AI Agents: Driving Enterprise Margin Gains

Global investments in artificial intelligence are surging, but a widening chasm is emerging between enterprise AI spending and demonstrable business value. New data from KPMG’s inaugural Global AI Pulse survey reveals a stark reality: while organizations worldwide are earmarking an average of $186 million for AI initiatives over the next twelve months, a mere 11% have successfully scaled AI agents to deliver enterprise-wide business outcomes.

This doesn’t signal AI’s failure; a significant 64% of respondents report that AI is already generating meaningful business results. However, the term “meaningful” often encapsulates incremental productivity gains rather than the substantial compounding operational efficiencies that truly move the needle on profit margins. For most companies, the journey from initial deployment to this level of impact remains a considerable undertaking.

The Architecture of a Performance Gap

KPMG’s research meticulously distinguishes between “AI leaders” – those actively deploying and scaling agentic AI – and the broader market. The disparity in achieved outcomes between these two groups is pronounced. Steve Chase, Global Head of AI and Digital Innovation at KPMG International, emphasized, “The initial Global AI Pulse findings underscore that increased AI expenditure doesn’t automatically translate to value creation. Leading organizations are moving beyond mere enablement, leveraging AI agents to fundamentally reimagine processes and transform decision-making and workflow across the enterprise.”

Among AI leaders, an impressive 82% attest to AI delivering significant business value. This figure drops to 62% for their less mature peers. This 20-percentage-point difference, while seemingly modest, compounds rapidly, reflecting not just superior tooling but fundamentally different deployment philosophies. The pioneering 11% are deploying AI agents capable of orchestrating cross-functional work, routing decisions autonomously, surfacing real-time, enterprise-wide insights from operational data, and proactively identifying anomalies before they escalate into critical incidents.

In IT and engineering functions, 75% of AI leaders utilize agents to expedite code development, compared to 64% of their counterparts. In operations, particularly for supply chain orchestration, this split stands at 64% versus 55%. These are not marginal differences in tool adoption; they signify distinct levels of process re-architecture. Many enterprises have adopted a strategy of layering AI models onto existing workflows, resulting in incremental gains. In contrast, organizations effectively closing the performance gap have inverted this approach, redesigning processes first and then deploying AI agents to operate within these optimized structures. The differential in AI return on investment over a three-to-five-year horizon is poised to become a defining competitive advantage across numerous industries.

What $186 Million Actually Buys—And What It Does Not

The reported investment figures warrant closer examination. A weighted global average of $186 million per organization is substantial, but regional variances offer a more nuanced perspective. The Asia-Pacific (ASPAC) region leads with an average planned spend of $245 million, followed by the Americas at $178 million, and EMEA at $157 million. Within ASPAC, organizations in China and Hong Kong are investing heavily at an average of $235 million, while US organizations in the Americas are investing $207 million. These figures encompass model licensing, compute infrastructure, professional services, integration, and the crucial governance and risk management frameworks essential for responsible AI deployment at scale.

The critical question is not the absolute amount invested, but rather the proportion allocated to the operational infrastructure required to extract value from AI models. Survey data suggests that this latter category is consistently being underweighted. Compute and licensing costs are more tangible and easier to budget for. However, the often-underestimated “friction costs” – the engineering hours dedicated to integrating AI outputs with legacy systems, latency introduced by retrieval-augmented generation pipelines operating on poorly structured data, and the compliance overhead for auditing AI-assisted decisions in regulated sectors – tend to surface late in the deployment cycle and frequently exceed initial projections.

Vector database integration serves as a pertinent example. Many agentic workflows depend on the ability to retrieve relevant context from vast, unstructured document repositories in real-time. Building and maintaining the necessary infrastructure—including selecting from providers like Pinecone, Weaviate, or Qdrant, embedding and indexing proprietary data, and managing refresh cycles as underlying data evolves—introduces significant engineering complexity and ongoing operational costs that are often absent from initial AI investment proposals. When this infrastructure is inadequate or poorly maintained, agent performance degrades subtly, making diagnosis challenging as the model may be functioning correctly based on the provided context, which is itself stale or incomplete.

