Only 11 percent of global organisations have moved AI agents beyond the pilot phase to deliver enterprise-wide results, despite planning to spend an average of $186 million each on AI over the next year.
Companies are pouring unprecedented capital into artificial intelligence, but the vast majority have little to show for it beyond incremental productivity gains and isolated use cases. KPMG's first quarterly Global AI Pulse survey pulls no punches: global organisations plan to spend a weighted average of $186 million on AI over the next 12 months, yet only 11 percent have successfully deployed and scaled AI agents in ways that generate measurable, enterprise-wide business outcomes. That disconnect is not a technology problem. It is an architecture and strategy problem.
Steve Chase, Global Head of AI and Digital Innovation at KPMG International, puts it plainly. He notes that spending more on AI is not the same as creating value, and that leading organisations are moving beyond basic enablement to deploy agents that fundamentally reimagine processes and reshape how decisions flow across the enterprise.
The KPMG data draws a sharp line between what it calls AI leaders, organisations actively scaling agentic AI, and everyone else. Among leaders, 82 percent report AI is already delivering meaningful business value. Among their peers, that figure falls to 62 percent. That 20-percentage-point gap might seem manageable at first glance, but it widens fast when you look at what it actually represents. These are not minor differences in tool adoption or licence counts. They reflect fundamentally different philosophies about how AI should interact with corporate infrastructure.
Companies in that leading 11 percent are not simply adding a copilot to an existing workflow or bolting a summarisation tool onto a legacy process. They are deploying agents that coordinate work across multiple functions, route decisions without human intermediation at every step, surface operational insights in near real-time, and flag anomalies before they become costly incidents. In IT and engineering, 75 percent of AI leaders use agents to accelerate code development compared to 64 percent of their peers. In operations, where supply chain orchestration dominates, the split is 64 percent versus 55 percent. These gaps compound over years into a structural competitive advantage that late movers will struggle to reverse.
This mirrors patterns seen in earlier enterprise technology shifts, from cloud migration to data analytics adoption. The companies that treated those investments as infrastructure redesign rather than vendor procurement pulled ahead decisively, and AI appears to be following the same trajectory, just at a faster pace.
Where the money actually goes
The regional investment figures deserve attention. ASPAC leads at $245 million per organisation on average, with China and Hong Kong averaging $235 million. The Americas come in at $178 million, with US organisations at $207 million, while EMEA trails at $157 million. These numbers cover the full stack: model licensing, compute infrastructure, professional services, integration, and the governance frameworks needed to operate AI responsibly at scale.
The critical question is not whether $186 million is the right number. It is what proportion of that budget is being spent on the operational infrastructure required to actually extract value from the models. Too many organisations are allocating the bulk of their AI budgets to model access and compute power while underinvesting in the process redesign, change management, and integration work that turns a capable model into a business advantage. Generative AI models from providers like OpenAI, Anthropic, and Google have become commoditised quickly. What differentiates companies now is how effectively they embed those models into redesigned workflows and give agents the autonomy to execute across siloed systems.
As the Financial Times recently noted in its coverage of corporate AI spending patterns, the gap between experimentation and measurable return on investment remains the central challenge for chief information officers navigating board-level expectations around AI-driven efficiency gains.
What separates leaders from the rest
The organisations gaining ground share a common trait: they redesign the process first, then deploy agents to operate within the new structure. Most enterprises do the opposite, layering AI onto legacy workflows and hoping for transformational results. That approach generates incremental gains, perhaps faster document drafting or marginally quicker data retrieval, but it does not produce the kind of compounding operational efficiency that genuinely improves margins over a three to five year horizon.
This distinction between process re-architecture and tool layering is likely to become the defining competitive variable across several industries in the coming years. Companies that treat AI as an infrastructure transformation rather than a software purchase will outpace those that simply distribute licences and wait for results. The data is already showing that gap, and it is growing.
