The most useful thing anyone said to me this week was not an insight. It was a rule.
An IT strategy lead at a gas producer told me his leadership team had already agreed, in writing, that time saved is not value. His words: “we basically got the leadership team to agree with us that time saved is not value … it’s not because you save two or three hours here and there that the bottom line is evolving, or the top line is evolving either.”
He got that agreed before a single use case was approved. It is the smartest AI decision I have seen a mid-market buyer make this year. It also breaks the business case that almost every provider, mine included, has been pitching for two years.
It is budget season, and every AI proposal leads with hours saved. In five days I heard the same objection from three organisations that have never spoken to each other.
The head of business systems at a construction group of just under a thousand people put it plainly: “I always find those ROIs a bit dodgy at the best of times in the time savings.” Then he did the maths: “how do you know that you’re spending ten grand on tokens? Cool. If you’re saving a hundred grand in time, fine — but actually reliably measuring that is…” He did not finish the sentence. That was the honest part.
An innovation lead at a not-for-profit of 800 to 1,000 staff has a monthly report that takes 28 to 30 hours. He knows the hours exactly. He does not know what happens to them once AI frees them up, and he admitted it without being asked.
Three buyers, one week, same conclusion. None of them are sceptical about AI. They are all spending. They are sceptical about measuring it in hours.
They treat hours saved as value. A saved hour is only worth something if one of three things happens: it goes into revenue-earning work, the team handles more volume at the same cost, or a role does not need backfilling. If you do not name which one up front, the hour just disappears into the day. The saving looks good on paper but never shows up in a number anyone reports.
Compare how the gas producer framed their invoice project. They did not pitch it as faster invoice processing. They pitched it as: today we match about ten per cent of invoices to contracts and purchase orders. We want to match one hundred per cent, without adding staff. That is a goal you can measure. It stands up to a board question. “Saves the team three hours a week” does not.
The second mistake is timing. Most organisations only decide how to measure value after the pilots come back with soft numbers. Agreeing the rule first is the cheapest governance decision in AI. It costs almost nothing.
Our own AI projects burn through risk and contingency faster than any other service we deliver. Our PMO flagged it this month. I would rather say that openly than have a client find out.
Why does that matter? If a provider cannot yet predict their own AI delivery costs, they have no business confidently predicting your benefits. There is also a simpler reason our proposals are measured in hours: hours are what we bill. The business case is measured the same way the invoice is. Few providers say that out loud, because it means admitting the measure was never neutral.
Before you approve anything, ask where the saved hour goes: revenue work, extra volume, or a role not backfilled. “Capacity” is not an answer. It is a way of avoiding one.
Ask your provider to show you a business case from the last twelve months that was not based on time saved. If they cannot, they are measuring what they sell, not what you buy.
Then get your leadership team to agree the rule before the use cases arrive. The gas producer’s edge over almost everyone I met this week is not technology, budget or appetite for risk. It is that they did things in the right order.
Do not ask what AI will save you. Ask which number the board will see change in ninety days, and who has put their name to it. Then watch how many proposals change shape.