Five Questions That Decide Whether Your AI Spend Becomes Revenue

Two organisations spend the same money on AI this year. One reports revenue growth from it. The other reports a pilot portfolio. The variable that separates them is not the budget, the vendor, or the model — it is whether a single named person is accountable for an outcome. That claim is now measurable.

Grant Thornton surveyed 950 senior business leaders between 23 February and 18 March 2026 — CEOs, CFOs, COOs, CIOs and CTOs, plus direct reports to the C-suite, across ten industries. The 2026 AI Impact Survey found that organisations with fully integrated AI are nearly four times more likely to report AI-driven revenue growth than those still piloting: 58% versus 15%.

Two caveats before that number does any work. It compares organisations at different stages of integration, not projects; it does not by itself prove that integration causes revenue. And the survey is of large and mid-large enterprises, not the ten-person company. What makes it useful anyway is the second number Grant Thornton reports alongside it, because that one is about accountability rather than technology: 74% of fully-integrated organisations are very confident they could pass an independent AI governance audit within 90 days. Among those still piloting, 7% are. A tenfold gap.

Meanwhile the money keeps moving. Gartner forecasts worldwide end-user spending on AI models and platforms at $64 billion in 2026, up 63.4% from $39 billion in 2025. Nobody is under-investing. The 15% group and the 58% group are both writing cheques.

The proof gap is not a technology gap. It is an ownership gap wearing a technology costume.

Here are the five questions we ask in a diagnostic. Each one has a right answer that is a person's name, and each maps to something Grant Thornton measured directly.

1. Who owns the P&L line this is supposed to move?

Not the tool budget. The outcome line. If AI is meant to lift gross margin in the service business, the person accountable for gross margin in the service business owns it — and their number changes if it works.

This is the question most companies skip, and the survey shows the cost. Only 22% of operations leaders are working with a fully developed and implemented AI strategy. Grant Thornton's phrasing of the failure is precise: organisations are "succeeding on breadth with more pilots, more use cases, more functions touched by AI, but they are failing on depth." Breadth is easy to fund and easy to report. It also has no owner, because a portfolio of eleven experiments belongs to everyone.

A named P&L owner does something a steering committee cannot: they will kill an initiative that is costing them margin. That is the mechanism, not the paperwork.

2. What is the exit criterion — written, before you start?

Grant Thornton's own recommendation is worth reading twice: the organisations pulling ahead "are not scaling more pilots — they are scaling fewer, with better measurement and clearer exit criteria."

An exit criterion is a sentence written before the pilot begins that says what result, by what date, justifies moving to production — and what result triggers shutting it down. Without it, every pilot has the same terminal state: it neither ships nor dies, it just quietly stops being mentioned. Companies then carry the run-rate and the risk of an initiative nobody believes in, which is how the pilots-never-reach-production problem becomes structural rather than technical.

3. Who signs the audit?

Ask your leadership team a specific question: if an outside party arrived in 90 days and asked how a particular AI-influenced decision was made, who produces the evidence and whose signature is on it?

Across the whole survey, 78% lack strong confidence they could answer that. And the board layer is thinner than the deployment layer: three in four boards have approved major AI investments, but 48% have not set AI governance expectations and 46% have not integrated AI risk into ongoing oversight. Approval without expectation is not governance; it is funding.

We have written separately about what happens when returns are expected and incidents have already occurred. This question is upstream of that one. It is not "are we safe" — it is "can we show our work," which is the thing that determines whether you can scale at all.

4. Who owns the workflow redesign — not the tool rollout?

The most revealing finding in the survey is a disagreement. CIOs and CTOs are five times more likely than COOs to say the workforce is fully ready to adopt AI. Those two executives are describing the same company and reaching opposite conclusions, which means at least one of them is wrong about the operation they jointly run.

The supporting numbers say who. Training is the most underfunded AI investment area in the survey, with 34% of finance leaders saying it is not getting enough. Only 6% of executives name change leadership and workforce enablement as a top skill for thriving in an AI-driven environment. Frontline employees (37%) and middle managers (30%) were identified as the groups needing the most support — the people closest to the work.

Grant Thornton's conclusion is blunt: organisations are "treating AI like an IT project," when it is a change-management initiative. Somebody has to own the redesign of how the work actually happens — the operating model, the handoffs, the decision rights — and that person is almost never the one who signed the software contract. This is also where the thinning manager layer bites, because middle managers are the mechanism by which redesigned work becomes actual work.

5. If an agent is wrong tomorrow, who explains it?

Nearly three in four organisations are giving agentic AI access to their data and processes. 20% have a tested AI incident response plan.

Most have not handed over the keys entirely — 5% allow agents to execute high-stakes decisions without human review, and 60% limit agents to moderate-risk task automation. But the exposure is already live, and the C-suite is not aligned on it: 54% of COOs are concerned about regulatory and compliance uncertainty around agentic AI, against 20% of CIOs and CTOs. When the people deploying the technology are not worried about what the people running operations are worried about, the gap itself is the risk.

The useful framing is that nearly every organisation has already built this capability once, for cybersecurity: a named owner, a tested runbook, a post-incident review. The elements transfer. What does not transfer automatically is the name at the top of the runbook.

What this costs to fix

Run the arithmetic on a mid-market company spending, say, $500,000 a year across AI tooling, integration and internal time. The Grant Thornton data does not let you multiply that by a return — the 58%/15% split is a stage comparison, not a per-dollar coefficient, and treating it as one would be laundering. What the data does support is narrower and more actionable: the organisations that can name an owner and produce an audit trail are the ones reporting the growth.

So the question is what it costs to put a name on it. For most companies in the $10M–$250M range, a full-time Chief AI Officer is the wrong instrument — the mandate is real but it is not a 40-hour-a-week mandate for the first year, and the compensation math reflects that. A fractional Chief AI Officer exists to hold exactly the five answers above: the P&L line, the exit criteria, the audit signature, the workflow redesign, and the incident runbook. Not to run pilots. To own outcomes.

And the sequencing matters more than the spend. Adoption rates vary enormously by company size, which means the right first move for a 60-person manufacturer is not the right first move for a 900-person services firm. The diagnostic exists to answer that before the tooling decision, not after.

There is a version of this argument that sounds like a governance lecture, and that is not the argument. Grant Thornton found that organisations with stronger governance adopt faster — they move decisively because they can defend what they built. The companies stuck at 15% are not being reckless. They are being diffuse. Eleven initiatives, no owners, no exit criteria, and a board that approved the budget without setting the expectation.

Fewer things, owned by name, taken all the way to production. That is the same discipline we apply to a search, and it works for the same reason: depth is what produces outcomes, and depth requires somebody whose name is on it.