Your AI portfolio is forty good ideas and no strategy

I’ve started opening leadership sessions by telling the room to stop wasting their time, and mine. It isn’t a polite way to begin, and I do it on purpose. By the time I arrive, most teams have already been through the exercise of hunting for use cases, and what they bring me is a list of ways to make the existing work faster or cheaper. Every item on it makes sense. Taken together, it’s a cost-saving programme dressed up as a strategy, and I’d rather say so in the first ten minutes than watch a capable team spend another year on it.

I’ve got two reasons, and the first is practical. The technology is moving so quickly that most efficiency builds are out of date before they’re finished: the tool you designed around at the start of the project has usually been overtaken by the end of it. The second reason matters more. When AI lets you do almost anything, a team that points it at everything through a cost-saving lens ends up doing nothing that changes its position.

Scarcity used to do the choosing

That second reason has a history. For most of my career, the hardest part of strategy wasn’t choosing what to do. It was getting permission to do anything at all. Every idea a leadership team had ran into a budget line, a headcount freeze or an IT backlog that stretched out past the next planning cycle, and those constraints did a quiet piece of strategic work that nobody ever gave them credit for. They forced the choice. You couldn’t do everything, so you did the few things that mattered most, or at least the few that survived the fight for resources.

AI has taken that constraint away, and I don’t think many leadership teams have noticed what went with it.

When resources were scarce, the constraint chose for you

When starting is cheap, nothing chooses

The wrong mental model

None of this is carelessness. These are smart people applying a mental model that has served them well for every technology they’ve bought before. In that model, AI is an automation tool: something that does the existing work faster and cheaper. So the natural question becomes where we can use it, and the honest answer is almost everywhere. Every function can find a task to speed up, and every one of those business cases stands up on its own. Approve them one at a time and you end up with forty initiatives, a respectable total of hours saved, and no change at all in how the business competes.

It’s the electric motor problem I wrote about in my last piece: factory owners put the new motor exactly where the steam engine had been, kept the old layout, and waited decades for a payoff that only arrived when someone redesigned the floor.

Why savings don’t compound

The automation model isn’t wrong about what AI can do. Where it goes wrong is on what AI is for. If the only value on offer is cost, then the right number of initiatives is as many as you can fund, because savings add up. Ten projects that each shave a little off a process give you ten small savings and nothing more, and your nearest competitor can buy the same tools and bank the same savings by the end of the next financial year. You’ve paid for parity and called it progress.

The better model starts from intelligence rather than automation. What AI actually offers an organisation is the ability to sense what’s happening, decide what to do about it and learn from the result faster than it could before, and learning has a property savings don’t: it compounds. A business that gets a little better every cycle at the thing that decides whether it wins (how it prices, how it reads a customer, where it puts its capital) ends up somewhere its competitors can’t reach by licensing the same software. The gap it opens is in what it knows, and that’s much harder to copy than a workflow. That’s where the real prize sits: new revenue, a different way of organising the business, and the kind of market position that efficiency on its own has never bought anyone.

Efficiency projects add up

Sense

Reason

Act

Learn

The question that creates focus

That model hands leaders a question the automation model never asks. Which capability, if we learned faster at it than anyone else in our market, would change our position? It’s a hard question, and most executive teams I work with haven’t sat with it for longer than a slide. But answer it honestly and focus stops being a discipline you have to impose from the top. It becomes the obvious consequence. Most of the forty initiatives fall away, not because they’re bad ideas, but because they teach the organisation nothing about the thing that matters. The handful that remain all point at the same problem, so each one makes the next one smarter. I’d bet most executive teams could cancel two thirds of their AI initiatives tomorrow and lose nothing they’d miss. When anything is possible, the strategy lives in what you refuse.

When anything is possible, the strategy lives in what you refuse.

That’s what I mean by cohesion. A tidier portfolio is a by-product. What you’re really after is a set of efforts that learn as one system, rather than forty experiments each learning alone and reporting to a different steering committee.

Isn’t experimenting the point?

I should own a tension here, because I’ve spent years arguing that learning velocity is the metric that matters, and you could read that as permission to run as many experiments as you can. Plenty of people I respect make exactly that case, and early on they’re right. You can’t find the strategic question without trying things and getting some of them wrong. But experiments that circle one question behave very differently from experiments that each answer a different one. The first kind compounds. The second kind produces activity, and a dashboard full of activity is remarkably good at looking like learning. I don’t think there’s a clean rule for when exploring should give way to focus. My instinct is that most organisations have stayed in the exploring phase a year longer than they needed to, mainly because nothing forced them out of it.

The teaching test

That piece was really about the story leaders owe their people: a picture of an AI future worth living in. This is the same problem from the inside. A story needs a direction, and inside a business, direction looks like saying no to a lot of things that would each, on their own, have been perfectly reasonable.

So here’s what I’d ask any executive team to do before their next planning round. I call it the teaching test. List every AI initiative you’re funding, and next to each one write a single line on what it’s teaching the organisation, rather than what it’s saving. Then read that second column on its own. If those lines point at the same capability, you’ve got the beginnings of a strategy. If they scatter, or half of them are blank, the conversation worth having next is about what you want to be better at than anyone else in three years’ time, and which of those initiatives is helping you get there.

The teaching test

Initiatives teaching forty different things

Initiatives pointing at one capability