Professional services firms deployed AI across 40% of their organizations in 2025, but only 18% bothered to track whether it’s actually working, according to Thomson Reuters’ 2026 AI in Professional Services Report. That’s not a measurement gap — it’s a financial accountability crisis hiding in plain sight.
The math doesn’t just fail to add up. It reveals something worse: enterprises are scaling AI faster than their ability to understand it, and the gap is widening as autonomous agents enter the mix.
The 82% who can’t prove their AI investments work
Here’s the uncomfortable truth about professional services in 2026. These are regulated, audited entities — law firms, consultancies, accounting practices — that wouldn’t buy office furniture without a three-bid process. Yet they’re deploying AI with less rigor than they’d apply to a coffee machine lease.
Even as AI adoption accelerates across high-skill professions, the infrastructure to measure its impact is lagging dangerously behind. 41% of executives now cite ROI measurement as their top priority for 2026 — which probably means it was their top priority in 2025 too, and they still haven’t fixed it.
The problem isn’t awareness. It’s action.
Why most AI deployments fail to deliver
Only 5% of enterprises achieve substantial AI ROI, according to research compiled in the Thomson Reuters analysis. Another 35% report partial returns. The rest? The data doesn’t say, which is itself revealing.
And the failure modes are predictable. 69% of tech executives reportedly made security and scalability trade-offs just to move faster. The 18% tracking rate doesn’t account for shadow AI deployments, where employees use tools IT doesn’t even know exist.
The security shortcuts are compounding as AI agents find vulnerabilities faster than humans can patch them — a problem that’s invisible without proper ROI tracking. But here’s what’s worse: firms know they’re flying blind. They just can’t stop long enough to build instruments.
Agentic AI is making everything unmeasurable
15% of professional services firms have already adopted agentic AI systems — autonomous agents that make decisions without human approval. Another 53% are planning or considering it. That adoption curve is faster than GenAI’s early trajectory, and it’s happening while most firms still can’t measure their basic chatbot deployments.
GenAI outputs are somewhat trackable. You can count documents generated, emails drafted, code written. But autonomous agents choosing their own tasks? 62% are experimenting with this technology, 23% are scaling it, and nobody in the Thomson Reuters research can name a single firm that’s solved the measurement problem.
No case studies. No success stories. Just aggregate data showing a systemic failure accelerating in real time.
The collision nobody’s prepared for
The honest limitation here isn’t technological — it’s organizational. Professional services firms are deploying autonomous AI agents at 15% adoption while 82% still can’t measure if their basic GenAI tools are working. Traditional ROI frameworks don’t apply when the AI is writing its own task list. And the firms that should know better — the ones auditing everyone else’s books — are the ones deploying tools they fundamentally can’t evaluate.
One of those numbers has to break.
