Thomson Reuters just released its fourth annual AI in Professional Services Report. The headline number: 40% of organizations now use generative AI, nearly double the 22% reported last year. That is not the interesting finding. The interesting finding is buried a few pages deeper. Only 18% of those organizations track return on investment. Another 40% say they do not know whether ROI is measured at all. In plain terms, the industry doubled its AI adoption rate and barely moved on the question that actually matters: is any of this working?
The measurement gap is the real adoption gap
The conversation about AI adoption usually splits companies into two camps. Those using it and those not using it. That split made sense in 2024 when adoption was at 22% and the question was whether AI would stick. It does not make sense anymore.
The meaningful divide in 2026 is between companies that adopted AI and can prove it delivers value, and companies that adopted AI and have no idea. That second group is four times larger than the first.
This is not a minor gap. It changes everything downstream. If you cannot demonstrate what AI is doing for the business, you cannot justify expanding it. You cannot make smart decisions about where to invest more and where to pull back. You cannot answer the CFO when she asks what the company got for its money. You end up defending AI budgets with anecdotes instead of evidence.
The Thomson Reuters data puts a number on something that has been visible for months: organizations are spending on AI the way they spend on conference sponsorships. They believe it is probably good for them, they have no proof, and nobody is asking hard questions about it yet.
What the 18% are actually measuring
Even among the organizations that do track results, the picture is thin. Thomson Reuters found that 77% of the organizations measuring AI ROI rely exclusively on internal metrics. Cost savings. Employee usage rates. Time spent per task.
These are activity metrics, not outcome metrics. Knowing that 200 employees use an AI tool every week tells you nothing about whether those employees are producing better work, serving clients more effectively, or generating more revenue. Usage is not value. Cost reduction is not growth.
Almost none of the organizations in the survey measure client satisfaction impact or revenue attribution from AI investments. The measurements that would actually tell you whether AI is delivering business value are the ones almost nobody is collecting.
This is the operations problem hiding inside the technology story. The organizations that deployed AI treated it as a technology decision. Buy the tool, roll it out, count the logins. The organizations that will get actual returns from AI will treat it as an operations decision. Define what “working” looks like before the deployment starts. Set the criteria, build the feedback loops, and measure against them from day one.
What happens to the companies that never measure
The 82% without measurement are not standing still. They are building on a foundation they cannot evaluate. Every quarter without data is a quarter where bad investments get renewed, good ones get overlooked, and the AI budget becomes a political line item instead of a strategic one.
Here is the compounding problem. The 18% measuring outcomes will iterate. They will see what works, cut what does not, and redirect resources toward the highest-return applications. Over 12 months, that feedback loop produces dramatically different AI portfolios than “we rolled it out and hope it is helping.”
The Thomson Reuters report found that 15% of organizations have already adopted agentic AI tools, with another 53% planning or considering them. Agentic AI is more expensive, more operationally complex, and harder to evaluate than basic generative AI. If organizations cannot measure the impact of a chatbot answering employee questions, they are not ready to measure autonomous agents making decisions on their behalf.
The gap between the measured and the unmeasured will accelerate. It always does. Operations without feedback loops do not improve. They just continue.
The question this week
If your organization is using AI, ask a simple question in your next leadership meeting: what is our evidence that AI is producing business value? Not usage numbers. Not time savings estimates. Actual business outcomes tied to AI investments.
If nobody in the room can answer, that is your real AI problem. Not which model to use. Not which vendor to pick. Whether anyone in the building knows if the thing you already bought is doing what you bought it to do.
Thomson Reuters surveyed 1,500 professionals across legal, tax, accounting, risk, and government. The numbers are specific to professional services, but the measurement gap extends far beyond one sector. McKinsey’s own 2025 research found similar patterns across industries: adoption rates climbing, measurement capabilities lagging behind. The gap is structural, not sectoral.
Adoption was the easy part. Measurement is where the actual work begins.