// in this post
If you've paid for AI licences and you're still waiting for revenue to move, you're in large company. In PwC's latest survey of 4,454 CEOs, 56% said AI hadn't yet delivered any significant financial benefit, and only 12% said it had both cut costs and raised revenue (PwC).
I think much of that disappointment comes from asking AI to do a job it was never going to do. AI doesn't increase revenue. People do, through the decisions they make, the judgement they use and the ideas they have. What AI can do is give them more time to do it.
While some firms that invest more in AI have grown their sales, a study in the Journal of Financial Economics found that growth came through new products, with no sign of it coming from lower costs (Babina et al.). Somebody still had to decide which products to build.
This piece looks at where a manager's time actually goes, why the hours AI saves tend to disappear, and four decisions that turn those hours into ideas that make money.
The people who know where the money is hiding have the least time to look for it
In my experience, a manager's job comes in three parts: operations, strategy and people. The mix changes with seniority, team size and resources, but at any given moment, that's the job. And nearly every manager I've worked with has told me the same two things. There's never enough time for strategy, and the people side is the hardest part.
The research backs this up. In a McKinsey survey of 706 middle managers, they said they spent nearly a full day a week on admin, under a quarter of their time on strategy and under a third on their people (McKinsey). A 2025 YouGov survey of 1,000 operational managers in Australia, the people running plants, sites, stores and venues, found they lose about six and a half weeks a year to low-value work such as email, scheduling and typing the same data into different systems (IT Brief).
A CEO's day, in my experience, leans towards decisions and strategy, while an operations manager's leans towards firefighting. CEOs aren't immune, though. In a Harvard study of 27 CEOs, about 36% of their time went on reacting to issues as they came up (HBR, MGMA). And how a CEO uses the rest seems to matter. A study of 1,114 CEOs in six countries found firms tended to do better when the CEO spent their time in high-level meetings across functions, as long as that style suited the firm (Bandiera et al.). In a 50 to 500-person business, the CEO is often close to day-to-day operations as well.
Put that together and most organisations have the same problem. The managers closest to the work know where the money leaks out and where the next good idea might be, and they're the ones with the least time to think about it.
What happens when people get time to think
About ten years ago, at HelloFresh, I ran a mentoring programme for some of the company's strongest managers. Each had a small group and a senior executive as a mentor, and each group was given a real business problem to solve.
The ideas that came out of it were often simple. One group looked at how the order of meals on the menu shaped what customers picked, given that some recipes cost far more to make than others. Another looked at what else customers might want to add to their box. Neither idea needed new technology. They needed people who knew the business, a reason to look and the time to do it.
The menu idea has research behind it. In a study at Flinders University, people choosing from a 12-item online menu were 30 to 40% more likely to pick a healthy item when it sat at the top of the list. On a printed menu, position made no difference (Appetite). It's the kind of pattern you only notice when you know the menu, the costs and the customers, and you've been given an afternoon to think about it.
The principle isn't new, either. 3M has long let its researchers spend 15% of their time on projects of their own choosing, and credits the Post-it Note to it (3M).
The catch: AI won't hand the time back by itself
This is where most of the AI argument goes wrong. It assumes that if AI saves someone an hour, they'll spend that hour thinking, and the early evidence says they mostly don't.
Researchers at Berkeley followed 200 people at a US technology company from April to December 2025. Using AI didn't lighten their work. It made it more intense: they ran more tasks at once, switched between them more often and spent more time checking what the AI had produced (HBR). In Denmark, a study of 25,000 workers found AI users saved about 3% of their working hours, and that the tools had created new tasks for 8.4% of workers (Humlum and Vestergaard).
Left alone, the time AI saves fills up with more admin. If you want it spent on strategy and people, someone has to decide that and then protect it.
