dilution is likely very low. If they had to pay cash instead then it would be a lot more destructive to shareholder value.

Similarly, if the stock trades at 1x ARR but management believes it should trade at 10x then that SBC is a lot more expensive because a lot more shares had to be given for that $12,500 GAAP expense.

The GAAP expense doesn’t tell us how it impacts FCF/share so comparing it across companies with different revenue multiples is pretty useless.

Shareholder Dilution:

Ultimately, shareholder dilution is what matters. There is no cash impact from SBC but existing shareholders’ slice of the pie shrinks and the FCF/share formula is negatively impacted.

There are a few ways you can look at dilution, but all of them are much more truthful than the GAAP expense.

4. Employee Equity’s Impact on Layoffs

Public companies treat employee equity like cash. It’s part of employee’s total comp. Essentially no different than cash. Employees typically get granted a variable number of RSUs to hit a target cash value on the grant date (e.g. $250k of RSUs).

So if a company’s stock price craters…the company has to incur a lot more dilution to provide the same cash value.

Let’s pretend that Amplitude (public SaaS co) hired 100 engineers when the stock price was $83 (glory days of 2021) and they gave $250K of RSUs that vested over 4 years. That’s $25M of SBC expense over 4 years and 301K shares being issued.

Fast forward to today and the stock price is only $5.90. If they wanted to give refresh grants to all those engineers of $250K, then they would have to issue 4.2M shares…The expense on the P&L would be the exact same for those new grants ($25M over 4 years) but dilution is 14x higher 🤯

Even if the company cuts back a lot on SBC, dilution would still be higher given how much software stock prices have fallen. You will also risk losing your best talent if you don’t pay good equity packages. You don’t want that either.

So what is the solution?

You likely have to do layoffs to help pay for the additional dilution required for competitive equity packages.

This is another reason companies are pushing AI adoption so hard internally. They need to lay off people to make the new math work.

• Revenue multiples have compressed

• A lot more dilution is required to pay same comp packages

• Increased investor focus on near term profitability (FCF/share)

• Terminal values have collapsed

If smaller revenue multiples are permanent (which seems likely…), then dilution from employee equity can quickly become a death spiral because the dilution would be significantly higher to maintain comp packages. Also, a lot of public company employees still have grants from when stock prices were much higher…when they get grant refreshes, promo grants, or company makes new hires then dilution will skyrocket.

Everything points to more layoffs coming….

Final Thoughts

The $250K of RSUs you pay to your engineer isn’t less real than their $250K cash salary. It just impacts a different part of the FCF/share equation.

But the analysis is the same…

Are you creating more long-term shareholder value by spending/diluting more today?

If the answer is yes, then keep spending. If no…then you need to cut back (on all expenses). SBC matters so be efficient where you can, but also…accelerate or die.

Footnotes:

• Check out these offshoring costs by country (both salary and employer burden) so you can decide where to focus global expansion. *from my friends at Deel

• Reply to this email if you want to sponsor OnlyCFO

Subscribe now

📚Things I Found Interesting This Week:

Private Credit

Still looks messy. Not sure what will happen but there are plenty of people panicking

And guess where private credit has the most exposure? The industry that seemed like the ultimate gift to private equity firms, software.

I can’t imagine most places will be increasing their exposure in software….

How Low Can SaaS Go?

I am going to put my own tweet here because I think this is insane. Software can apparently go much lower…Domo trades at half of its contractual future revenue!

There are a number of companies that are barely trading above their contractual remaining revenue. This was unthinkable a few years ago, but investors are now saying the terminal value of some of these businesses is near zero. And future FCF is at risk.

Payroll is entering a new era with AI-driven compliance, real-time visibility, and higher expectations from the board.

Deel’s Payroll Strategy Toolkit gives finance leaders a step-by-step roadmap to modernize payroll, reduce manual risk, and build a scalable strategy with confidence.

Payroll Strategy Toolkit

Suing Your Board of Directors

Common shareholders (current and former employees) are suing their company’s board because of a pushed fire sale that allegedly undervalued the company and screwed common stockholders.

The accusation?

Six out of the seven board members had conflicts of interest and did not act in the best interests of the common stockholders (which is required by law). As such, the board members breached their fiduciary duties and are therefore liable to the common stockholders.

How did they breach their fiduciary duties?

Board members can be conflicted in several distinct ways. Below are the most common groups:

1. The VC Directors:

VCs have a “dual fiduciary” problem. They must wear two hats: 1) GPs of their VC firm (need to maximize returns for their LPs) and 2) Directors of the company (need to maximize the value of common stockholders).

