Stop buying AI tools. Start buying AI returns.

There is a gap in this market that nobody seems to want to fill.

Every vendor, every consultancy and every training provider sells the same unit of value: hours saved. It is easy to measure, easy to demonstrate, and it is not a business result.

What the market sells

Look at how AI enablement is priced and pitched and you find a consistent shape. Tools per seat. Training per head. Time saved per person per week. All inputs.

Of thirty-eight firms in a 2026 UK pricing benchmark, only three published a price at all. Most require a call before they will tell you what anything costs, which tells you something about how confident they are in the value.

What nobody sells

The question after hours saved is the one that matters: where did the hours go, and what were they worth there?

That question is harder. It requires knowing which lines of your P&L are movable, what each one is worth per pound, and being willing to say afterwards whether it worked. It also requires the person selling to understand a P&L, which narrows the field considerably.

Why the gap persists

Partly because it is uncomfortable. A vendor who ties their work to EBITDA can be measured against it. Hours saved cannot really be disproved.

Partly because it is a different skill. Knowing which model to use is not the same as knowing whether overhead efficiency or sales mix is the bigger prize in a particular business.

What to do instead

Three changes, none of them technical.

Decide the destination before the tool. Name the driver the workflow is meant to move, and the number you expect it to move, before anyone opens an application.

Take a baseline. Adoption, process time, and the outcome metric. Before, not after.

Redeploy deliberately. Freed capacity that is not consciously reallocated gets absorbed. Decide where it goes, then check in ninety days whether it went there.

None of this is about AI. It is ordinary operating discipline, applied to a category that has so far been allowed to avoid it.

The test

If your AI programme cannot answer "what did it return" in pounds, it is not finished. That is not a criticism of the technology, which mostly works. It is a gap in how the work is being scoped and sold.

Business fundamentals first. Technology second.