By Brad Ferris · 25 July 2026
The most useful number in AI this quarter is not a benchmark score. It is buried in KPMG's Global AI Quarterly Pulse Survey for Q2 2026: organisations with full visibility into what their AI actually costs to run are five times more likely to report established ROI than those without it. Not five per cent more likely. Five times.
That finding sits inside a broadly optimistic read. KPMG reports that 76 per cent of organisations now say they are getting real business value from AI, up 12 points on the previous quarter, and that 22 per cent have reached what it calls the "driving adoption" stage, embedding AI into daily workflows rather than piloting it at the edges. After two years of the productivity-versus-returns argument, more businesses are finally landing on the returns side of it.
But the cost-visibility stat is the one worth pinning to the wall, because it explains the split. The businesses pulling ahead are not the ones spending the most on models. They are the ones who can tell you, to the dollar, what a given AI workflow costs to operate and what it returns. Everyone else is running on faith.
If you have deployed AI in the last eighteen months, there is a decent chance you cannot answer a simple question: what does that assistant, that automation, that agent cost you per month, and what has it saved or earned in return?
The reason is structural. AI costs arrive scattered. A per-seat licence here, a usage-based API bill there, a cloud line item somewhere else, plus the staff time to maintain prompts and review outputs. Each piece is small enough to wave through. Added together, and set against a return nobody has measured, they form a number no one owns. So the workflow keeps running, quietly, unexamined, and when the CFO finally asks whether it pays, the honest answer is a shrug.
Faith is a fine way to start a pilot. It is a poor way to run a business function. And it is precisely the gap KPMG's data exposes: value is available, but only to the organisations disciplined enough to measure the exchange.
Cost visibility is a symptom of something larger. McKinsey's State of AI Trust in 2026 puts the gap plainly: employee readiness for AI sits at around 70 per cent, while organisational readiness sits at just 27 per cent. Only 11 per cent of organisations have reached what McKinsey calls the reinvention stage, where work is genuinely redesigned around AI rather than bolted on to it.
Read those two numbers together. The people are ready. The operating model is not. Staff are willing to use the tools; the business has not rebuilt the process, the measurement, or the accountability around them. Cost invisibility is what that unreadiness looks like in the ledger.
This is also where the "bolt-on" tax shows up. Boston Consulting Group's work on redesigning the operating system of work found that end-to-end process redesign delivers cost reductions of around 60 per cent, against under 20 per cent for point automation that simply drops a tool into an existing workflow. Bolting AI onto a process you have not rethought captures a fraction of the value and, because it is a fraction, it rarely justifies the attention needed to measure it properly. The under-measurement and the under-return feed each other.
Here is the part that should encourage anyone running a business between five and three hundred staff. Cost visibility is a scale problem, and at your scale it is far more solvable than it is for the enterprise.
A large corporate has AI spend spread across dozens of departments, hundreds of vendors, and procurement layers that make a single view of cost a multi-quarter project. You do not. You can see your whole cost base. You can put every AI workflow, its monthly cost, and its measured return on one page, and you can do it in an afternoon rather than a fiscal year. The thing the KPMG data says separates the winners is the thing your size makes easy.
So the practical move is not to buy more AI. It is to take the AI you already run and give it a meter. List each workflow. Attach a monthly cost, including the staff time. Attach a return, even a rough one. Then look at the page. Some workflows will clearly pay. Some will not, and you will stop them. And the ones worth expanding will announce themselves, because you will finally be able to see them.
The organisations getting real value from AI this quarter are not smarter or better funded than you. They can just read the meter. That is a discipline, not a budget, and it is one a focused mid-market operator can adopt faster than almost anyone.
Where does your business stand? The free AI Scorecard takes three minutes and shows you. If you want a straight steer from a person, book an AI Opportunity Call.