Harvard Business Review published a piece today that anyone signing off an AI budget in education should read.

It's by three economists: Justine Hastings (Auger and the University of Washington), Chad Syverson (Chicago Booth) and Seth Zimmerman (Yale School of Management). Their question is simple. If almost everyone is using AI, where are the results?

The numbers don't add up yet

They pull together some uncomfortable figures.

An NBER working paper surveyed nearly 6,000 senior executives in early 2026. 69% said their company actively used AI. About nine in ten reported no measurable effect on productivity or employment.

McKinsey's State of AI survey, published late in 2025, found 88% of organisations using AI regularly. Only 39% reported any impact on profit at enterprise level.

And one 2025 industry report put the share of AI pilots with a measurable bottom-line effect at just 5%. That figure has been quoted a lot, and argued over a lot. It's useful as a warning light. I wouldn't lean on the exact number.

Where the value is hiding

The authors' diagnosis is about how AI gets deployed. Most tools improve work inside a single function. Finance gets quicker at finance. HR gets quicker at HR.

The bigger costs sit in the work between functions. They call it coordination work: pulling data out of different systems, reconciling information that doesn't agree, and getting teams aligned before anyone can act.

The article cites research suggesting employees spend 37% of their time on this kind of work. 77% would like to spend less.

If you've worked in a school, you'll recognise it straight away.

What coordination work looks like in a school

Take a pupil whose attendance is slipping.

The attendance data is in the MIS. Behaviour logs might be in another system. Safeguarding notes sit in a third. The SENCo keeps their own records, the pastoral lead has a trail of emails, and the form tutor has a hunch.

Getting a clear picture means somebody pulling it all together by hand. Usually under time pressure, and usually more than once.

Now look at how AI tends to arrive. Teachers get a tool for lesson planning. The business manager tries one for finance. The office uses Copilot to draft letters home.

Each might save time in its own corner. None of them touches the hours spent joining the dots between those corners.

Trusts have the same problem at a bigger scale. Central teams spend a lot of time collecting returns from schools, chasing differences between spreadsheets and building reports that are out of date by the time anyone reads them.

What the authors recommend

Their advice to leaders, in short:

  • Measure coordination costs before you automate anything
  • Put AI where the hand-offs between functions happen
  • Reward results across the whole organisation, so teams aren't judged only on their own patch
  • Redesign roles as AI takes on some of the coordination work
  • Track decision quality, cycle time and operating performance, and stop counting adoption

That last point is the one I'd underline for education.

Translating it for schools and trusts

A few practical places to start:

  • Map one process that crosses teams. Admissions, attendance follow-up, EHCP reviews or end-of-year reporting all work. Note who touches it, which systems are involved, how many hand-offs there are and how long it takes.
  • Ask staff where they re-key, chase or reconcile. They'll tell you, and they'll be specific.
  • Ask a new question before buying a tool. Which hand-offs does it remove? If the answer is none, but one team will be quicker, it may still be worth having. Just don't expect it to shift much at whole-school level. There's more on this in Before you buy that AI tool.
  • Change what you report. "80% of staff now have a Copilot licence" says nothing about whether pupils got support sooner. The time from a concern being raised to action being taken tells you far more.
  • Bring your Data Protection Officer in early. Joining up data across systems is exactly where pupil information can end up somewhere it shouldn't.

Microsoft made a similar point in its UK report this month. The bigger productivity gains come when organisations redesign how work gets done across a whole process. I wrote about that in The UK ranks eighth for AI use.

Read it with the caveats

Much of the HBR article sits behind a paywall. This piece draws on the summary and opening sections, which are free to read.

The 37% coordination figure is cited as research, but the free sections don't name the source. The survey data is from businesses, not schools. And one of the authors works for a company that builds AI products, which is fair to keep in mind.

Still, the central idea travels well. Schools and trusts are full of teams that each work hard and talk to each other through spreadsheets and email. That's where the time goes.

Which process in your school or trust has the most hand-offs, and is anyone measuring how long it takes?