In this example, the company was considered a zombiecorn. Growth was too slow and the VCs stopped believing in it. In 2026, you are either a hot, fast-growing AI company or you are dead. There is little in between. The VC Directors wanted liquidity and to move on from a company they stopped believing in. They just wanted to focus on AI stuff so they pushed a fire sale.

After the first potential acquirer expressed interest, the VC Directors pushed hard for board approval to move forward with the sale. The acquisition price was not great. A fraction of what the company was last valued at, but the VCs (with their preferred shares) would get 90% of their money back while common shareholders got nothing.

See the problem?

The lawsuit alleges that the VC Directors created a fire sale so they could quickly be done with the company. The accusation was that a more thoughtful, planned sale would have resulted in a much more favorable outcome for everyone (particularly common stockholders) but that the VC directors didn’t care given their contractual preferred treatment.

Preferred vs Common Exit Waterfall

In the example below, anything between a $35M and $175M acquisition price doesn’t impact the VCs financially. They get the same amount either way in an exit. But common stockholders start to benefit after $35M. If the VC Director isn’t fully wearing their fiduciary hat they might see a $35M acquisition offer and be quick to accept it because they know there is no way the company is worth more than $175M so they would rather take the guarantee now.

2. CEO and CFO May Not Be Innocent Either

In theory, these individuals represent the interests of common stockholders because that is the equity that they have.

But in this case, both the CEO and CFO also had a conflict. The CEO was tired of the grind and was supportive of the fast sale, with contingencies. He had close previous relationships with the PE firm that they sold to and knew they would give him a very large joining bonus and equity in the new entity. The CFO got a similar deal. So while their equity would be worth nothing, they were guaranteed a large cash bonus because of the transaction.

3. Independent Director

Even the “Independent” Director was conflicted…He was referred by one of the VCs. He sat on two other boards that were the VC’s portfolio companies and he was an LP in one of their funds that invested in the company.

I wonder how he voted?….

There are lots of Independent Directors who are not so “independent”.

So What Happened?

The above scenario came from the facts of two separate lawsuits for a breach of fiduciary duties:

Mindbody (public software company): They were public and taken private by Vista (PE firm). Mindbody’s CEO favored Vista and wanted to exit to them so he deliberately suppressed the stock price to make it more attractive and fed Vista insider info about a formal sales process before it began. He wanted liquidity and thought Vista was the best firm to give him more upside post-merger. He was ordered to pay $48M in damages for a breach of fiduciary duties…

Trados (VC-backed software company): Private company that sold for $60M where preferred stockholders received $52M, management received $8M through a retention incentive plan, and common stockholders got nothing. 6 of 7 Directors were conflicted and were found to not have considered the interests of common stockholders and breached their fiduciary duties. Despite a failure in fiduciary duties, they escaped liability because it was determined that common stock had no value before the merger (so receiving zero was fair).

Understanding Your Fiduciary Duties

Many board members and company executives don’t understand their fiduciary duties. There is no required training once you become a board member that tells you what to do (or not to do). A lot of it is common sense. Don’t do unethical stuff… But some of it is not very intuitive.

Who are the fiduciaries?

All board members and the CFO. Even if the CFO isn’t a board member they are a fiduciary.

What does it mean to be a fiduciary?

A fiduciary is legally required to put someone else’s interests ahead of their own.

The three main “fiduciary duties”:

1. Duty of Loyalty: Don’t use your position for personal gain. No self-dealing. No hiding conflicts. If you’re on both sides of a transaction, the board needs to know.

2. Duty of Care: Make informed decisions. Read the materials. Ask the hard questions. “I didn’t know” is not a defense if you didn’t try to know.

3. Duty of Good Faith: Act honestly, don’t disregard obligations, don’t intentionally violate the law, etc.

Who are fiduciaries supposed to protect?

This can be the tricky part, especially when so many board members can be conflicted.

Fiduciary duties run to the common stockholders (as the residual equity claimants), not to preferred stockholders. Preferred stockholder protections come from contract (their stock purchase agreements), not fiduciary duty.

Lessons for Fiduciaries:

You must explore all possible trajectories the company could travel

This is essential to fulfilling Director fiduciary duties when evaluating strategic options (such as M&A), especially when it’s not a great price.

• What is the potential company value under each path?

• What is the probability of each outcome?

I talked to a public company CEO that was taken private and he said the work they had to do around this was exhausting. They knew that once a deal was announced, the lawsuit trolls would appear because it inevitably wouldn’t be good enough for someone (even though the price seemed great). They needed to fully consider every possible outcome and document everything.

Most VC-backed boards are conflicted

Just because the entire board approves something doesn’t mean they have met their fiduciary duties because most boards (particularly VC-backed boards) are conflicted:

• VC Directors are always conflicted as dual fiduciaries.

