For AI Transformation

Most responses to that are not answers

“We need an AI strategy” is a budget line. “We can’t fall behind” is a fear, and a fair one, but it won’t tell anybody what to build on Monday. “We’re rolling out copilots across support” names a tool and a department, which is a plan for this quarter rather than a description of the company you’re trying to become.

A real answer names something that changes outside your building. What can a customer do afterward that they can’t do now? What can you sell that you couldn’t sell before? Who can you serve that you have been turning away? Write that down and you have a sentence worth measuring everything else against, including the vendors.

Two projects, one name

“AI transformation” gets used for two ambitions that have very little to do with each other.

One is doing what you already do, for less. Fewer hours in the same process, the same product out the other end, a smaller number at the end of the quarter. That one measures itself, because last year’s cost tells you whether it worked. If that’s the whole plan, you don’t need us. It rarely is, though: the hours you free up can go into building something new, and deciding what is already a question about the future shape of the company.

That’s the other ambition: building something you couldn’t have sold two years ago. There’s nothing to measure that against, because it doesn’t exist yet. So the only thing that can tell you whether an idea is the right one is a clear sense of what this company is for, and what you want it to become. Without that, every proposal sounds reasonable, and the one that wins is the latest or the loudest.

Treat the capabilities like new colleagues

Think of these capabilities as collaborators on the team rather than as tools you buy. We agree with Neil Redding (opens in a new tab), who speaks about AI and leadership and makes this point eloquently in his talks. A tool is something you configure once and then forget. A collaborator is someone (or something!) you brief, challenge, and reason with.

Try it for a moment. When a senior hire starts on Monday, you don’t hand them a laptop and hope. You tell them what the company stands for, who it serves, and what it won’t do, and then they make a hundred small calls you never see, mostly correctly.

Your AI collaborators are making those same calls, just faster and with nobody watching. They work from whatever brief they were given, and when nobody has written one down, their brief is whatever the person at the keyboard assumed, or happened to remember, that morning.

If you skip the question

Everyone answers it anyway, privately and differently. The product lead has one answer, the COO has another, and the vendor in the room has a third that happens to match what they sell. A few months later the work shows it.

• A dozen pilots, each one defensible, but none of them pointing in the same direction.
• Every initiative reports success, and yet the P&L barely moves.
• Two teams built overlapping capabilities and nobody can say which was right.
• Ask ten people what the company is for and four companies come back.

Nobody in that picture is doing a bad job. Each team filled in the missing answer for themselves, sensibly enough, and built for the company they had in mind. You end up funding four companies at once, and none of them gets far enough to matter. That’s what a positioning problem looks like from the inside: a company that never settled what business it’s in, and shows it in everything it builds.

Worse, nothing gets canceled. Every initiative is defensible on its own terms, and a defensible project doesn’t die, it gets renewed. The portfolio only grows, the spending compounds, and your position in the market doesn’t.

At least you will find out sooner than you would have five years ago. Running twelve initiatives at once is cheap to start, so what used to take three years to surface takes two quarters. AI changed the clock, but without direction you just end up nowhere, faster. The long version

What we are not

We don’t build AI, sell it, or audit your data.

What we do is settle what the company is for, in writing, specific enough to be wrong. Then the people building have something to build against, and so do the collaborators they’re building with.

That earns its keep in the practical decisions. In the months of follow-through that come with either engagement, when you’re down to three vendors or two architectures, we’re in the room asking which one gets you to the company you said you were becoming. It’s a question about you, not the model, and it’s usually the one nobody in the room is assigned to ask.

If you already have your North Star, written down and shared, and the initiatives still don’t add up, your problem is somewhere else. Usually it’s incentives that quietly reward something other than the destination, teams without the capacity to build what you asked for, or data that won’t support it. While we don’t fix those, we will tell you what we think you’re dealing with, and where we can, point you to people who do that work.