Large companies review AI costs more widely than they write role rules for staff, KPMG finds

Rahsaan Shears, a Georgia-based AI enterprise transformation leader at KPMG LLP ^[16], counts the guidance figure as progress. "The role-specific guidelines, this 21 to 55%, I think that speaks to maturity because you have to understand how it's integrated into the flow of work," she said ^[5]. On the year-on-year figures, guidance has kept pace. Workforce adoption rose 34 points over the year ^[17]. Role-specific guidelines for working with AI agents rose by the same 34 points, from 21% ^[3][18].

KPMG counts firms "developing" those guidelines ^[3], and the survey does not say how many have finished them. Controls on money are further along. Cost reviews in AI approval processes rose 13 points in a single quarter, from 61% ^[2][21], and now lead role guidance by 19 points ^[19]. Seventy percent of firms use AI monitoring dashboards, and 43% set usage or token budgets that cap how much AI processing a team can use ^[6].

Shears gave the reason spending gets watched first. "You could easily create an agent that costs more to run than it is to pay somebody to do the work if you're not careful," she said. "And when you go from pilot to scale, it becomes more and more important." ^[7]

Limits on what agents may decide are thinner. Forty-nine percent of firms have set out the high-risk uses in which agents are barred from making decisions on their own ^[9]. If that share and the share working on agents come from the same 314 respondents ^[1], at least 13% of the firms are building, deploying or developing agents without having drawn that line ^[20].

Gallup surveys employees and KPMG surveys executives ^[11][1]. The two ask different questions, so the distance between them cannot be computed. They point the same way. A Traliant report covered by HRD America in July 2026 found 51% of HR teams reporting clear AI use guidelines and employee training ^[12]. Shears described what employees ask themselves when access arrives before clarity: "There is this potential, like, am I gonna be in trouble? Is it okay?" ^[13] Gallup's research has found employees more likely to use and value AI when their organisations give clear direction ^[15].

The trade-off this quarter is where the next round of governance work goes. One option is more spend control. The other is finished guidance that tells each role what it may hand to an agent. I think the guidance should come first, because the incentive Shears expects to drive adoption depends on staff knowing the rules. "When you hear leaders say, good performance equates to you demonstrating what you can do to get the most from AI, that'll be the carrot," she said. "It's not a stick." ^[14]