Hebbia is an ambitious genAI doc analysis startup, until now focused on the financial sector. With backers such as Peter Thiel and Jerry Yang, and with big chunks of VC cash, it is now targeting the legal world. Artificial Lawyer spoke to Ryan Samii, its new Head of Legal, about the move and also saw the product at work.
But, before we get into the legal sector strategy, it was reported earlier this week by TechCrunch that Andreessen Horowitz and others were putting $100m into the company for a Series B. (It did a $30m Series A investment previously). However, Hebbia told this site that the TechCrunch story ‘didn’t have all the details right’. That said, let’s just take it as read that Hebbia is getting the attention of some very significant investors and plenty of more money is going into the business.
Right, back to the legal strategy.
To understand that, you first have to understand what Hebbia, which started in 2020, does in the financial world. As demonstrated to AL by Samii, one of its key features is what they call the Matrix, which is where they are able to place multiple, long and complex documents side-by-side in rows on your screen, and then extract and compare data across the entire corpus, via multiple prompts.
This allows what would otherwise be a very intensive and challenging piece of work to be reduced to a much more fluid process, topped off with natural language Q&A. For example, a bundle of side letters from participants in a major investment can be examined and queried all at once in unison, with key data extracted and comparisons made based on what you want to know.
This allows Hebbia to ‘take on the rote work of financial analysts, which could include comparing earnings reports, or [investor] call summaries, so investors can make decisions’ Samii explained.
The company also stresses that it is multi-modal and multi-model, i.e. it can work with different LLMs and also different types of data, in fact it states: ‘AI should work over any type of information and any modality of data. Whether it’s a scan of a picture, or tables nested inside other tables, Hebbia seamlessly ingests your most complex documents.’
The company adds that it can ‘leverage your firm’s knowledge over any LLM (OpenAI, Anthropic, and more) orchestrated to cite its answers’. And that the Matrix ‘reads all file types and hooks into any source of private data in addition to millions of built-in sources, e.g. SEC Filings, earnings transcripts and expert calls’.
Samii also said that they have been developing ‘AI agents to take on specific steps in KM work’. They have also developed what they believe to be a highly accurate approach that goes beyond traditional RAG methods.
‘RAG does not lend itself well to multi-step queries,’ said Samii, ‘if there are multiple sections and multiple docs RAG runs into issues.’
He explained that the company’s approach was not ‘compute heavy’ and allowed ‘for more analysis of the materials’, which it then ‘mapped back to’. This site mentioned that it still sounded rather like RAG, however, Samii stated that their approach did not need the same type of ‘context’ seen in RAG methods.
‘Our approach is simultaneous and parallel,’ he added. ‘There is query decomposition and there are rules-based elements. This approach is lean and the compute costs are affordable.’
He also noted that their approach leads to very high levels of accuracy.
The Legal Market
So, that’s a taste of what they do. Now they have hired Samii, who previously led his own legal AI startup, Standard Draft, and before that he was a lawyer at Paul Hastings. He will now lead the legal sector push. He has been Los Angeles-based, but has moved to New York where Hebbia is based.
Now, whether the company has bagged $100m or a bit more, or a bit less, doesn’t really matter. The key point here is that we have an AI company, with serious backers, with an established approach to handling complex document sets, now seeking to carve out some market share in legal.
Samii told this site that they have already had multiple meetings with AmLaw 100 firms. And what they do in the financial sector could easily find uses in legal. Want to compare multiple contracts, whether in a big corporate deal, a financing, a real estate deal? Want to compare patents for IP work? They can do it. Want to compare depositions at a very granular level? They could do it as well. They could also do the same for legislation, perhaps case law too.
The reality is they are just getting going in legal and the use cases will no doubt keep growing. Samii also noted that they are putting together a sales team to sell into the legal world and surprisingly given the investment levels, they are only around 50 people at present, although that is sure to grow.
So, there you go. This is not going to be the last we hear of Hebbia, that is for sure. Moreover, as with other new generation AI startups, one aspect that seems clear to this site is that although they have started with one or two key features, in this case their Matrix approach, as money arrives, time passes and more engagement with customers takes place, then new features will evolve. After all, Hebbia started in 2020 and it’s only just started to engage with legal. They could take this very deep into the legal needs that relate to doc and textual database analysis in the next couple of years.
Whatever happens, it’s another sign that legal is still seen as a growth area for startups and we can expect the ecosystem here to get both busier and to keep on evolving.