Enterprise interest in AI agents has moved far beyond experimentation, but widespread deployment remains rare. KPMG’s Global AI Pulse finds that organisations plan to spend a weighted average of $186 million on AI over the next 12 months, while only 11% are scaling agents across the enterprise.
That contrast explains why spending alone is a weak measure of progress. KPMG reports meaningful business outcomes at 64% of organisations overall, compared with 82% among AI leaders. The difference is not simply access to better models. Leading companies are redesigning work, funding integration and treating governance as an enabler of scale.
Agents can automate a task, but margin gains emerge when companies reconsider the process around it. Deploying an assistant on top of a slow or fragmented workflow may reduce a few minutes of labour without changing the economics of the operation.
Process redesign separates leaders from tool buyers
KPMG’s leaders tend to redesign processes before introducing agents. Less mature organisations are more likely to layer new tools over existing structures, preserving handoffs and bottlenecks that automation could have removed.
The pattern appears in specific use cases. Leaders report coding agents at a rate of 75%, compared with 64% among peers. In supply-chain orchestration, the figures are 64% and 55%. These are not isolated chat interfaces; they are systems working inside activities that affect product delivery, operating capacity and cost.
This distinction matters for return on investment. A company can buy an agent quickly, but connecting it to trusted data, permissions and business applications takes engineering and organisational work. Hidden integration and governance costs can erode the expected saving if they are not included in the business case.
The practical starting point is therefore a measurable constraint: a delayed approval, repetitive reconciliation or overloaded service queue. Teams can then map the entire process, decide what the agent may do and establish where a person must remain accountable.
Regional ambition is high but takes different forms
Planned investment varies considerably. KPMG puts the weighted average at $245 million in Asia-Pacific, $178 million in the Americas and $157 million in Europe, the Middle East and Africa. China and Hong Kong average $235 million, while the United States reaches $207 million.
Scaling patterns are closer but still distinct: 49% in Asia-Pacific, 46% in the Americas and 42% in EMEA report scaling agents. Asia-Pacific also leads in multi-agent adoption at 33%. Regional preferences for collaboration and deployment differ, which means a global company cannot assume one operating model will fit every market.
The numbers suggest that agent adoption is becoming a competitive infrastructure decision rather than a collection of local pilots. Yet rapid regional expansion can magnify inconsistent security, data and approval standards. An agent working well in one unit may produce different outcomes when connected to another market’s records and rules.
Common controls are essential, but implementation must reflect local regulation, language and work practices. Central architecture and regional operating knowledge need to reinforce each other.
Governance creates confidence to move faster
Data security, privacy and risk concern 75% of surveyed organisations. Confidence in managing those issues rises from only 20% among experimenters to 49% among leaders. The finding challenges the idea that governance mainly slows innovation.
Clear permissions, monitoring and escalation make it safer to give an agent more responsibility. They also make performance easier to compare. Without those controls, a successful demonstration can stall because nobody is confident enough to connect it to consequential systems.
KPMG also finds that 74% of executives would keep AI among their highest priorities during a recession. That resilience raises the standard for investment: programmes must show durable operating value, not novelty.
The strongest agent playbook is consequently both technical and managerial. Redesign the workflow, account for integration, define ownership and measure the business result. Enterprise leaders are not winning because they release the largest number of agents. They are building the conditions in which a smaller number can safely change how work is done—and turn automation into margin rather than another software expense.