• Management Directors may also be conflicted if they receive material benefits not available to the rest of the common stockholders in a transaction.

• Independent Directors may be conflicted based on all their relationships

Get non-conflicted party consensus

The key is that, in an interested party transaction, you need to get a majority of the people who aren’t getting any ‘special benefits’ to approve the deal. If you can’t, then you’re asking for legal pain.

Boards that document their deliberations, get independent fairness opinions on M&A, and create special committees can often escape liability even in messy transactions.

The Rule of Common Maximization

Directors owe their primary duty to common stockholders when the interests of the preferred and common come into conflict.

The logic is that common stockholders are the "residual claimants". They're the ones with locked-in, illiquid equity who can't exit. Preferred stockholders (VCs) have contractual protections (liquidation preferences, anti-dilution rights, etc) that are negotiated and enforceable regardless of what the board does. Common stockholders have none of that. All they have is the board's fiduciary obligation to them.

Directors often must act against their own self-interests.

They must do the right thing for the company, its shareholders, and its stakeholders. There is no wiggle room on this rule. For Directors, this is the golden rule.

Being a fiduciary carries significant potential personal liability

The Mindbody CEO had to personally pay $48M in damages. Board members and CFOs can face personal liability, not just company liability.

Make sure the company has good D&O insurance. While it won’t cover fraud or intentional misconduct (like in Mindbody’s case), it will cover honest mistakes.

Fiduciary Duty Lawsuits are Coming

When companies are highly successful and have good exits, then everyone celebrates. It’s easy for Directors to act in the best interest of everyone because it’s good for them too.

But times are tough for many in 2026…I know several companies being pushed to explore M&A today by their VCs. Most of these companies are likely worth less than (or maybe just above) the pref stack so common would get nothing. And that adds a lot of risk for lawsuits and being found breaching fiduciary duties.

Directors must prioritize their fiduciary hat and make sure they are protecting the interests of the common stockholders. And take these responsibilities seriously…

Footnotes:

• Download this guide and up-level your payroll for faster, more compliant processing and building a scalable strategy in the age of AI

• Reply to this email if you want to sponsor OnlyCFO

Subscribe now

*nothing in this post should be considered investment, tax, or legal advice. Do your own diligence.

📚 This Week’s Interesting Things:

OpenAI CFO’s Fiduciary Duty

OpenAI CFO has reportedly expressed concerns about their massive spending and the IPO timeline being discussed. She is trying to keep her fiduciary duties by communicating concerns about the CEO’s targets. Good for her. Many CFOs just shut up and nod to whatever the CEO says.

But apparently Sam Altman (OpenAI CEO) has been excluding her from financial plan conversations, like a recent meeting with top investors about server spend plans. If true, that would seem pretty bad governance…

Software’s Continued Meltdown

When the stock market is down, SaaS stocks are way down. When the stock market is up, SaaS stocks are still down. Whether SaaS is dead or not….SaaS investors are.

When will SaaS catch a break?!?

Global payroll costs extend far beyond salaries. This guide breaks down international payroll expenses, compliance costs, and hidden fees so you can forecast accurately, reduce risk, and improve cost efficiency as you scale. Built for finance leaders managing multi-country operations.

Guide to International Payroll Costs

The 8th Wonder of the World Is Crumbling

Einstein allegedly once said that compound interest was the 8th wonder of the world. Einstein was wrong. At least there was something that took its place for the last 20 years…

Annual recurring software revenue was the 8th wonder of the world. But now AI seems to be destroying such a beautiful thing that so many of us were able to witness (and profit from…)

There are two questions that every company needs to answer in 2026:

1. How recurring is your revenue, really? Not based on historical retention metrics. Those aren’t nearly as meaningful right now. But based on AI today and your current products.

2. If it isn’t truly recurring, what does that mean for your company?

To help answer these questions, I went to the person who understands the value of retention and recurring revenue better than anyone else I know…

Nick Mehta (former CEO of Gainsight) is the Godfather of Customer Success, which largely exists because of the concept of recurring revenue. As such, few people understand this topic better.

So you should read what he has to say…

Lies, Damned Lies and Recurring Revenue

-A guest post by Nick Mehta

Games, Golf and Gartner

In 5th grade, I co-founded a tech company, if you use the term “tech” loosely. My co-founder and I would solve PC video games (“King’s Quest”-types, to completely date me) and print up a guide to solving the game. We sold them in the cafeteria at lunchtime for a whopping $5 a pop! Alas, our empire came crashing down whenever Sierra Online issued a new title. All of that work (work being playing video games) had to be done again.

In college, our golf dot com startup had to convince you to misguidedly buy new clubs each season. Beyond demand for golf balls, customers had no need to come back.

