Creating an Autonegotiations Strategy for Investors Rights Agreements

When I created Ozeki, I decided to build software for negotiating SaaS MSAs.  I have negotiated thousands of these agreements, so it made sense to start with something close to home.  However, our tech has been improving so rapidly, I wanted to see if I could push its limits and broaden my horizons at the same time.

In my career, I have been fortunate enough to help place about a quarter of a billion dollars of venture debt and equity for my clients.  And, while I have experience in capital formation, I don’t have expertise.  I have experience working with two or three investor rights agreements.  I have negotiated one, but only as second or third chair.  I wanted to become an expert on IRAs. So, time to boot up the Ozeki Machine.  

I started by deciding I needed a true north agreement for my strategy so I would have a best case for any negotiated outcome.  I can read any IRA and develop an informed opinion as to whether it is favorable to my client.  What I can’t do is draft an IRA from scratch or tell what’s missing from an incomplete agreement.  To get a reliable template IRA, I didn’t run off to my favorite foundational model and prompt it to draft an agreement.  Instead, I went to the National Venture Capital Association and downloaded their template Investor Rights Agreement from their model documents center. I then proceeded to harvest 15 IRAs from the Securities and Exchange Commission’s EDGAR database from filings in connection with S-1 registration statements. One important thing to note is that while the NVCA IRA is a standard template, the other 15 sample IRAs were completely heterogenous.  These agreements were created and negotiated by top-tier law firms, including (but not limited to) Cooley, Wilson Sonsini, Latham & Watkins, and Fenwick & West.

I uploaded all of the documents to Ozeki, which took less than a minute.  Ozeki’s AI processed each IRA and compared it to the NVCA’s template, taking between two and ten minutes each to complete the analysis.  The result was a strategy consisting of thousands of different dealpoints across 83 different contract clause types to negotiate over.  Some of these dealpoints were not in the original NVCA template, but Ozeki collected them from the sample IRAs.  Further, Ozeki created a strategic score for each dealpoint so I would know what would be important to trade away to get the optimal deal. When I started this experiment a few hours previously, I had some familiarity with these types of agreements.  However, I could not be confident that I would get bested by another party whose lawyer had a more extensive background in these types of transactions.  With a universe of thousands of dealpoints spread across 83 of clause types, I am confident that my computer-augmented expertise can handle such negotiations and come out with a solid deal. 

Ozeki’s computer-augmented expertise is derived from the experience of the top talent at the top Silicon Valley law firms.  This goes way beyond training an LLM.  Any fool can feed documents into a transformer and get it to spit out the next most probable words based on a Bayesian model.  We have done something way more powerful.  We used our AI to conduct an analytical comparison of the best chess moves of the best players.  I also did not just run off to my favorite foundational model and give it prompts expecting it to do my job for me.  Rather, I used AI to supplement my knowledge and judgment to develop deep subject matter expertise in just a very short time.  If you use the technology properly, AI won’t take your job.  It will, however, turn you into a super expert.