Nice article, very thoughtfully written. Yes, COGS will go up but the salary expense will go down because we will use less number of people because of AI, not proportionally though!
When i was working in a company that did many projects, I coded each project in the general ledger so that we know at the end of the day whether we are making money in the project or loosing our pants & shirts. Perhaps, we need to code AI expense by department so that we can account for it correctly?
As usual, we seem to be thinking about the same things, you and I. Though in my case -- at least for this post, I veered away from the gross margin discussion (though I've been thinking and talking a lot about that over the past 6-12 months).
One thing I'm wondering about -- and you'd know more about this than me -- is there a world where if the token consumption is your customer (as opposed to your employees) using tokens which they buy through you because you are delivering the overall solution... are there some scenarios where you as the vendor, can book the NET and not the gross?
Analogous to how some SaaS vendors have historically booked payments revenues (net vs. gross).
If true, this might mitigate the negative impact on gross margins associated with the cost of the customer-facing token usage.
the answer will likely be no 99% of the time though because they aren't simply reselling tokens. It's part of their product and the vendor controls how it is delivered. And the vendor is responsible if the product sucks or fails. Not the base model companies.
There are scenarios in which this revenue could be booked on a NET basis, but only if the reporting entity concludes that it does not control tokens before transfer to customers. If you are able to negotiate terms in which any pricing changes from the upstream vendor are passed through to your customer, and you do not need to purchase tokens in advance of customer orders. With carefully designed contractual terms, it is definitely possible.
The SaaS valuation framework was built on 77% gross margins. Every revenue multiple and every LTV/CAC model that justified premium SaaS valuations assumed those margins were structural and permanent. When AI token costs move into COGS and compress margins toward 52%, the entire valuation architecture that priced a generation of software companies has to be rebuilt from the gross margin line up.
Thats the investment consequence your P&L article implies without crossing into. The CFO sees a classification problem. The investor should see a repricing event. A company trading at 15x revenue with 77% margins is priced very differently from the same company at 15x with 52%, but the market hasnt adjusted because the AI COGS line is still blended into hosting where the board can't see it. Your advice to break it out separately would fix the accounting and force the valuation problem into the open simultaneously.
Nice article, very thoughtfully written. Yes, COGS will go up but the salary expense will go down because we will use less number of people because of AI, not proportionally though!
When i was working in a company that did many projects, I coded each project in the general ledger so that we know at the end of the day whether we are making money in the project or loosing our pants & shirts. Perhaps, we need to code AI expense by department so that we can account for it correctly?
Love this... By coincidence, today I created a post on token flow that probably hit LinkedIn within minutes of yours: https://www.linkedin.com/feed/update/urn:li:activity:7465081155164258304/
As usual, we seem to be thinking about the same things, you and I. Though in my case -- at least for this post, I veered away from the gross margin discussion (though I've been thinking and talking a lot about that over the past 6-12 months).
One thing I'm wondering about -- and you'd know more about this than me -- is there a world where if the token consumption is your customer (as opposed to your employees) using tokens which they buy through you because you are delivering the overall solution... are there some scenarios where you as the vendor, can book the NET and not the gross?
Analogous to how some SaaS vendors have historically booked payments revenues (net vs. gross).
If true, this might mitigate the negative impact on gross margins associated with the cost of the customer-facing token usage.
Just wondering...
goes to principal vs agent guidance in ASC 606
the answer will likely be no 99% of the time though because they aren't simply reselling tokens. It's part of their product and the vendor controls how it is delivered. And the vendor is responsible if the product sucks or fails. Not the base model companies.
There are scenarios in which this revenue could be booked on a NET basis, but only if the reporting entity concludes that it does not control tokens before transfer to customers. If you are able to negotiate terms in which any pricing changes from the upstream vendor are passed through to your customer, and you do not need to purchase tokens in advance of customer orders. With carefully designed contractual terms, it is definitely possible.
The SaaS valuation framework was built on 77% gross margins. Every revenue multiple and every LTV/CAC model that justified premium SaaS valuations assumed those margins were structural and permanent. When AI token costs move into COGS and compress margins toward 52%, the entire valuation architecture that priced a generation of software companies has to be rebuilt from the gross margin line up.
Thats the investment consequence your P&L article implies without crossing into. The CFO sees a classification problem. The investor should see a repricing event. A company trading at 15x revenue with 77% margins is priced very differently from the same company at 15x with 52%, but the market hasnt adjusted because the AI COGS line is still blended into hosting where the board can't see it. Your advice to break it out separately would fix the accounting and force the valuation problem into the open simultaneously.