And for my first job, we sold backup software that you paid for one time per server, plus a small annual maintenance fee. All that mattered was getting analyst firm Gartner to like your product.

Romancing Recurring Revenue

As such, when I ran my first SaaS company, I had to google this idea of “recurring revenue.” I learned from the master, as I tore through Marc Benioff’s classic, Behind the Cloud.

From my last startup’s launch in 2013 to its sale to private equity firm Vista Equity Partners in 2021, Annual Recurring Revenue (ARR) became Awesomely Reliable Religion.

To state the obvious, the beauty of the SaaS business model was:

• Invest Research & Development (R&D) upfront to build a product

• Deliver it at good Gross Margins at scale (70%)

• Spend one time dollars to acquire customers

• Capture revenue annually with a high rate of retention

• In the mature phase, you can cut Sales and Marketing and R&D and drink from the fountain of ARR

• You could even raise prices, since customers couldn’t leave

In this way, some companies with sticky businesses eventually got to operating margins over 50%. Alternatively, if TAMs were big, companies could defer monetization and keep growing R&D and S&M, with the comfort that the tap of recurring revenue will be there when they need it.

Times were good!

A Tale of Two Cities: AI and SaaS ARR

It certainly is the best of times and the worst of times.

For AI startups, ARR is more of a “concept of a plan.” Don’t be a buzzkill and ask questions!

The A isn’t always really annual (since most revenue is consumption-based). The first R is certainly not contractually recurring. It’s unclear how much will recur but for now, there is so much top down pressure from companies to buy AI technologies that growth is off the charts. And the last R? Let’s just make sure the auditors don’t ask any questions yet.

For SaaS incumbents, to paraphrase another Dickens tale, ARR feels like the ghost of Christmas past. Just 4 years ago, investors drooled over recurring revenue. A few months ago, I was at a charity event with mostly hedge fund types and every time I told them I worked in software, they shouted out “software sucks!” Nice to meet you too?

Death of Terminal Value

On the surface, it looks like investors simply irrationally corrected SaaS stocks. Under the covers, the truth is harsher. If we go back to SaaS playbook from a few sections back:

• Invest Research & Development (R&D) upfront to build a product

• Deliver it at good Gross Margins at scale (70%) => unclear if margins hold given inference costs

• Spend one time dollars to acquire customers

• Capture revenue annually with a high rate of retention => huge concern that customers attrit quickly given ease of competitive offerings being built with agentic coding

• In the mature phase, you can cut Sales and Marketing and R&D and drink from the fountain of ARR => as such, it’s not obvious you can cut ongoing costs and radically increase profits

• You could even raise prices, since customers couldn’t leave => raising prices seems almost impossible in this climate

Welcome to the Annual Recurring Restart

In the most depressing version of this story, SaaS companies now have to have the following strategy:

• Invest significant R&D every year to keep up as the market evolves at light speed

• Accept the risk of lower Gross Margins

• Think of customer acquisition and retention as more similar than not - every “renewal” is a new sale

• Forget about price increases

• As such, if we assume 60% Gross Margins, 15% G&A (better at scale but scale might be tough), 15% R&D and 20% S&M, we might be looking at well-run companies having 10% Operating Margins (and we haven’t even touched Stock-Based Comp yet!)

Think that’s crazy? Most businesses don’t have the historical stickiness and moats of the high-margin software industry:

Unfortunately, if we assume a 4.3% 10-year US Treasury risk-free yield and the current 5% equity risk premium from Kroll (as of March 2026), we get a discount rate of 9.3%.

While SaaS companies can still grow, investors are skeptical. So a 10% operating margin SaaS business with 5% perpetual growth and a 9.3% discount rate (it would probably be higher since there are concerns about SaaS) would be valued at 2.33x revenue.

What are your options beyond that ridiculous plan? I see 2 at this point for SaaS companies.

Option 1: Man of La Mancha

Perhaps the critics are wrong. Maybe the current storm is temporary. Some CEOs are clearly trying to fight the good fight, to badly stretch Cervantes.

They might take growth expectations down and take a big hit. But they use that money to invest in the current R&D, Sales and Customer Success motions. They batten down the hatches during this current craziness. Wow I’m mixing metaphors.

Option 2: Eternal Sunshine of the Spotless Mind

One advantage of terminal values being so low is that many companies have very little to lose. It’s hard because these businesses have been built with a leadership team that was used to the classic model. It’s incredibly difficult to pull off, but the best move for most SaaS companies is to find the Lacuna mind-erasing machine from the Charlie Kaufman classic and wipe out their old way of thinking.

