AI works. Does your organisation?

Twenty months ago I wrote my first newsletter on AI. About Laloux, about how we see people in organisations. Five editions later the question has become a good deal flatter: is any of that money coming back? McKinsey published its State of AI at the end of August. Nine in ten organisations use AI. Eighty per cent of employees say it makes them more productive. And the share of companies seeing it in the profit line: 37 per cent. Exactly the same as last year. The front-runners, with at least five per cent of EBIT from AI: six per cent. Also unchanged.

So AI works. For people. For organisations, it does not. And that difference has an address. The gap is at the top

MIT put it more bluntly last year: 95 per cent of organisations get zero measurable return from their AI pilots. Meanwhile, at more than ninety per cent of those same companies, employees simply use their own ChatGPT or Claude for work. The shop floor has already crossed over. The executive floor is still standing on the bank.

Let me say it as I see it: I meet alarmingly few executives and non-executives who have genuinely immersed themselves in AI. Who know what a language model can and cannot do, what it does to their business model, and how to put an organisation to work on the subject in a structured, systematic way. Plenty who call it "important", have a pilot running and have parked the rest with the CIO. That is not a strategy. That is delegating what you do not understand.

The questions you should be asked

The supervisory board wants to know three things: which bet are we making, who is personally accountable, and where do I see it in the numbers? The executive board has to make the choices beneath that: which three domains, not thirty? Build or buy? Do we redesign the work, or bolt AI onto what is already there? Anyone answering "bolt on" belongs with the 63 per cent who still see nothing.

What the winners do

In its management playbook, McKinsey studied twenty companies that are pulling it off: DBS, Freeport-McMoRan, LATAM, Toyota. Twenty per cent more EBITDA, three dollars for every dollar invested, cash-positive within two years. Their recipe is dull. At most three domains, precisely where a small improvement makes a lot of money. Systems, not stand-alone tools. Workflows redesigned end to end: three-quarters of the front-runners do it, a quarter of the rest. And at the top: a C-suite that understands AI itself. Not delegates. Understands.

A small jab: the firm that measures that 94 per cent create no value sells, in the very same article, the second edition of its book and the transformation that fixes it. The figures hold up. The self-assurance you take with a pinch of salt.

When to change course

Two checks. One: cash-positive within two years? Anyone still reporting "promising pilots" after that does not have a pilot; they have a hobby. Two: do not believe your own forecasts. Last year 32 per cent expected AI to shrink their workforce. Fourteen per cent saw it happen. Now 39 per cent expect a decline. Steering on expectations rather than realised EBIT is steering on air.

What this demands of people

The decisive capability is not a technology but a person: the business leader who combines domain expertise, an understanding of data and the ability to lead change from start to finish. That profile is scarce. And it rarely appears in the search brief.

From board to shop floor, the requirements are shifting. Non-executives must be able to judge an AI bet without being dazzled by slides. Executives must choose what not to do. And the middle layer carries the load: 47 per cent of middle managers and individual contributors report workload, insecurity and fatigue caused by AI, against 31 per cent of executives. Devised at the top, felt at the bottom. As long as the top regards AI as something you have done for you, that will not change.

In short: AI works; that much is proven. Whether your organisation works depends on one thing that comes before everything else: a top team that grasps the subject itself and takes personal ownership of it. Only then come focus and redesign.

You are welcome to exchange thoughts on this — on the leaders who can carry an AI transformation from boardroom to shop floor, and on finding and selecting the best, right candidates for leadership roles in your organisation.

Warm regards,

Aegeus