The problem
A law firm sells expert time, and much of that time goes to work that is necessary but repetitive: reading and summarizing documents, first drafts, research and review. When that work takes longer than a client will pay for, the firm writes off the hours. CMS names the business pressures directly: it wants to reduce write-offs, reduce the load on lawyers, offer competitive prices to clients and maintain profitability (CMS).
The firm also had to decide how to buy. It runs a rolling program of testing generative AI tools and has evaluated more than a dozen, so the question was not whether to use AI but which tool deserved an enterprise-wide commitment.
What they did
CMS began using Harvey about a year and a half before the firmwide deal, with a small trial of around 300 licences, then expanded to about 3,000 lawyers roughly a year before the announcement, according to John Craske, chief knowledge and innovation officer at CMS UK, speaking to Legal IT Insider (Legal IT Insider). He said adoption and perceived value kept pushing licence numbers up, and that Harvey worked well as a general-purpose legal AI tool, which made an enterprise rollout the sensible way to get lawyers involved with AI. He credited the vendor’s support with helping drive adoption.
Before the expansion, Harvey already had more than 1,100 daily active users and nearly 3,000 monthly users at CMS (CMS). The December 2025 rollout extended access to all 21 CMS member firms in more than 50 countries, which the firm says is the largest law firm rollout of Harvey in Europe, the Middle East and Africa by seats. CMS says the rollout followed extensive collaboration with the vendor on reliability, performance and security.
The firm has not made Harvey its only tool. It says it will keep working with a portfolio of generative AI tools to identify the most effective use cases and measure benefits.
Results
CMS says its internal analysis shows that 93% of users have reported productivity gains equating to up to 117.9 hours saved per lawyer per year (CMS). Two qualifiers matter. The gains are self-reported by users, and 117.9 hours is an upper estimate (“up to”), not an average. CMS has not published the method, the sample or the average saving.
Usage is the firmer signal. More than 1,100 daily active users and nearly 3,000 monthly users before the expansion show that lawyers were choosing to use the tool, not just holding a licence. The firm’s leadership treats the saving as capacity to redeploy: Stephen Millar, managing partner at CMS UK, said the firm’s lawyers are “already seeing meaningful time savings”, which it is reinvesting into higher-value work for clients.
The December 2025 rollout put the tool in the hands of more than 7,000 lawyers and staff. CMS has not disclosed the contract cost, changes in write-offs or realization rates, or any effect on headcount or pricing.
What didn’t go to plan
Public sources report no setbacks at CMS. The open risks are familiar for legal AI. Generative tools can produce confident errors, and in law an invented citation or misread clause becomes a professional-conduct problem, so every output still needs a lawyer’s review. Saved hours in a business that bills by the hour create a commercial tension: CMS says it will offer competitive prices and maintain profitability, but has not said how the time saved will be shared between the firm and its clients. The headline figure is an upper bound from the firm’s own analysis, announced alongside the vendor’s CEO, which makes independent measurement more important as the rollout matures.
Lessons
- Earn the enterprise deal with usage. CMS grew from around 300 licences to about 3,000 lawyers before going firmwide, with more than 1,100 daily users.
- Keep a portfolio while you standardize. The firm has tested more than a dozen tools and says it will keep measuring alternatives.
- Tie the tool to a business pain. CMS links the time saved to fewer write-offs, lighter workloads and competitive pricing.
- Report averages, not only ceilings. “Up to 117.9 hours” is a best case; an average and a method would make the claim usable.
Adapt it in your organization
Is this for you?
- Your professionals spend a large share of time on drafting, review, research and summarizing.
- You write off hours that clients will not pay for, and you track write-offs by matter.
- You can measure active use, not just licences issued.
- Your quality and professional-conduct rules can accommodate AI-assisted work with human sign-off.
- You have a knowledge or innovation team to run trials and support adoption.
A 90-day version
Pilot: give a legal AI assistant to one practice group of about 50 lawyers, with a matched group working as usual, and track hours recorded, hours written off and partner-rated quality on comparable matters. Owner: chief knowledge and innovation officer. Success measure: written-off hours per matter down at least 10% with no fall in quality ratings.
Metric to track
Weekly active users as a share of licences, alongside write-offs per matter. CMS reports more than 1,100 daily users before its expansion and self-reported savings of up to 117.9 hours per lawyer per year. For a case where self-reported time savings did not match observed productivity, see the UK Government case.
Failure modes to pre-empt
- Reporting a best-case “up to” figure to the partnership as the expected saving.
- Buying firmwide licences before usage shows lawyers want the tool.
- Letting AI-drafted work reach a client without a lawyer’s review.
- Saving hours without deciding how the value is shared with clients.
Questions to take to your leadership team
- Which work do we write off most today, and could an AI assistant do a first pass on it?
- How will we share time saved between our margin and our clients’ fees?
- What evidence of usage would we need before committing to firmwide licences?
Executive FAQ
How much time does CMS say Harvey saves its lawyers?
CMS says 93% of users reported productivity gains worth up to 117.9 hours per lawyer per year, a self-reported upper estimate.
How did CMS roll out legal AI?
It started with around 300 licences, grew to about 3,000 lawyers, then extended access to more than 7,000 lawyers and staff in December 2025.
About the curator & AI disclosure
Curated by Luciano Ayres
Luciano Ayres is Director of Artificial Intelligence at ArcTouch, with more than 20 years of experience in software and technology leadership and specialized studies in deep learning. Connect with him on LinkedIn.
AI-generated briefing. This case was produced through an automated AI research and writing workflow curated by Luciano Ayres. Sources are linked for independent verification; summaries and analysis may contain errors and do not replace professional due diligence.