Lessons that appeared when the plan met reality.

Most transformation programmes don't have a technology problem

Every stalled programme I've seen gets diagnosed the same way. Wrong platform. Wrong integration. Wrong vendor. So the fix becomes another tool, another migration, another year.

The tool is rarely the thing that's stuck.

What's stuck is the decision. Twenty markets, one proposed standard, and no clear answer to a plain question: who actually gets to say yes? When that answer is vague, every market assumes the answer is "me." You don't end up with one operating model. You get twenty, wearing the same logo.

I learned this the expensive way. I built a clean global standard, then granted one market a sensible exception. Within a quarter, three more markets had pointed at that exception and asked for their own — and the standard I'd shipped was already fiction. I'd never decided who got to say no, so nobody did.

So I've learned to design the decision before the system. For any real choice — pricing logic, a customer journey, a single data definition — three things have to be explicit and written down: what is global and not up for debate, what is genuinely local and theirs, and who breaks the tie when the two collide. Most governance skips the third. That's the one that matters.

Done well, this doesn't slow anything down. The opposite. A market that knows exactly where its freedom ends stops arguing every standard and starts moving. The teams that feel most autonomous are usually the ones working inside the sharpest lines.

Committees don't fix this. A committee is where a decision goes when no one will own it.

So before I touch the technology, I ask the uncomfortable question first: when this goes live and two leaders disagree, who decides — and does everyone already know their name?

That's usually where the real programme begins.

AI needs a place in the commercial system

Most AI in commercial teams dies the same quiet death. A promising pilot, a good demo, a slide that gets applause. Then nothing moves, because the pilot was never attached to anything that actually runs.

That's the mistake. Treating AI as a project instead of a place.

AI doesn't create value floating next to the business. It creates value when it improves a workflow that was already going to happen — a lead getting prioritised, a next action being chosen, a rep deciding who to call, a customer getting an answer. If there's no governed workflow underneath, the model is just a clever answer to a question no one was accountable for.

So the question I ask isn't "where can we use AI?" It's narrower and more useful: which decision, made many times a day, is currently made on gut or on a stale report? That's where AI belongs. Not because it's impressive there — because that's where a small improvement, repeated at scale, actually shows up in the number.

In a seven-market e-commerce business I led, AI became useful when it stopped being a separate initiative and entered the commercial rhythm: audience selection, personalisation and the decisions that followed. The value was not the model itself. It was better decisions, repeated inside a P&L someone actually owned.

And it only holds if the guardrails come first. Who owns the workflow the model is touching. What the model is allowed to decide and what stays with a person. What happens when it's wrong. In a regulated business those aren't slowing-down questions — they're the reason the thing is allowed to run at all. Skip them and you don't get speed. You get a capability legal makes you switch off six months in.

The teams getting real value from AI aren't the ones with the most models. They're the ones who were honest about which decisions were worth improving, and disciplined about who stays accountable when a machine gets a vote.

AI doesn't need a strategy of its own. It needs a job inside the one you already have.