Most firms starting on AI ask what their AI strategy should be. It’s the wrong first question, or at least the wrong first question in the wrong order.
The question that should come before it is quieter and more awkward: do we understand our own organisation well enough to know where AI actually belongs in it? Those two questions feel like the same thing. They aren’t. A strategy sets a direction. The question underneath it is investigative, and it has to be answered first, because a direction set without it can be confidently, expensively wrong.
Why the order matters
A strategy built on a shaky understanding of your own organisation isn’t wrong because the thinking was poor. It’s wrong because it’s built on a foundation nobody checked. Optimise a business for AI on top of that foundation and you don’t get a slow drift in the wrong direction. You get a fast one, because AI’s whole function is to make things move quicker, including the mistakes.
Understanding where you actually stand means getting close enough to the work to see where value is really created and lost, where the genuine bottlenecks are, which decisions are high-leverage and which only feel that way, and where human judgment is non-negotiable versus where it’s being spent on work that doesn’t need it. None of that comes from a strategy document. It comes from looking honestly at the organisation as it is.
The questions worth sitting with
A few are worth answering honestly before any budget is committed.
Where are decisions actually made, and how fast do they move? Can the people building things get the data they need without filing a request and waiting three weeks? Is technical debt being managed as a discipline or quietly deferred? Are leaders creating the conditions for honest work, or the kind of low-grade distrust that has people working around new systems instead of with them? And is there anything AI is already influencing that would surprise the people it affects?
None of these are new questions. Anyone who’s been through a serious transformation of any kind will recognise all of them. That’s not a coincidence. The signal you’re looking for in AI readiness is the same signal, in the same places, as the one you’ve seen before.
Failure is information
For firms already partway in, this is still readable. The tools that aren’t delivering, the pilots that quietly failed, the adoption that never stuck. Those aren’t just disappointments. They’re data, and they point at exactly the places the questions above point at. A failed pilot is telling you something specific about where the organisation wasn’t ready. It’s worth reading rather than burying.
The firms that get real value from AI aren’t the ones that adopted it fastest. They’re the ones that used it as the reason to finally do the organisational work they’d been putting off. The question isn’t whether AI is worth doing. It’s whether you know your own organisation well enough to know where it belongs, and most firms haven’t checked. That check is cheap compared to the alternative, which is finding out after the money’s gone.