One of the few successful examples of this approach is Intercom. Intercom was able to stare its core business in the face and sacrifice short term results to go all in on agents. Today, Intercom is one of the fastest growing private incumbent SaaS companies around. I’ll let their founder Eoghan McCabe tell the story, because no one else can do it justice:

Tip Your Wait Staff!

I’ll admit this wasn’t the most cheery post. I hope you take away that:

• The concept of ARR was fun while it lasted.

• We no longer can take it for granted.

• One approach is to fight in the old world.

• Another is to “burn the boats,” Intercom-style, and go to the new world.

In short, software isn’t dead. But it’s no longer recurring.

OnlyCFO Final Thoughts

Recurring revenue didn’t just mean that customers stuck around (although that was an important piece!). It also implied that there were only small, incremental costs required to keep them as customers. That’s what enabled software companies to become so profitable (and valuable).

If you have to substantially reinvent each year, then the current model doesn’t work…

That doesn’t mean all hope is lost, but you do have to adapt. It means more refactoring for AI in order to survive and probably leaner teams as well.

Nick’s definition, “Annual Recurring Restart”, might be the most truthful version of ARR for most companies in 2026.

Footnotes:

*Thanks Nick for sharing your insights!

• Want to know the cost of offshoring by country? Check out this International Payroll Guide that breaks down all the costs you should be considering.

• Reply to this email if you want to sponsor OnlyCFO

Subscribe now

Payroll is entering a new era with AI-driven compliance, real-time visibility, and higher expectations from the board.

Deel’s Payroll Strategy Toolkit gives finance leaders a step-by-step roadmap to modernize payroll, reduce manual risk, and build a scalable strategy with confidence.

Payroll Strategy Toolkit

Figma’s Wild Ride

Figma has been on a wild rollercoaster ride since going public just 8 months ago:

• IPO on July 31, 2025

• 250% IPO pop

• AI disruption fears and stock volatility

• Heavily invests in AI

• 80%+ stock decline since

• Shareholder lockup expired on Jan 27th

Figma went from a 50x NTM revenue multiple right after the IPO to just 12x today…Falling from the second highest revenue multiple (2nd only to Palantir) to the 13th highest (out of 75 public software companies).

Did Figma’s Key Metrics Deteriorate Since IPO?

2nd Highest Revenue Growth

Figma’s revenue growth is still the second highest on an LTM and NTM basis. So this isn’t the culprit of the falling stock price. Although growth fell more than some were hoping (more on this below).

Weakening Faith In Revenue Durability

Figma has decent revenue growth endurance (prior year growth rate / current year growth rate), but it is declining faster than others.

Relative to all other public companies with higher revenue growth, Figma has nearly the weakest growth endurance. In Figma’s defense though, they had a higher starting revenue growth rate so higher growth endurance rates are harder.

Revenue durability is the #1 concern I have when investing today. And it’s not just NTM revenue. It’s 2+ years out. Understanding this far out feels like a bit of guesswork in the age of AI (and it is), but there are some early warning signs…like current weak growth endurance.

When Amazing Retention Might Also Signal a Revenue Durability Issue

Figma has crazy good retention numbers. And they have gotten even better! Almost no one else is doing that right now. The increase was driven by the release of new products (particularly the AI features).

The median public company NRR is much lower (just 109%) and Figma has the second highest NRR (again, second only to Palantir).

So why would increasing/strong NRR be a bad thing? It’s always good, right?!?!

Strong NRR isn’t bad in isolation. But it may signal potential issues when evaluated with other metrics….

Figma’s revenue growth rate has weakening growth endurance but their already strong NRR is rising. This means that revenue growth from new customers is slowing even faster and they are relying on expansion revenue to hold up overall growth.

Figma raised prices last year, introduced new products, and pushed AI products hard. All of this is good for NRR in 2025. But most of it will be hard to regularly repeat.

The ratio of new customer growth vs expansion is an important indicator for long-term growth. It’s why I always review the break out of growth rates at my company.

We don’t have this level of detail for Figma, but we know the new business growth is dropping faster than the overall growth rate. And that is a potential problem for long-term growth rates if it continues.

Rapidly Declining Free Cash Flow:

Figma’s decline in FCF margins is obviously a concern…

Looking at FCF on a quarterly basis gives an even clearer view. FCF margins were amazing in the quarter immediately before the IPO (surprise surprise), but it has been on a rapid decline since then and projected to be even lower in FY26 (just 8%)

So what happened? What is the justification for the shrinking FCF margins?

Expenses ran hot in 2025: With slowing revenue growth you generally expect that expenses will not outpace gross profit additions, but that is what happened. Figma increased gross profit by $208M but added $228M of operating expenses (excluding SBC).