There's a second catch. In October, Anthropic published a study in which 50 professional data analysts worked through an analysis with an AI agent. Without the analysts knowing, each agent had been set at random to lean one way on each judgement call. The analysts went with the agent's lean 86% of the time, most of them explained their choices using the agent's arguments, sometimes word for word, and only 15% mentioned the agent at all. Run with no lean, the same agent made the same five choices 91% of the time, which matters if your competitors are asking it the same questions (Anthropic). Freed-up time only pays off if the judgement stays with your people.
Here's how: four decisions that make AI pay off
1. Point AI at the admin, and say so
Be specific about which operational work AI should take off your managers: the weekly report, the rota, the chasing, the re-keying. The firms making real money from AI changed how the work flows. In McKinsey's 2026 survey, only 6% of firms got 5% or more of their profit from AI, and about three-quarters of those had redesigned their workflows, compared with about a quarter of everyone else (McKinsey). If you want a starting point, my Manager Prompt Pack has 60 prompts for the jobs that fill a manager's week.
2. Decide what the time is for, in writing
Agree with each manager how much time AI should free up and what it's for, then put it in the diary like any other commitment, the same way you'd protect time for AI champions. Time that's promised "when things calm down" goes to the next fire.
3. Give the time a purpose
This is the decision that turns saved hours into ideas, and it's the one most businesses skip. Free time with no shape gets filled, as the Berkeley study shows. The HelloFresh programme worked because the time had a shape, and you can copy it.
A real problem with money attached. Ask your leadership team for the questions they'd most like answered this quarter: why a product line's margin is slipping, where customers drop off, where the waste is. A training exercise won't get the same effort.
A small group close to the work. Four to six managers from different parts of the business, picked because they know the detail rather than for their grade.
A senior sponsor. Someone from the leadership team who meets the group regularly, opens doors and asks awkward questions. Giving a group an executive's time tells everyone the work matters.
Their view first, then the AI's. Ask each person to write down what they think before they ask AI, then ask the AI to make the case for the option it didn't pick. In Anthropic's study, the agents rarely explained why a good expert might choose differently, so the analysts rarely heard the other side.
A deadline and an audience. Six to eight weeks, then a slot to present to the leadership team. A date in the diary does more for focus than any amount of encouragement.
A plan for good ideas. Decide before the groups start who tests a promising idea and with what budget. Otherwise the best ideas get a round of applause in the presentation and stall the week after.
AI's part in all this is the hours. If it takes most of a day a week of admin off a manager, half a day of that can go to the group.
4. Measure the ideas, not the usage
Most AI dashboards count logins and prompts. They tell you people are using the tools and very little about whether anything has changed. Count what the time was meant to produce: problems looked at, ideas tested, processes rebuilt, decisions made. In McKinsey's survey, the firms getting the most from AI were twice as likely to have a defined way of measuring its impact (McKinsey).
Where to start
Pick one problem, one group and one sponsor, and put the presentation in the leadership team's diary for eight weeks' time. Then work out which admin AI can take off those managers so they have the hours to do it.
If you'd like a hand setting it up, my AI champions programme takes up to 12 people through three months of fortnightly clinics on real work. The first conversation is a free 30-minute call.
FAQ
Why isn't AI paying off for most businesses?
Mostly because the time it saves isn't put to use. In PwC's 2026 survey, 56% of CEOs said AI hadn't yet delivered a significant financial benefit. Saved hours tend to fill up with more admin, and the firms getting real profit from AI are the ones that redesigned how work gets done and gave the freed-up time a purpose.
Why isn't AI increasing revenue?
Because revenue comes from decisions about products, customers and prices, and a tool doesn't make those. AI saves time and takes on routine work. Revenue grows when people use that time to find better ideas and act on them. In McKinsey's 2026 survey, the few firms getting a real share of their profit from AI were the ones that had redesigned how their work gets done.
What should managers do with the time AI saves?
Decide before it's saved. Left alone, the hours fill up with more tasks. Agree what the time is for, give it a real business problem and a deadline, and put a senior sponsor behind it.
How do you know whether AI is paying off?
Count what changed because of the time it freed up: problems looked at, ideas tested, processes rebuilt, decisions made. Logins and prompt counts tell you people are using the tools and very little else.