Only 18% track AI ROI, even as agentic AI rolls into professional services

AI use in professional services is no longer the hard part. Proving what it does for the business is.

The Thomson Reuters Institute’s 2026 AI in Professional Services Report pegs organization-wide AI usage at 40% this year, up from 22% in 2025, based on a survey of more than 1,500 respondents across 27 countries. Yet the same report finds only 18% of respondents say their organization is tracking return on investment for AI tools. Adoption has hit critical mass, measurement hasn’t.

Deloitte’s 2026 State of AI in the Enterprise report lands on the same tension from the broader enterprise side. It reports that efficiency and productivity gains are the most common realized benefit so far, with 66% of organizations reporting gains, while revenue lift remains far less common, 20% say they’re already achieving revenue increases and 74% hope to in the future. In other words, AI is in the workflow, but it still struggles to show up cleanly in the P&L.

The adoption gap is now between use and governance

Thomson Reuters Institute measures “agentic AI” adoption in professional services for the first time this year, and reports 15% of organizations have already adopted some type of agentic AI tool. Another 53% say they’re actively planning for agentic AI tools or considering whether to use them. That’s a fast pipeline for systems that can act, not just generate.

Deloitte’s survey points to a similar trajectory, but includes a warning operators should take at face value: governance is falling behind. Deloitte reports that just one in five companies has a mature governance model for autonomous AI agents, even as agentic AI use is expected to increase over the next two years.

For legal ops, tax leaders, and shared-services owners, this is where the work shifts. It’s less about which model is “best” and more about whether the organization can specify, test, and monitor what an agent is allowed to do, which data it can touch, and how it proves it did the right thing.

Client-firm AI rules are becoming a contract clause, not a hallway conversation

One of the clearest operational takeaways from the Thomson Reuters Institute report is that communication about AI remains inconsistent between corporate departments and the outside firms they hire. The report says that more than half of corporate legal departments and corporate tax departments want their outside firms using AI on client matters, yet fewer than one-third of respondents in those departments say they know whether their firms are actually doing it.

From the firm side, the report finds 40% of firm respondents have received conflicting instructions from clients, being told both to use AI and not to use AI on matters. That contradiction isn’t a culture problem. It’s a sourcing and governance gap: rules aren’t being translated into requirements that survive intake, engagement letters, matter management systems, and billing.

A practical implication: outside-counsel guidelines and MSA language are becoming the control plane for AI. If a corporate department wants AI-enabled speed but also wants defensible confidentiality, privilege, and work-product treatment, it needs a written, auditable position on allowed tools, approved environments, logging, and human review, then it needs that position reflected in panel requirements.

The new benchmark is workflow redesign, not tool access

Both reports point to a similar failure mode: AI spreads via individual use while enterprise design lags. Thomson Reuters Institute reports that for the first time a majority of individual professionals are using publicly available tools such as ChatGPT. But it also finds organizations are still thin on broader business metrics, noting that even among those tracking AI, measurement tends to focus on internal operational metrics rather than outcomes like client satisfaction, external revenue generation, or new business won.

Deloitte frames the redesign challenge another way. It reports that only 34% of organizations are “deeply transforming” with AI, while 30% are redesigning key processes and 37% are applying AI at a surface level with little or no process change. Deloitte also notes that organizations see the AI skills gap as the biggest integration barrier, and that education ranked No. 1 in how companies adjusted talent strategies due to AI, rather than role or workflow redesign.

That split matters for operators writing requirements today. For organizations stuck in “surface level” deployments, vendor selection tends to focus on seats, usage, and security checklists. For those redesigning processes, the procurement artifact shifts toward integration patterns, auditability, and outcome metrics. The cost center changes too: licenses move to platforms, and platforms move to data engineering and governance.

There’s also a trust layer, and it’s not optional in professional services. A June 2026 paper in the Journal of Responsible Technology (via ScienceDirect) describes AI in law and accounting as reshaping professional judgment, and argues that responsibility and trust will be defined by how AI’s role is interpreted by professionals, clients, regulators, and technology developers. That’s academic language for an operational reality: buyers will keep asking for proof, not promises, and providers will need repeatable controls to answer.

Where this lands in 2026 vendor reviews and panel renewals

• For legal ops and tax procurement: add an AI-use disclosure field to matter intake and panel RFPs, then decide what “disclosure” means (tool category, environment, data handling, and human review), using the Thomson Reuters Institute finding that fewer than one-third of corporate departments know whether firms use AI as the baseline problem to fix.
• For CIOs and risk leaders: treat agentic AI like a privileged integration, not a productivity app. Deloitte’s “one in five” mature agent governance figure is a sober benchmark for how much policy, monitoring, and testing work is likely missing.
• For professional services COOs: pick two outcome metrics beyond internal efficiency, such as cycle time to draft, first-pass acceptance, client satisfaction, or win rate, and instrument them. The Thomson Reuters Institute’s 18% ROI tracking figure suggests the competitive gap may come from measurement discipline, not model choice.
• For finance and procurement: separate “license ROI” from “workflow ROI.” Deloitte reports 66% see productivity gains but only 20% see revenue gains so far, which indicates many business cases will be justified on capacity and throughput before they are justified on growth.