This is rich in insight and truth! From the other side of the table: across 57 deals, I watched buyers forgive ugly gross margins plenty of times. What they never forgave was a CFO who couldn't explain clearly where the margin was going and what was driving it (and path to remedy). A credible path at 40% beats a messy story at 60% every time in diligence.
The key test is whether pricing and routing gains survive a full cohort of heavier users. Would you split gross margin by usage decile to distinguish durable unit economics from light-user subsidies?
Full disclosure that maybe this was addressed in the X responses and I didn’t bother checking, but a good law firm better be paying for Enterprise and not just pro plans for each employee. The math assumes a $20 / month pro to get to $12k a year and that means the law firm isn’t paying for the security and privacy or enterprise tooling that you actually need when dealing with private legal matters
yeah, I had the same thought - would definitely reduce the cost delta for enterprise. But wouldn't be surprised if some small law firms were just on the $20/month subcription
I agree. I think that's true for most SMB's who offer Claude or Codex to their employees. Besides security and governance (which a law firm needs to prioritize), the enterprise versions allow for easier standardization of skills or agents across an organizations which is the exact kind of approach that a law firm needs if they're going to opt out of Legora or Harvey.
lots of diversity in practice from what I have seen/heard, but it should all costs to develop net new functionality just like before.
you can argue that most fine-tuning (for example) is making iterative improvements on existing functionality, but I would think something like Harvey did (post-training a model) to the extend they did would mostly get capitalized and amortized into COGS. or at least it should...But then fine-tuning after that probably wouldn't for the most part
A very insightful read. I think this is also because of the VC mentality of growth over profitability. The LegalTech market is extremely competitive as well with many players like SpellBook, Clio, Thomson Reuters, OpenAI, Anthropic and Ironclad etc. Companies are trying to capture the market because of the intensifying competition. Profitability is key to building long-term success. I think Legora has shifted to a consumption-based model as well so we might see gross margins improve.
I write on LegalTech analyzing startups, trends and opportunities in the space. I'd love to know your thoughts on my blog.
The thin wrapper test is my favorite part of this. Watching what happens to demand as margins go from 0% to 40% tells you much more than arguing over whether something is "just a wrapper." The COGS list stood out too. If free trials and model subscriptions land in the wrong bucket, the gross margin number everyone is debating could be wrong before pricing even comes into it.
Great breakdown of when bad AI gross margins are acceptable. PMs increasingly have to make that call too. Disclosure, I built https://www.allthingspm.app/, an AI PM course with a chapter on AI cost and unit economics.
This is rich in insight and truth! From the other side of the table: across 57 deals, I watched buyers forgive ugly gross margins plenty of times. What they never forgave was a CFO who couldn't explain clearly where the margin was going and what was driving it (and path to remedy). A credible path at 40% beats a messy story at 60% every time in diligence.
Very true. Have seen a messy gross margin story cause a big haircut on valuation in an acquisition
The key test is whether pricing and routing gains survive a full cohort of heavier users. Would you split gross margin by usage decile to distinguish durable unit economics from light-user subsidies?
100% - figure out what the heavy users are doing and how reduce AI costs based on their usage
Full disclosure that maybe this was addressed in the X responses and I didn’t bother checking, but a good law firm better be paying for Enterprise and not just pro plans for each employee. The math assumes a $20 / month pro to get to $12k a year and that means the law firm isn’t paying for the security and privacy or enterprise tooling that you actually need when dealing with private legal matters
yeah, I had the same thought - would definitely reduce the cost delta for enterprise. But wouldn't be surprised if some small law firms were just on the $20/month subcription
I agree. I think that's true for most SMB's who offer Claude or Codex to their employees. Besides security and governance (which a law firm needs to prioritize), the enterprise versions allow for easier standardization of skills or agents across an organizations which is the exact kind of approach that a law firm needs if they're going to opt out of Legora or Harvey.
Great post! I’ve heard about Harvey’s negative gross margins but your post provides much more nuance!
Thanks Les!
Thanks *Les Barclays* - got cut off which makes less sense :)
Great article
Have you seen anything on what portion of model development or related R&D might be capitalized and roll through COGS
lots of diversity in practice from what I have seen/heard, but it should all costs to develop net new functionality just like before.
you can argue that most fine-tuning (for example) is making iterative improvements on existing functionality, but I would think something like Harvey did (post-training a model) to the extend they did would mostly get capitalized and amortized into COGS. or at least it should...But then fine-tuning after that probably wouldn't for the most part
A very insightful read. I think this is also because of the VC mentality of growth over profitability. The LegalTech market is extremely competitive as well with many players like SpellBook, Clio, Thomson Reuters, OpenAI, Anthropic and Ironclad etc. Companies are trying to capture the market because of the intensifying competition. Profitability is key to building long-term success. I think Legora has shifted to a consumption-based model as well so we might see gross margins improve.
I write on LegalTech analyzing startups, trends and opportunities in the space. I'd love to know your thoughts on my blog.
The thin wrapper test is my favorite part of this. Watching what happens to demand as margins go from 0% to 40% tells you much more than arguing over whether something is "just a wrapper." The COGS list stood out too. If free trials and model subscriptions land in the wrong bucket, the gross margin number everyone is debating could be wrong before pricing even comes into it.
Great breakdown of when bad AI gross margins are acceptable. PMs increasingly have to make that call too. Disclosure, I built https://www.allthingspm.app/, an AI PM course with a chapter on AI cost and unit economics.