KPMG's New AI Data Has a Blunt Message: Stop Layering AI on Top of Broken Processes

KPMG just published its first Global AI Pulse survey — a global snapshot of how organizations are deploying AI agents in 2026. The headline number is uncomfortable: despite companies planning 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 agents in ways that produce measurable enterprise value.

That's not a technology problem. That's a sequencing problem.

The survey didn't just surface the gap — it identified the mechanism. The organizations in that 11 percent are not deploying better models or spending more on compute. They're doing something structurally different: they redesign the process first, then deploy AI agents to operate within the redesigned structure. Everyone else layers AI onto whatever process already exists and wonders why the returns are incremental.

This is the single most important operational insight for any business owner thinking about AI right now.

The Copilot Trap

Most SMBs and mid-market companies start their AI journey the same way: someone reads about ChatGPT or sees a demo, and within a week there's a company-wide account and a Slack message telling everyone to "use AI in your work." A few weeks later, a few tools get bolted on — a copilot for email, an AI summarizer for meetings, maybe a chatbot for the website.

These tools are not bad. But KPMG's data shows exactly what they produce: incremental productivity gains. Employees save 20 minutes here, draft a reply a bit faster there. The aggregate ROI is real but modest, and the gains plateau quickly because the underlying process — the actual workflow those tools sit inside — hasn't changed.

This is what KPMG calls "layering models onto existing workflows." It's the path of least resistance and the path of lowest return.

Among the 11 percent of AI leaders in the survey, 82 percent report that AI is already delivering meaningful business value. Among the other 89 percent — most of them also using AI tools — that figure drops to 62 percent. Twenty percentage points is the gap between "AI is helping a bit" and "AI is compounding our advantage every quarter."

What Process-First Actually Looks Like

The inversion sounds abstract until you see it with a specific function.

Take a professional services firm with a proposal process that currently looks like this: sales call → notes sit in someone's inbox → account manager manually drafts proposal in Word → it gets reviewed, revised, sent. The firm buys a writing assistant and uses it to draft proposals faster. Result: proposals get written in two hours instead of four. The process still requires the same human touchpoints; it just moves faster.

Now apply the process-first approach. The redesigned workflow: intake call auto-transcribed and structured → AI extracts key requirements and matches against past successful proposals → agent drafts a complete, scoped proposal using your pricing logic → human reviews exceptions only → proposal sends. Result: proposals in 20 minutes, human time drops from four hours to 15 minutes, volume the firm can handle triples without adding headcount.

Same tools, completely different outcome — because the second version was designed around what the process could be, not what it already was.

In KPMG's data, IT and engineering functions show this clearly: 75 percent of AI leaders are using agents to accelerate code development, versus 64 percent of their peers. In operations and supply chain, the split is 64 percent versus 55 percent. These aren't marginal differences in tool adoption rates. They're differences in what those tools are actually doing.

The Hidden Cost of Skipping This Step

There's a common objection: redesigning a process before deploying AI requires more upfront work. That's true. It's also why most businesses skip it — and why most businesses stay in the "incremental gains" category.

The cost of skipping process redesign isn't just lower ROI. KPMG's data surfaces a subtler problem: it creates integration debt. Every incremental AI tool you add to an existing process is another component that needs to be ripped out or rebuilt when you eventually do redesign the process properly. The friction costs — engineering hours to retrofit AI outputs into legacy systems, the latency of retrieval pipelines built on top of poorly structured data, the compliance overhead of systems never designed to handle AI-generated decisions — compound silently.

Among organizations still in the experimentation phase, just 20 percent feel confident in their ability to manage AI-related risks. Among AI leaders, that figure rises to 49 percent. Risk confidence and deployment velocity are not in tension; they move together. Organizations that have embedded governance into their process redesign are the ones moving fastest — because they're not stopping for a fresh compliance review every time they want to extend AI into a new function.

The SMB Advantage Nobody Talks About

Here's what the enterprise-focused KPMG survey doesn't say explicitly but implies: smaller organizations have a structural advantage in this transition.

An enterprise with 10,000 employees and decades of accumulated process has enormous inertia. Redesigning any core workflow requires buy-in from multiple departments, legal review, change management across hundreds of people, and integration work across systems that were never designed to talk to each other.

A company with 50 employees can redesign its proposal process in an afternoon and have an AI agent running in the new workflow within a week. The $186 million average AI spend in the KPMG survey includes infrastructure, licensing, professional services, governance overhead, and integration work at scale. An SMB doing this right can achieve equivalent process leverage for $15,000 to $40,000 — because the redesign is simpler, the systems are fewer, and the team is small enough to actually change how they work.

The KPMG data also shows that 74 percent of global leaders say AI will remain a top investment priority even in a recession. If that's true — and the competitive dynamics suggest it is — the window for businesses still in the "adding copilots" phase is shorter than it looks. The 11 percent who are scaling agents are compounding their operational advantage every quarter.

Where to Start

The practical version of the process-first rule isn't complicated, but it does require honesty about which of your current processes are actually worth redesigning versus which ones just need a better tool.

Start with one process that has a measurable output — proposals sent, invoices processed, leads qualified, support tickets closed — and ask: if there were no constraints, how would this process work? Not "how can we make the current process faster?" but "what is the ideal version of this process, assuming AI can handle any step that doesn't require a human judgment call?"

That question usually surfaces a process that looks very different from what you're doing today. The tools to execute that redesigned process already exist. What most organizations lack isn't the technology — it's the willingness to start with the process instead of the tool.

That's the gap the KPMG data is actually measuring.

If you're not sure where to start, a structured AI audit is the fastest way to identify which of your processes have the highest redesign leverage — and which ones would just be faster broken processes with an AI attached. Learn more at /services or reach out at /contact.