Legal AI, ranked by the data

Legal AI has the clearest commercial logic of any market in the catalog. Firms bill by the hour, so time saved converts into money without an argument — a luxury almost no other AI buyer offers. And yet only 83 products are competing for it, and more than one in eight of the ones we have tracked are already gone. Both halves of that are worth explaining.

A market with an obvious buyer and few sellers

Most AI companies have to construct their return on investment. They estimate hours saved, apply a loaded salary cost, and hope the buyer accepts the arithmetic. A law firm does not need the argument constructed: it already knows precisely what an hour of associate time is worth, because it sells them.

So why have only 83 products taken the opportunity? Because the same thing that makes legal work valuable makes it unforgiving. A product that invents a citation has not produced a slightly worse answer; it has produced a professional-conduct problem for the person who relied on it. Passing that bar takes retrieval over verified sources, an audit trail, and the patience to earn a partner’s trust — none of which a good prompt provides. The market’s average wrapper-risk score of 12.1 is that barrier showing up in the data.

The four jobs the leaders have taken

The top of this market by opportunity score divides cleanly. There is research and drafting — the assistant that reads case law and produces a first version. There is contract lifecycle work, where the volume is high and the judgement per document is low. There is review at scale, reading thousands of documents for a diligence or disclosure exercise. And there is compliance, which is really a monitoring product wearing legal clothes.

What none of the leaders do is offer general legal advice to consumers. That market has the worst properties available: unlimited liability, no budget holder, and a free general-purpose assistant as the incumbent competitor.

The uncomfortable number

Legal sits eighth from the top of our failure table. 12 of the 95 legal products we have tracked no longer resolve to a working product — 12.6%, against a catalog-wide 8.4%. A market can be genuinely under-served and still bury the people who entered it, and legal is the clearest case of that in the catalog.

The explanation is in the sales cycle rather than the product. A firm evaluates for months, pilots with one practice group, and buys on a committee. A company with eighteen months of runway and a twelve-month sales cycle has one shot. That is a financing problem, not a technology problem, and it is the single thing most likely to kill a legal AI company that has built something good. The full ranking is in our mortality report.

Nobody here is doing bottom-up

Only 26 of 80 offer a free tier, against roughly two thirds across the catalog. Free trials do not work well here because the person who would try the product is not the person who can approve it, and because a lawyer cannot evaluate a drafting tool on anything other than real, confidential matter documents.

Still an early-stage market

Seed leads at 34 companies, which makes legal noticeably earlier than healthcare, where Series A dominates. Read together with the failure rate, that is a market where a lot of companies are currently between their first product and their first repeatable sale — the exact stretch the mortality data says is dangerous.

If you are building here

Two things follow from the data. First, raise or price for a long cycle: the companies that died here mostly did not run out of product, they ran out of months. Selling into one practice area with a sharply narrow product beats selling a platform to a whole firm, because it shortens the committee.

Second, decide early what you own that a model provider does not. Verified sources, a document corpus with rights attached, an integration into a practice management system, or a partner who will put their name on the pilot. In a market this unforgiving, the tools that survived had one of those before they had scale.

The market, product by product

Ranked by opportunity score. Each name links to the product’s own site.

Photo: Kaboompics / Pexels. Colour-graded for Falcoscan.

← All articles