Investing in AI: Their main excuse for declining FCF margins is investing in AI. And it’s theoretically a great reason. If Figma doesn’t invest in AI disruption, then their long-term valuation is toast.

• Decreased gross margins because of higher infra and inference spend for AI products. Figma (like many AI companies) is currently subsidizing their AI features to gain adoption.

  

• Increased R&D costs: Need more people and higher infrastructure costs to build out AI stuff.

• Increased S&M costs: More free users playing with AI stuff (drives S&M costs up). And they need more sales and marketing to sell new products and meet growth targets.

Making these hard financial choices is really hard (especially when you are public), but if you mentally accept that your terminal value will be zero if you don’t then it is a bit easier. This is easier for some companies than others. Figma has to move faster than most…

Also, if Figma can’t see a path to beating their 8% FCF expectations (and revenue growth), then they will likely feel pressure to do a layoff. There isn’t much investor appetite for weak FCF margins without extraordinary revenue growth in 2026.

What’s Next For Figma?

Figma needs to adapt faster to AI than most other companies because 1) their products have a higher AI disruption risk and 2) their customers can more easily leave (i.e. churn).

There are a lot of concerns investors can have with Figma, but based on their financial investments they seem to be making the right (albeit hard) decisions to rebuild as a real AI company.

The big question for Figma (and so many others) is if what they are building will add enough incremental value beyond the base models to be worth the extra cost and can their unit economics survive the pricing pressure that is coming.

Figma has customers that love them and a fantastic team (I know many of them) so I definitely wouldn’t bet against them.

Footnotes:

• Download this guide and up-level your payroll for faster, more compliant processing and building a scalable strategy in the age of AI

• Reply to this email if you want to sponsor OnlyCFO

Subscribe now

*nothing in this post should be considered investment, tax, or legal advice. Do your own diligence.

I (OnlyCFO) recently took a demo with Paraglide and it blew my mind. My team has been wasting incredible amounts of time on billing inquiries and back-and-forth with customers on collections.

Paraglide can automate the majority of this work. It’s high-volume, boring, and repetitive so it’s a great place to leverage AI. Talk to my friends at Paraglide and see what accounts receivable AI agents can do.

See Accounts Receivable AI Agents

What Paths Remain for Software?

David George (GP at a16z) wrote an article titled “There Are Only Two Paths Left for Software” where he argues that there are only two paths left for durable equity value creation (down from 3 paths):

1. Accelerate revenue by 10+ percentage points YoY (through AI-native products)

2. Messy middle of slower growth and not very profitable

3. Rebuild the company to 40-50% operating margins (including SBC)

I agree that the “messy middle” path leads to certain death (of equity value) in 2026. For the past 10+ years, the messy middle path still often turned into a decent outcome (usually via acquisition by PE). But those days are gone. PE isn’t coming to rescue you.

So that leaves two remaining paths…

However, I take a more aggressive stance. There is really only one path remaining. Any other path will be destructive to long-term equity value.

The Profitability Path Still Ends in Tears

Lots of companies are trying to get on the profitability path. It’s the natural thing to do for companies as their growth has significantly slowed. But there are two potential traps that most folks on this path fail to grasp:

1. They are actually on the messy middle path (which is an even worse fate)

   and/or

2. The path feels good until they realize they are driving off a cliff

Let’s go through both of these traps:

1. Escaping the Messy Middle

I believe low-growth SaaS companies can get to 40%+ profit margins in 12-18 months. If they really want to…

Many companies have already made massive FCF improvements as their revenue growth has slowed. Public company median FCF has exploded from 0% to 22% in just ~3 years.

But…an important point that David makes is that most companies that have shifted to strong FCF margins are still unknowingly in the messy middle after considering the impact of SBC expenses. True high profitability requires you to also consider SBC/dilution.

Most public SaaS companies aren’t growing fast enough (or profitable enough) to outrun their high SBC expense. While the increase in FCF margins is great, it isn’t enough…And shareholder returns will be the casualty.

In addition to SBC expense, most companies won’t get their FCF margins high enough because they will experiment with AI products just enough to burn more money without seeing real long-term ROI.

I am already seeing a lot of this.

Companies have a FCF margin target so they can't spend much on innovation, but they need to spend some so they dip their toes in innovation. FCF margins are ~5%+ lower than what they could be. Problem is that customers don’t care about their AI products because it isn’t nearly as good as the competition. Burning cash with no added value.

👉 The penalty for being mediocre has never been larger…and it’s only going to get worse.

2. The Mirage of Profitability Will Lead You Off a Cliff

Becoming highly profitable can feel really good. “We are in control of our destiny”. “We don’t have to rely on VC money”.

