The Legal AI Return on Investment Reality Check 2026: What Law Firms and Legal Departments Are Actually Measuring — and How It Compares to What They Projected at Deployment

Research briefing — a review of public sources. The Legal Stack did not conduct a survey, benchmark or interviews for this piece.

Updated October 2026. An earlier version described "projected vs realized" ROI gaps without sources. This version relies only on published surveys and studies, and labels our analysis as analysis.

What the public record can and cannot tell you

No independent study has tracked a cohort of legal AI deployments and compared the ROI projected at purchase with the ROI realized 18 to 36 months later. That is the question in this briefing's title, and the honest answer is that the data to settle it is not public.

The public record does contain three useful things:

forecasts of the value of AI time savings
self-reported data on whether organizations see returns and whether they measure them
independent evidence on the costs that tend to erode projected value

Taken together, they show a clear pattern. Expectations are high, measurement is rare, and the people paying for legal work are not yet seeing the savings.

What is being projected

Thomson Reuters' Future of Professionals report, announced on 26 June 2025, surveyed 2,275 professionals in February and March 2025. Respondents predicted that professionals using AI would save about 5 hours per week within the next year, up from 4 hours predicted in 2024. Thomson Reuters valued this at an average of $19,000 per professional per year. It estimated a combined annual impact of $32 billion across U.S. legal and CPA professionals, $20 billion of it in legal.

These numbers are predictions by respondents, not measured outcomes. They are the kind of number that ends up in internal business cases. A firm that uses them should label them as forecasts and test them against its own baseline.

The same report found that:

Only 22% of respondents said their organization had a visible, defined AI strategy.
Organizations with such a strategy were twice as likely as those with informal or ad hoc adoption to report revenue growth from AI.
53% of professionals said their organization already sees ROI from AI, directly or indirectly.

What is being measured

Measurement is the weak link. Thomson Reuters' 2025 Generative AI in Professional Services report was published on 15 April 2025. Only 20% of respondents said they knew their organization was measuring the ROI of generative AI. Organization-level active use of generative AI rose to 22% in 2025, up from 12% in 2024. Those figures cover all professional-services respondents, not legal alone.

Clio's 2025 Legal Trends Report, summarized by the Institute for the Advancement of the American Legal System (IAALS), shows a similar picture for law firms:

79% of legal professionals use AI.
36% say AI has positively affected revenue, rising to 69% among firms that have adopted it widely.
About a third say it is too early to tell.
Among wide adopters, 45% have changed their pricing and 20% report difficulty meeting billable targets.

That last figure is a direct sign of the hourly-billing problem. For a firm billing by the hour, time saved is revenue lost unless pricing changes.

What clients are seeing

From the client side, the evidence is more sobering. In October 2025, the Association of Corporate Counsel (ACC) and Everlaw surveyed 657 in-house professionals in 30 countries and found that:

nearly 60% had seen no noticeable savings yet from outside counsel's use of generative AI
58% said firms had not adjusted pricing to reflect AI efficiencies
61% said they were likely to push for different pricing, such as value-based fees or alternative fee arrangements (AFAs)

The same report found that 52% of in-house teams now actively use generative AI themselves, up from 23% in 2024.

Wolters Kluwer's 2024 Future Ready Lawyer survey, which covered 712 professionals in 10 countries, points the same way. Sixty percent of respondents expected AI to reduce the dominance of the billable hour.

The cost side: verification and fee rules

Two documented factors reduce the value that projections assume.

Verification overhead. Stanford RegLab and Stanford HAI found in May 2024 that Lexis+ AI and Ask Practical Law AI hallucinated on more than 17% of benchmark queries, and Westlaw AI-Assisted Research on more than 34%. Accuracy has improved since then. Vals AI's October 2025 legal research benchmark reported AI tools scoring 79–81% against a lawyer baseline of 71%. But accuracy below 100% means a lawyer still has to review the output, and that review time belongs in the cost side of any ROI model.

Ethics rules on fees. ABA Formal Opinion 512, issued 29 July 2024, says lawyers billing hourly may bill only for time actually spent. Lawyers on flat fees should take AI efficiencies into account. Lawyers generally may not bill clients for time spent learning a tool they will use regularly. A tool that supports general practice is overhead, not a chargeable expense. In practice, the opinion moves much of the efficiency gain to the client unless the pricing model changes.

