Across North's own deployments, firms that left AI to individual initiative before working with North saw real usage settle at about 9% of the firm. After a structured, firm-wide rollout, that figure lands closer to 80%. Those are North's internal numbers from its own client work, not industry research.
The more useful question is why a range that wide is even possible, and what a law firm AI adoption rate would have to count before it meant anything. Two firms can buy the same model, on the same terms, in the same month, and end up in completely different places. The difference sits in what happened after the licenses were issued, and in whether anyone was measuring the right thing while it happened.
The industry's adoption numbers measure access
Here's the problem with how the law firm AI adoption rate usually gets reported. Almost every headline figure counts availability, not use.
The ILTA 2025 Legal Technology Survey, released in September 2025 and covering nearly 600 firms, is the clearest illustration available. Eighty percent of respondents said their firms are using or exploring generative AI. That's the number that ends up in the trade press.
Then look at what deployment actually looks like in the same survey. Microsoft 365 Copilot is present at 68% of firms, but nearly 60% of those are still testing or piloting it, two-thirds issue it on request rather than by default, and only 8% have it deployed for all employees. The pattern repeats across the legal-specific tools: 71% of Harvey users are still in pilot, over 90% of iManage AI users, over half of Lexis+ AI users, and 60% of CoCounsel Core users. Westlaw AI-Assisted Research at 40% fully deployed and Laurel.AI at 36% are the outliers, and they're outliers at well under half.
Eighty percent and eight percent come from the same survey, and both are true. The eighty is every firm using or exploring generative AI anywhere in the building. The eight is the share of Copilot firms that have actually put it on every desk. The survey is careful about which is which. The trade coverage is not.
This is the reporting practice worth being skeptical of, and it isn't confined to any one vendor. "Using or exploring" is a category a firm joins by running a three month pilot for eleven people. Seats licensed is a purchasing metric. Neither tells a managing partner whether the firm's work changed.
What the people using it actually report
Frequency data suggests the tools genuinely are in daily hands. Thomson Reuters surveyed 1,816 professionals across 62 countries in law, tax, audit, accounting, compliance, risk, and global trade in March and April 2026 for its Future of Professionals Report 2026, released in June. It found that 74% now use AI tools several times a week, and 44% report using them multiple times a day.
That's real usage, and it should be read alongside what the same report says those tools are producing. Ninety-one percent said their organizations are falling short of what the technology could deliver. And even inside organizations that had named an AI strategy, 35% of professionals said that strategy wasn't visible in their day-to-day work.
So a profession that touches these tools daily largely believes the firm isn't getting the value.
What separates the two ends
The same Thomson Reuters research points at the difference, and it's structural rather than technical. In firms and departments with a named AI strategy, 66% of professionals said AI was meeting or exceeding expectations. At organizations with no active strategy, that figure dropped to 22%.
Look at what firms in the low group tend to have in common and the picture gets concrete. Access was granted on request, which quietly makes usage a function of individual curiosity, not firm process. The tool was introduced as a capability rather than attached to a named piece of work, so no partner could say which matters it was for. Training was never provided at all: in the 8am 2026 Legal Industry Report, published in the March/April 2026 issue of ABA Law Practice Magazine, 54% of legal professionals said their firm had given no training on responsible generative AI use and had no plans to. And nothing was instrumented, so six months in, the honest answer to whether it was working was that nobody could say.
Here's what that looks like on a matter. A firm buys access in January. A corporate associate uses it to summarize a diligence index and finds it useful. Nobody tells the real estate group, who are still abstracting leases by hand, or the litigation paralegals, who are still building deposition prep binders the same way they did in 2023. The associate's habit never becomes the firm's process, because nothing was in place to carry it there. By June the firm has a licensed tool, one satisfied user, and no way to describe what it changed.
There's a related cost to leaving that space empty. In the Thomson Reuters data, 34% of professionals said they use AI tools their organization hasn't sanctioned, rising to 41% among those who say their organization is moving too slowly. The report describes that association without assigning a reason for it. For a law firm the risk is the same regardless of motive: client material moving through systems the firm hasn't reviewed and can't see.
What a real adoption number counts
If a firm wants to know its actual adoption rate rather than its access rate, the metrics have to survive a partner's skepticism. Four hold up.
Active users in a normal month, measured well after launch. Launch month numbers are close to worthless. Month four is a real reading.
Matters touched. Not sessions, not queries, but the count of live matters where the system did something that ended up in the file. This is the metric that connects AI usage to the thing the firm actually sells, and it's the one most firms don't have.
Depth per user. One lawyer running forty document reviews and thirty-nine lawyers running one apiece produce the same total and mean entirely different things about the firm. Averages hide this completely, which is part of why they get reported.
Breadth past the enthusiasts. Every firm has four people who'd have adopted anything. The real test is whether the fifth through fiftieth are using it, and whether people who never attended the original training are running skills that somebody else built.
That test should apply to any number a firm is asked to believe, including North's. The 9% and 80% at the top of this article are North's own deployment data, measured across its client firms after rollout, and a reader is entitled to put the same four questions to them that they'd put to any vendor's press release.
Reporting on any of this requires the deployment to be governed in the first place, which is the practical argument for running AI inside a controlled environment instead of handing out logins. North OS exists to make firm-wide use consistent and reviewable, and the measurement follows from that. If the firm's skills for NDA review, closing binders, conflict checks, and diligence indexes are governed and shared, then what's being used, by whom, and on which matters is a question with an answer.
What this means for a managing partner
The uncomfortable version of all this is that a firm can be entirely typical, sit inside every favorable industry statistic, and still be getting close to nothing back.
If your firm has AI access, the diagnostic is short. Can you say how many distinct matters it touched last month? Can you name who's using it outside the group that volunteered? Has anything about how a piece of work gets done actually changed?
If those questions don't have answers, then the firm doesn't have an adoption rate at all. It has a purchase.
Start by picking a single workflow the firm runs constantly, an NDA review cycle, an intake and conflicts process, a closing checklist, and instrument it properly: who runs it, on how many matters, and what happens to the output. One measured workflow tells you more about whether your firm can adopt AI than a year of pilots.
That's the problem North OS is built to solve, and the training half of it is covered in Claude Training for Law Firms: How to Turn Access Into Adoption — companion post, drafted and ready, link once both are live on north.law. For the wider implementation picture, see AI for Law Firms: A Practical Guide to Firm-Wide Implementation.
Sources: ILTA 2025 Legal Technology Survey, as reported by Law360 Pulse (released September 2025); Thomson Reuters Future of Professionals Report 2026 (global survey of 1,816 professionals, fieldwork March–April 2026, released June 2026); 8am 2026 Legal Industry Report, via ABA Law Practice Magazine (March/April 2026 issue).