Global AI investment is accelerating, but KPMG data shows that the gap between enterprise AI spending and measurable business value is rapidly widening.
The headline numbers in KPMG’s first quarterly Global AI Pulse survey are straightforward. Even though global organizations plan to spend a weighted average of $186 million on AI over the next 12 months, only 11 percent have reached the stage of deploying and scaling AI agents in a way that drives enterprise-wide business outcomes.
The central finding, however, is not that AI is failing. 64% of respondents say AI is already delivering meaningful business outcomes. The problem is that “meaningful” in this sentence means doing a lot of heavy lifting, and for most organizations there’s still a huge gap between incremental increases in productivity and compounding operational efficiencies that move the needle in the right direction.
Performance gap architecture
KPMG’s report distinguishes between what it calls “AI leaders” (i.e., those that are scaling or actively operating agent AI) and everyone else. The difference in outcomes between these two cohorts is striking.
Steve Chase, global head of AI and digital innovation at KPMG International, said: “The results of the first Global AI Pulse confirm that increasing spending on AI is not the same as creating value. Leading organizations are moving beyond enablement and deploying AI agents to rethink processes and reshape decision-making and work flows across the enterprise.”
82% of AI leaders report that AI is already delivering meaningful business value. Among peers, this number drops to 62%. This 20 percentage point spread may seem modest when viewed in isolation, but it quickly becomes complicated when you consider what it reflects. It’s not just an improved tool, it’s a fundamentally different implementation philosophy.
Of these, 11% of organizations have deployed agents that coordinate work across functions, guide decisions without human intervention every step of the way, uncover enterprise-wide insights from operational data in near real-time, and report anomalies before they become incidents.
In IT and engineering, 75 percent of AI leaders use agents to accelerate code development, compared to 64 percent of other leaders. In operations, where supply chain orchestration is the primary use case, the percentages are 64 percent and 55 percent. These are not small differences in tool adoption rates. These reflect different levels of process restructuring.
Most companies implementing AI do so by layering models onto existing workflows (e.g., co-pilot here, summarization tool there, etc.) without redesigning the processes the tools reside within. That gives you incremental benefits.
Organizations that are closing performance gaps are reversing this approach. We are first redesigning the process and then introducing agents to work within the redesigned structure. The difference in return on AI spending for these two approaches over a three- to five-year period could be the determining competitive variable in some industries.
What $186 million actually buys you, and what it doesn’t.
The investment numbers in KPMG Data are worth scrutiny. While a global weighted average of $186 million per organization may seem like a significant amount, the regional differences tell a more interesting story.
ASPAC leads with $245 million, followed by the Americas with $178 million and EMEA with $157 million. Within ASPAC, organizations, including those in China and Hong Kong, invest an average of $235 million. Within the Americas, the size of the US organization is $207 million.
These numbers represent planned spending across model licenses, computing infrastructure, professional services, integration, and the governance and risk management apparatus needed to operate AI responsibly at scale.
The question is not whether $186 million is too much or too little. It is the percentage of that number that is allocated to the operational infrastructure required to extract value from the model itself. Survey data suggests that most organizations continue to underestimate this latter category.
Compute and licensing costs are visible and relatively easy to budget for. Frictional costs, such as engineering time spent integrating AI output with traditional ERP systems, delays introduced by search-enhanced production pipelines built on poorly structured data, and compliance overhead for maintaining audit trails for AI-assisted decision-making in regulated industries, tend to surface late in the implementation cycle and often exceed initial estimates.
Vector database integration is a useful example. Many agent workflows rely on the ability to retrieve relevant context from large unstructured document repositories in real time. Building and maintaining the infrastructure for this (choosing a provider like Pinecone, Weaviate, Qdrant, etc., embedding and indexing proprietary data, and managing update cycles as the underlying data changes) adds significant engineering complexity and ongoing operational costs that are largely absent from early AI investment proposals.
If that infrastructure is absent or poorly maintained, the model’s behavior is correct for the context it receives, but that context is outdated or incomplete, resulting in poor agent performance and often difficult to diagnose.