Analysis: why projected and realized ROI diverge

Based on the sources above, our analysis is that the gap between projected and realized ROI comes mainly from three causes. This is analysis, not measured data.

The baseline is missing. If only about one in five professionals know ROI is being measured, most "realized ROI" figures are reconstructed after the fact. Time saved is easy to claim and hard to verify without pre-deployment cycle-time and hours data.
The value lands in the wrong place. Under hourly billing and Opinion 512, efficiency gains go mostly to the client. The ACC/Everlaw and Clio data suggest pricing has not yet caught up. Clients see no savings and firms see pressure on billable targets, so both sides can be disappointed at once.
Review costs are left out. Business cases often count drafting or research time saved and leave out the senior review the hallucination data says is still needed.

A practical measurement framework

The following framework is our analysis. We suggest it for legal ops teams and firm management:

Baseline first. For each target task (for example, NDA review or first-draft research memos), record volume, elapsed cycle time and total attorney hours, including review hours, for at least one quarter before go-live.
Measure net time, not gross time. Count the time for the AI-assisted first pass plus verification, and compare that with the baseline.
Track who captures the value. Firms should record whether savings went to lower client bills, write-down reduction, capacity for new work or margin on fixed fees. Legal departments should track outside counsel spend on the targeted matter types.
Count the full cost. Include licences, integration, training time (which, under Opinion 512, generally cannot be billed to clients), and governance overhead.
Treat vendor and survey forecasts as hypotheses. Use figures like Thomson Reuters' 5-hours-per-week forecast as a hypothesis to test against your own baseline, not as the expected benefit.
Tie pricing to measured efficiency. Clio found that 45% of wide adopters had changed their pricing. Fixed-fee or subscription arrangements built on measured efficiency are the most direct way for firms to turn time saved into retained value.

Bottom line

The public evidence does not show that legal AI fails to deliver ROI. What it shows is that most organizations cannot yet show whether it does:

Forecasts are confident.
Measurement is rare.
Clients largely report no savings yet.
Ethics rules on fees push efficiency gains toward clients unless pricing changes.

Firms and legal departments that set a baseline and measure net time will be the first to produce credible ROI numbers, whichever way those numbers turn out.

Sources

Thomson Reuters, "The AI adoption reality check: firms with AI strategies are twice as likely to see AI-driven revenue growth" (26 Jun 2025): https://www.thomsonreuters.com/en/press-releases/2025/june/the-ai-adoption-reality-check-firms-with-ai-strategies-are-twice-as-likely-to-see-ai-driven-revenue-growth-those-without-risk-falling-behind
Thomson Reuters, "2025 Generative AI in Professional Services report" (15 Apr 2025): https://www.thomsonreuters.com/en-us/posts/technology/genai-professional-services-report-2025/
2Civility (Illinois Supreme Court Commission on Professionalism), "2025 Clio Legal Trends Report" (29 Oct 2025): https://www.2civility.org/2025-clio-legal-trends-report/
Association of Corporate Counsel, "ACC GenAI report: corporate law departments' AI use (with Everlaw)" (14 Oct 2025): https://www.acc.com/about/newsroom/news/acc-genai-report-corporate-law-departments-ai-use-everlaw
LawSites (Robert Ambrogi), "Latest Future Ready Lawyer survey from Wolters Kluwer finds broad adoption of gen AI by legal professionals" (24 Oct 2024): https://www.lawnext.com/2024/10/latest-future-ready-lawyer-survey-from-wolters-kluwer-finds-broad-adoption-of-gen-ai-by-legal-professionals.html
Stanford HAI, "AI on Trial: Legal Models Hallucinate in 1 out of 6 (or More) Benchmarking Queries" (23 May 2024, updated 30 May 2024): https://hai.stanford.edu/news/ai-trial-legal-models-hallucinate-1-out-6-or-more-benchmarking-queries
LawSites, "Vals AI's latest benchmark finds legal and general AI now outperform lawyers in legal research accuracy" (23 Oct 2025): https://www.lawnext.com/2025/10/vals-ais-latest-benchmark-finds-legal-and-general-ai-now-outperform-lawyers-in-legal-research-accuracy.html
Freeman Mathis & Gary, "ABA issues formal guidance for lawyers' use of generative AI" (2024): https://www.fmglaw.com/professional-liability/aba-issues-formal-guidance-for-lawyers-use-of-generative-ai/