Defining AI Strategy

The thing that genuinely fascinates me about AI strategy as a concept is how many organisations have produced documents with that title that contain, functionally, no strategy whatsoever.

That's a bit sweeping. Plenty of organisations have thoughtful AI strategies.

I'm sure they do. And then there are the rest of them, who have a slide deck with a circle diagram and the word 'transformation' used as a full stop at the end of every sentence. Those are the ones we are, statistically, more likely to encounter.

The circle diagram is actually doing real work here though. Four components — business goals, data and infrastructure, AI capabilities, risk and ethics. That's not decorative. Those genuinely are the load-bearing walls of any serious AI strategy.

I don't dispute the framework. I dispute what people do with it, which is typically to nod at all four components and then go and do whatever they were already planning to do, now with the word 'AI' attached.

Which is exactly why the lesson makes a point of saying alignment isn't incidental — it's the whole mechanism. If your AI initiative doesn't connect to a specific business objective, it's not a strategy, it's a hobby.

That is, I will admit, a reasonable way to put it.

I'll have that framed.

The part that people reliably mishandle is the risk and ethics component — not because they disagree with it in principle, but because they schedule it for later. 'We'll address ethics once we've got the thing working.' That is precisely backwards and yet it is the default approach in a depressing number of organisations.

It's not always cynicism, to be fair. Sometimes it's just that risk feels abstract at the planning stage and very concrete after something goes wrong.

Yes. And the lesson content makes the obvious-but-apparently-necessary point that proactive risk mitigation costs considerably less than retrospective crisis management. The reason it needs to be said is that people keep discovering it the expensive way.

There's something worth flagging here that the lesson touches on but I think deserves emphasis — the framing of AI strategy as a business transformation plan, not a technology plan. That distinction does a lot of work. It shifts who owns the strategy.

It shifts who is accountable for it, which is why it makes a number of people in technology roles quietly uncomfortable. If AI strategy is a business transformation question, then the CTO does not get to run it from a server room and present findings quarterly.

And equally, business leaders don't get to hand it off entirely and then complain when the outputs don't match their objectives. It goes both ways.

Which is, when you lay it out plainly, not a complicated idea. Decide what you want to achieve. Check you have the data and infrastructure to support it. Understand what AI can and cannot actually do. Manage the risks before they become incidents. The framework is not subtle. The execution is where organisations discover things about themselves that they would apparently prefer not to know.