AI agents are spreading fast, but most firms still can’t price the gain

In three corners of the economy that sell expertise by the hour, a pattern is getting hard to ignore: AI adoption is racing ahead, budgets are holding, and the hardest part is no longer getting tools into people’s hands. It’s proving what they’re worth, governing how they’re used, and writing commercial terms that survive once the productivity gain becomes the baseline.

The most uncomfortable number in the stack is not an adoption stat. It’s a measurement stat. Tommaso Maria Ricci, citing Thomson Reuters’ 2026 AI in Professional Services Report, wrote that only 18% of organizations track return on investment for AI tools, even as firm-wide AI use jumped to 40% in 2026 from 22% in 2025.

Professional services bought AI, then forgot the dashboard

Ricci’s August 2026 guide, citing Thomson Reuters’ professional services survey data, highlights how adoption by individuals is outpacing companywide deployment. In Thomson Reuters’ Future of Professionals 2026 report, 74% of professionals report using AI several times per week, and 44% say they use it multiple times per day. By contrast, only 40% of organizations report AI in use across the firm.

That spread, heavy usage at the desk and lighter adoption in the enterprise, is where governance and economics go to die. If tools are used informally, firms get the risk and the spend, but not the proof. And without proof, pricing conversations default to client expectations rather than internal facts about cycle time, quality, and rework.

Ricci also pointed to the policy clock operators already feel: high-risk obligations under the EU AI Act became enforceable on Aug. 2, 2026, per the European Commission framework he referenced. For firms serving EU clients or handling EU data, the compliance work is now part of the rollout plan, not a later clean-up.

Recruiting shows what happens when efficiency turns into noise

Hiring offers a useful preview because recruiting has already gone through several waves of automation, yet the cost math has barely changed. In an April 2026 column, Staffing Industry Analysts’ Jacob Claerhout wrote that vendors marketing AI agents cite 30% to 40% lower cost per hire and 340% ROI within 18 months. His counterpoint: SHRM’s 2025 benchmarking still puts the average non-executive cost per hire at $4,700, roughly where it stood before the pandemic.

Claerhout’s most operational point is about constraints. When the bottleneck is recruiter capacity, automation can compress time to fill and increase throughput. When the bottleneck is candidate scarcity, more outreach volume does not change the market clearing price for talent.

He also flagged an early sign that results are normalizing: in a single year, cold response rates fell 27% as AI-powered messaging pushed more outreach into inboxes, but overall response volume did not increase at the same pace. The productivity lift was higher send counts, not more conversations.

Indeed’s Recruiting Trends study, cited by Staffing Industry Analysts, put numbers on the early upside: among early automation adopters, recruiters filled 64% more roles per recruiter and submitted 33% more candidates. Claerhout said the context has changed, with 61% of staffing firms using AI for recruiting operations, up from 48% a year earlier. What used to separate teams is now table stakes.

Finance budgets are steady, and agents are moving into production

Financial services is showing the same directional shift, but with larger budgets and stricter controls. WealthManagement’s Davis Janowski reported in January 2026 that NVIDIA published its sixth annual State of AI in Financial Services report, based on a survey of more than 800 industry professionals worldwide.

Two figures stand out for CIOs and risk teams. Janowski wrote that 65% of respondents said their company is actively using AI, compared with 45% the prior year. On spending, respondents were effectively unanimous that AI budgets will not shrink, instead forecasting either flat funding or an increase over the next year.

Agentic AI is also where costs can be hardest to spot. WealthManagement reported that 42% of respondents are using or evaluating agentic AI, and 21% said they have already deployed AI agents. It also reported that 84% consider open-source models and software important to their AI strategy, a signal that procurement discussions often move from “which vendor?” to “which stack, which controls, and who runs it?”

NVIDIA’s survey breakdown of challenges, as summarized by WealthManagement, reads like an implementation checklist. The top two reported issues were performance reliability (34%) and lack of internal skills or experts to manage or monitor AI agents (33%), followed by data issues such as privacy, sovereignty and disparate locations (30%). Implementation and integration challenges were cited by 28%, and regulatory and ethical concerns by 28%.

The connective tissue across staffing, professional services, and finance is that “AI spend” is no longer mostly a license question. It is becoming a monitoring, workflow integration, and commercial-terms question. In practice, the cost center shifts from seat counts to operating controls: evaluation harnesses, human-in-the-loop review design, audit trails, and the internal team that watches for drift and failure modes.

Contracting and measurement are becoming the real rollout work

For operators, the throughline is simple. Adoption can be fast and still be unmanaged. Pricing can stay the same and still shift who captures value. And the more agentic the system, the more the budget migrates into ongoing operations.

• Define what “ROI” means before the renewal. If only 18% of organizations track AI ROI in professional services, per Thomson Reuters data cited by Ricci, set a minimum measurement standard for each tool: cycle-time delta, rework rate, and exception volume are often easier to instrument than “hours saved.”
• Separate “capacity gain” from “market gain” in recruiting. Staffing teams evaluating AI agents can test Claerhout’s constraint logic by running pilots where candidate supply is healthy versus scarce, then measuring whether time to fill moves without increasing paid outreach volume.
• Budget for agent monitoring as a first-class line item. NVIDIA’s financial services survey, as reported by WealthManagement, flags reliability (34%) and skills gaps (33%) as top challenges. That should translate into a named owner for agent performance, explicit escalation paths, and procurement language that covers drift, logging, and incident response.