Companies that go down this path may feel pretty good for 6-12 more months, but then, without warning, you will discover that the path led you right off a cliff…

Here is what will happen over the next 1-2 years on the profitability path:

Lots of change will be required to get to 40-50% profitability (including SBC)

• Adopt a ton of AI internally

• Massive layoffs

• Cut lots of legacy software

• Raise prices on products

So what’s the problem with this path? It’s worked for PE firms for decades.

AI has changed things. By the time you get within reach of those juicy 40%+ profit margins, they will start to fade. Below are three major things coming that I don’t believe most people on the profitability path are fully considering. And they will eat away at their profits, pushing them back to the “messy middle”, and ultimately cause equity destruction:

Pricing Power Weakens:

If you are not accelerating product, then customer willingness to pay a premium will drop because you aren’t a top product anymore. Customers also won’t accept your annual price increases.

This has always been true to a degree, but expectations are so much higher now.

Pricing Pressure Grows

The number of competitors in the “good enough” category is skyrocketing. If you are not top ~5% in your space, then AI will birth hundreds of copycats that are also “good enough”.

It’s worth repeating: The penalty for being mediocre is massive today (and growing).

Churn Is Coming

Most company leaders are not concerned enough about churn. They are complacent because the churn hasn’t really shown its full impact yet because of:

• Multi-year contracts delaying churn

• Deeply embedded software is making it hard to churn (yet)

• Companies are slow to adopt AI so they haven’t experienced AI use cases yet

Churn will come slowly and then all at once for companies that choose mediocrity.

The Value of the Profitability Path

If you are admitting defeat that your company can’t compete and innovate with AI, then maybe the profitability path is the optimal outcome.

But is it worth it?

If my earlier scenario holds true, then lifetime future profits are maybe 1.5-2x current annual revenue — 40% FCF margins for the first year but then they get rapidly eaten away until they hit zero and you eventually shut down.

Look where “low growth” public company multiples are at…2.7x. And falling further every day.

EV/FCF margins have been on a rapid decline since 2021. It’s at its lowest level ever at 14.3x. Why? Because investors don’t have faith in revenue durability. These companies are generating a lot more FCF, but how long will it last?

We have already seen what the market does to companies with weak growth endurance (ignore C3 which is benefiting as a perceived AI beneficiary). Their valuation multiples have been crushed.

But the future mediocre companies will get punished even worse because future durability will be even worse and profits will be eaten away even faster.

Mediocrity will create an eventual death spiral…

The Only Path Left: Accelerate

“Accelerate or die” is likely the only path left. Any other path just changes the timing and how you eventually destroy equity value.

I don’t have the answers on how you must accelerate and rebuild to become a true AI company, but you need to figure it out. It’s the only path that creates long-term equity value.

For companies with imagination, you will do more with more. For companies where the leadership is just out of ideas, they have nothing else to do. They have no reason to imagine greater than they are. When they have more capability, they don't do more. — Jensen Huang

Final Thoughts

If you don’t think you can accelerate the top line and build AI-native products that will win in the market, then maybe path #2 is right for you.

The reason I say there is only one path is because path 2 (high profitability) leads to outcomes that are rarely worth the journey. Fire most of the company and drastically cut costs to maybe be worth 1.5-2x ARR (or 4-6x FCF)? Not a terrible outcome if you are bootstrapped, but if you raised any amount of money then your equity is going to be worth close to zero anyway. If you are going down path 2, then get highly profitable asap and sell the company as soon as you can. If you wait, then revenue will struggle and profits will eventually fall. Then there will definitely be no buyers.

Get on path 1. It will be a lot more fun (and hopefully more valuable) than the alternative. You can be both accelerating AND highly profitable (many are), but it still requires you to be focused on path 1.

The #1 question teams should be asking:

What changes do we need to make now to actually get on path 1?

Footnotes:

• Check out Paraglide’s AI agents in accounts receivable (today’s sponsor)

• Check out David George’s article: There Are Only Two Paths Left for Software

• Reply to this email if you want to sponsor OnlyCFO

Subscribe now

📚 This Week’s Interesting Things:

Redpoint’s Market Update

Logan Barlett and Redpoint team always do a great job on their annual market updates. There are lots of good slides, but the two below are interesting:

• Riches in the niches….vertical software has been more resilient than horizontal due to the presumably more durable revenue

• The categories at the most risk for disruption and less durable revenue

Thoma Bravo’s Software Update

Below are a couple of interesting slides from Thoma Bravo. Not all software is created equal, but the big question is how durable any existing software revenue will be. Some is certainly more durable than others, but….like I said in this newsletter, I don’t think it really matters unless companies are truly accelerating and building an AI company.