Governance as an Operational Variable, Not a Compliance Exercise

Perhaps the most practically insightful finding from KPMG’s survey is the strong correlation between AI maturity and risk confidence. While only 20% of organizations in the experimentation phase feel confident in managing AI-related risks, this figure jumps to 49% among AI leaders. Across all maturity levels, 75% of global leaders identify data security, privacy, and risk as ongoing concerns, but maturity dictates how these concerns are operationalized. This distinction is vital for boards and risk functions that often view AI governance as a restrictive measure. KPMG’s data suggests the inverse: robust governance frameworks do not impede AI adoption in mature organizations; they enable it. The confidence to accelerate deployment into higher-stakes workflows and expand agentic coordination across functions directly correlates with the sophistication of the surrounding governance infrastructure.

Practically speaking, organizations treating governance as a post-deployment compliance exercise are at a distinct disadvantage. Their deployment cycles are slower, as each new use case necessitates a fresh governance review. Furthermore, they face increased operational risk because the absence of embedded governance mechanisms means that edge cases and failure modes are discovered in production rather than in controlled testing environments. Companies that have integrated governance into the deployment pipeline itself – through mechanisms like model cards, automated output monitoring, explainability tools, and human-in-the-loop escalation paths for low-confidence decisions – are operating with the assurance that facilitates scaling.

“Ultimately, there is no agentic future without trust, and no trust without governance that keeps pace,” explains Steve Chase. “The survey clearly indicates that sustained investment in people, training, and change management is what enables organizations to scale AI responsibly and capture value.”

Regional Divergence and What It Signals for Global Deployment

For multinational corporations managing AI programs across diverse geographies, KPMG’s data highlights significant differences in deployment velocity and organizational posture that will inevitably impact global rollout strategies. ASPAC is demonstrating the most aggressive progress in agent scaling, with 49% of organizations actively scaling AI agents, compared to 46% in the Americas and 42% in EMEA. ASPAC also leads in the more complex capability of orchestrating multi-agent systems, at 33%.

Barriers to adoption also vary, with tangible operational implications. In both ASPAC and EMEA, 24% of organizations cite a lack of leadership trust and buy-in as a primary impediment to AI agent deployment, a figure that drops to 17% in the Americas. Agentic systems, by their nature, make or initiate decisions without per-instance human approval. In organizational cultures where decision-making authority is highly centralized, this can foster institutional resistance that technical solutions alone cannot overcome. The remedy lies in governance design: clearly defining the categories of decisions an agent is authorized to make autonomously, establishing triggers for escalation, and delineating accountability for agent-initiated outcomes.

The divergence in expectations around human-AI collaboration is also a critical consideration for those designing agent-assisted workflows on a global scale. East Asian respondents anticipate AI agents leading projects 42% of the time, while Australian respondents prefer human-directed AI at 34%. North American respondents lean towards peer-to-peer human-AI collaboration at 31%. These cultural nuances will necessitate tailored designs for agent-assisted processes across different regional deployments of the same underlying system, adding a layer of localization complexity that centralized platform planning often underestimates.

A key data point for CFOs and boards to note is that 74% of respondents believe AI will remain a top investment priority, even in the event of a recession. This could signify a genuine conviction in AI’s role in optimizing cost structures and enhancing competitive positioning, or it might reflect a collective commitment yet to be rigorously tested against actual budget constraints. The reality is likely a combination of both, varying in proportion across organizations. What is clear is that the window for organizations still in the experimentation phase is not limitless. If the top 11% of AI leaders continue to amplify their advantages – and KPMG’s data suggests the mechanisms for this are firmly in place – the imperative for the remaining 89% shifts from whether to accelerate AI deployment to how to do so without exacerbating the integration debt and governance deficits that are currently capping their returns.