Poor governance, not readiness, is biggest barrier to AI ROI, survey finds

Poor governance is emerging as the bigger drag on AI returns than workforce readiness or data readiness, and that should change how KPMG teams sell, build, and audit AI work. The survey result cuts against the familiar story that adoption stalls mainly because employees need more training or companies need better data. It suggests a more basic failure: organizations can buy the tools and hire the talent, yet still struggle to turn AI into value when accountability, policy, monitoring, and decision rights are unclear.

For consultants and auditors, that means AI work is becoming a governance exercise as much as a technical one. A client can have a model in production and still be unable to answer who owns it, how it is reviewed, or what happens when it changes. That gap slows adoption inside the business and leaves workers guessing about when AI is allowed, who approves its use, and whether output can be relied on in client-facing or control-sensitive work.

The practical implications reach well beyond experimentation. If governance is the binding constraint, then client engagements need to cover model oversight, control design, escalation paths, risk acceptance, change management, and outcome measurement. Those are not abstract policy topics. They shape how employees use AI day to day, how managers sign off on outputs, and how much confidence a firm can place in the results during busy season, transformation projects, or audit delivery.

That is also why AI keeps getting discussed alongside ethics and compliance. The organizations most likely to capture real returns are the ones that can show clear ownership and review processes, not just broad enthusiasm for innovation. In a Big Four environment, that plays directly to work already centered on audit quality, internal controls, operating model design, and risk advisory.

The shift is a reminder for KPMG staff that the most valuable AI professionals may not be the ones who can simply build the tool. They may be the ones who can govern it, document it, and make it usable without creating new control gaps. In a market full of talk about readiness, the harder issue is whether anyone has actually designed the rules that let AI scale.