Paraglide (AI platform handling billing inquiries, collections, and disputes) just raised a round led by Bessemer to enable companies to get paid on time with less manual involvement.

My AR teams have always spent way too much time in their finance inbox: answering billing queries, chasing customers, chasing down PO numbers, and handling disputes…It’s high-volume, boring, and repetitive. Which makes it a perfect place to use AI agents.

Get Billing & Collection AI Agents

2026 State of GTM

Below are my favorite slides from ICONIQ’s recent GTM report (2026 State of GTM), some relevant public company GTM metrics, and some thoughts on how I am thinking about the data.

Revenue Growth

Companies with <$100M ARR are showing an acceleration in revenue growth. I think we can likely attribute this to all the new AI-native hyper-growth companies. The $100M+ category is still being dragged down by “legacy software” companies that are slower growth.

Note: below are ICONIQ’s top-quartile growth rates and ICONIQ is already a top-decile VC firm. So it is the best of the best.

There are A LOT more non-AI-native software companies with much lower growth than this. But if you aren’t near ICONIQ’s top-quartile growth, good luck raising money in 2026.

While at a much larger scale, public company revenue growth is just 16% today. However, it seems like we have bottomed and are even slightly accelerating, so we will take the win!

Retention

NRR seems to be holding steady, but…remember that this is top-quartile NRR. The difference between top-quartile and everyone else is going to grow much larger over the next 12-18 months, especially with NRR.

NRR can be a very lagging metric on the health of a business. Things like long contracts and vendor price increases can maintain NRR temporarily but eventually churn will hit.

Public company NRR is still declining and it lines up with the $100M+ ARR bucket from ICONIQ’s report (both 109% NRR).

The thing I worry most about software company valuations is the durability of revenue. NRR and GRR are already lagging metrics, so if these keep declining then valuation multiples are going to continue to compress…It’s why public software stocks keep falling. No one is confident on revenue durability.

GTM Efficiency

The average dollar costs in ICONIQ’s report don’t mean much, but the trend is interesting. Cost for leads and opportunities dropped significantly over the past year, except in the SMB space.

While still up significantly from 2021, CAC payback period has been fairly steady for several quarters for public companies.

CAC payback period should be a top focus metric for companies in 2026. The durability of revenue is uncertain and that has implications for customer lifetime value. Reducing the risk period by cutting the payback period is important.

Sales Commissions

There is no single right answer on how to set quotas and pay commissions, but the rise of usage-based pricing likely means changes from what you were previously doing.

Commission clawbacks are really hard to perform, so I am not surprised most don’t do them. The ideal process is to only pay commissions when the likelihood of a clawback is small, but sometimes that can be hard to balance in usage-based models and you want to pay the rep upfront. I have seen too many cases where the lack of clawbacks creates bad sales rep behaviors…

Quota Attainment

Great to see attainment is trending higher. All else being equal, higher quota attainment here means higher GTM efficiency.

But…a lot has changed in GTM efficiency. For example, higher attainment on an AI product with 30% gross margins looks very different than lower attainment on 75% gross margins. Make sure you understand true complete GTM efficiency.

Pricing Model

Interesting data points as usage-based pricing becomes more popular.

It’s so much easier to try new AI products when it’s pay-as-you-go. Competitive pressure likely drives more companies to that model.

Contract Lengths

There is a combination of things driving shorter contract lengths in 2026:

• Uncertainty in software. No one wants to get stuck in a multi-year contract with a tool that becomes irrelevant in 12 months.

• Pricing model changes. As more companies shift to usage-based pricing, multi-year deals won’t be as common.

• Competitive spaces with unproven ROI. Companies don’t want to overcommit for the new shiny AI thing until value is proven first.

There are only a handful of vendors that I would consider for 2+ year contract terms…Unless a company is mission-critical, the sales reps should not be pushing multi-year deals hard. Let the customer pull you into the conversation if they want it. I dislike SPIFFs for multi-year contracts, especially for SMB deals.

I worry about the competence of any CFO/leader signing up for long duration contracts of non-critical software right now….

Public companies don’t disclose exact contract lengths but they are required to report RPO (remaining performance obligations), which provides an idea about average contract terms.

• “Current RPO” = contracted revenue to be recognized over the next 12 months.

• “Non-current RPO” = contracted revenue to be recognized after 12 months

Below is a sample of software company RPOs that I put together. Long-term contracts are quite common and a reason many public companies haven’t really felt the full impact of churn due to AI yet.

Footnotes:

• Check out this billing and collection AI agent. If you want to cut costs (everyone needs to) and collect more cash, then you should have an AI agent in accounts receivable. Paraglide is today’s sponsor.

• Reply to this email if you want to sponsor OnlyCFO

Subscribe now