83% of Companies Can't Prove Their AI Spending Pays Off — the Opening for Founders

Research commissioned by Air IT Group and reported by PCR on November 10, 2026 found a striking gap at the heart of enterprise AI spending. Of 504 UK decision makers surveyed, 83% either could not demonstrate returns from some of their AI spending or did not measure returns at all. Companies are spending heavily on artificial intelligence, and most of them cannot show what they are getting for it.

The same period offered a vivid illustration of how fast the spending is growing. IBM and Microsoft launched Imagine Enterprise on November 10, a platform offering more than 6,000 prebuilt agents. And PYMNTS Intelligence found that agentic AI is already in use at 89% of large U.S. enterprises — but that 78% of those enterprises would restrict agents to read-only directory access.

Adoption is nearly universal. Measurement is rare. Trust is limited. That combination is the defining feature of the enterprise AI market today — and the biggest opening for founders.

The Gap Between Adoption and Measured Return

The 83% figure deserves careful reading. It does not say that AI spending is wasted; it says that most companies cannot demonstrate what it returns. Some of that spending is certainly paying off. But if a company cannot measure the return, it cannot distinguish the spending that works from the spending that does not, and it cannot make a rational decision about what to buy next.

This is a familiar pattern in enterprise technology. A new capability arrives, excitement builds, budgets flow, and measurement lags behind. The companies that eventually win are the ones that close the measurement gap — and the vendors that help them do it become essential.

For founders, the implication is clear: selling AI tools is becoming a crowded business, but selling measurable outcomes is not. A product that can show, in numbers, what it saves, earns or improves has a decisive advantage over one that can only promise.

Why Measurement Is So Hard

There are genuine reasons why AI returns are difficult to measure. AI systems are often deployed across many workflows at once, making it hard to attribute a specific outcome to a specific tool. Their benefits — faster decisions, better drafts, fewer errors — are real but diffuse. And many organisations bought AI tools before they defined what success would look like, so there is no baseline against which to measure improvement.

But difficulty is not an excuse, and buyers are losing patience with it. Finance teams, who ultimately approve technology budgets, want numbers. A vendor that arrives with a measurement framework — baselines, metrics, dashboards, audit trails — is speaking the language of the people who sign the cheques.

Measurement, Auditability and Guardrails as Product Features

Guardrails are the other half of the equation. The PYMNTS finding that 78% of large enterprises would restrict agents to read-only directory access is a trust problem stated in numbers. Enterprises want the benefits of agentic AI, but they do not trust agents with write access to their systems. They fear mistakes, misuse and loss of control.

Founders who solve the trust problem will unlock the much larger market of agents that are allowed to act, not merely to read. That requires guardrails that are demonstrable: permission systems, approval workflows, rollback mechanisms, and audit logs that show exactly what an agent did and why. Trust, in this market, is a product feature — and it is one that most vendors have not yet built properly.

The Opening for Founders

The combination of the three facts — 83% cannot prove returns, 6,000 prebuilt agents are now on offer from IBM and Microsoft, and 78% of enterprises restrict agents to read-only access — describes a market that is saturated with supply and starved of proof and trust.

Founders who enter this market should not compete on the number of agents they offer or the sophistication of their models. They should compete on outcomes. Sell a measured result: hours saved, errors reduced, revenue recovered. Build the measurement into the product so the customer can see the return without hiring a consultancy to find it. Build the guardrails so the customer can let the agent act without fear.

The enterprise AI market is not short of tools. It is short of evidence. The founder who supplies the evidence — and the trust that makes the evidence credible — will find customers who are ready to buy.

Angeled.com connects founders with investors.