Governance as an operational variable rather than a compliance exercise
Perhaps the most practical finding of the KPMG study is the relationship between AI maturity and risk confidence.
Only 20% of organizations still in the experimental stage are confident in their ability to manage AI-related risks. Among AI leaders, this number rises to 49%. 75% of global leaders, regardless of maturity, cite data security, privacy, and risk as ongoing concerns, but maturity changes how these concerns are operationalized.
This is an important distinction for boards and risk functions, which tend to frame AI governance as a constraint to deployment. KPMG’s data suggests the opposite dynamic. In short, governance frameworks should not slow down AI adoption in mature organizations. they make it possible. The confidence to act more quickly, including introducing agents into more critical workflows and expanding agent collaboration across functions, is directly correlated to the maturity of the governance infrastructure surrounding those agents.
In practice, this means that organizations that treat governance as a retrospective compliance layer are doubly disadvantaged. Each new use case triggers a new governance review, slowing implementation. Additionally, the lack of built-in governance mechanisms means edge cases and failure modes are discovered in production rather than testing, making them more exposed to operational risk.
Organizations that build governance into the deployment pipeline itself (model cards, automated output monitoring, explainable tools, human escalation paths for unreliable decisions, etc.) operate with the confidence that allows them to scale.
“Ultimately, there is no future for agents without trust, and there is no trust without consistent governance,” explains Steve Chase, Global Head of AI and Digital Innovation at KPMG International. “This research shows that with continued investment in talent, training, and change management, organizations can responsibly scale AI and capture value.”
Regional differences and their signals for global development
For multinational companies managing AI programs in multiple regions, KPMG data shows significant differences in adoption speed and organizational structure that impact global deployment plans.
ASPAC has been the most active in agent expansion. 49% of organizations are scaling AI agents, compared to 46% in the Americas and 42% in EMEA. ASPAC also leads in more complex capabilities coordinating multi-agent systems with 33%.
Barrier profiles also differ in that they have real operational implications. In both ASPAC and EMEA, 24% of organizations cite a lack of trust and buy-in from leaders as a key barrier to adopting AI agents. In the Americas, this number drops to 17%.
Agent systems, by definition, make or initiate decisions without human approval on a case-by-instance basis. In organizational cultures where decision-making responsibility is concentrated at senior levels, this can create organizational resistance that no amount of technical ability can resolve. The solution is governance design. Specifically, we predefine the categories of decisions that agents are authorized to make autonomously, the triggers for escalation, and who is responsible for the outcomes they initiate.
The gap in expectations regarding human-AI collaboration is also noteworthy for those designing agent-assisted workflows on a global scale.
Respondents in East Asia expect 42% of projects to be led by AI agents. 34% of Australian respondents preferred human-driven AI. 31% of North American respondents lean toward peer-to-peer human-AI collaboration. These differences impact how agent-assisted processes need to be designed in different regional deployments of the same underlying system, adding localization complexity that is often underestimated in centralized platform planning.
One data point from the KPMG survey that CFOs and boards should take note of is that 74 percent of respondents say AI will remain a top investment priority even in the event of a recession. This is either a sign of genuine conviction about the role of AI in cost structures and competitiveness, or it reflects a collective effort that has yet to be tested against real budgetary pressures. Both are likely present in different proportions in different organizations.
This shows that the grace period for organizations still in the experimental phase is not indefinite. If the 11 percent of AI leaders continue to extend their advantage (and KPMG data suggests there are mechanisms in place to do so), the question for the remaining 89 percent is not whether to accelerate AI adoption, but how to do so without exacerbating the integrated debt and governance deficiencies that are already constraining returns.
SEE ALSO: Hershey applies AI across supply chain operations
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expos in Amsterdam, California, and London. This comprehensive event is part of TechEx and co-located with other major technology events such as Cyber Security & Cloud Expo. Click here for more information.
AI News is brought to you by TechForge Media. Learn about other upcoming enterprise technology events and webinars.
