The manifesto

Companies don't have a data problem. They have a decision problem.

Why the next wave of AI won't save organizations that still can't decide — and what to build instead.

I've worked with data for more than thirteen years. In a research lab, at IBM when Big Data was just emerging, in agriculture, at an Aon-owned insurtech, and today at an education group growing by acquisition. Sectors with nothing in common.

And in every one of them I ran into the same problem, always disguised as something else.

The scene repeats. Dashboards everywhere. A busy data team. Reports nobody argues with… until a real decision has to be made. Then everyone falls back on gut, on their trusty spreadsheet, or on whoever talks loudest in the room. The company has more data than ever and still decides as badly as it did ten years ago.

When that happens, almost everyone reaches the same wrong conclusion: “we’re missing data.” So they buy more. Another tool. Another dashboard. Another data lake. And now, of course, AI.

But the problem was never a lack of data. The problem is that the chain from data to decision is broken — and no product you buy fixes that for you.

What's actually broken

When you look closely at an organization that's “good with data but bad at deciding,” you almost always find the same things:

Nobody agrees on the metrics. Every department calculates the same indicator differently, so meetings are spent arguing over whose number is right instead of what to do about it.

There's no shared language. “Customer,” “active,” “margin” or “enrollment” mean different things depending on who's asking. Without shared semantics, every query is a negotiation.

And as a result, nobody fully trusts the platform. And if you don't trust the number, you decide on instinct. Again.

The data was there. The decision wasn't.

The reporting factory

There's a tell that gives these organizations away: the data team lives fighting fires. Tickets, requests, “pull me this spreadsheet by tomorrow.” It becomes a reporting factory.

And here I want to be blunt, because it's one of the things I care about most: a data team that only answers tickets isn't badly managed for lack of effort. It's badly designed.

You don't fix it with more people or more dashboards. You fix it by changing the operating model: moving from a service that fields requests to a partner that co-creates decisions with the business. It's a problem of organizational design, not technical capacity.

The chain that does work

After seeing the same pattern across such different contexts, I stopped treating it as a series of projects and started treating it as a single chain. Build it whole and in order, and the organization decides well. Miss a link, and everything below it wobbles:

Strategy — what the business is actually trying to win, stated clearly enough to build on.
Metrics — the few numbers that define “winning,” governed and shared, beyond dispute.
Semantic models — one language for the business, so every question means the same thing to everyone.
Data platform — reliable infrastructure that people and AI agents alike can build on.
AI — applied where it changes the decision, not where it looks good in a demo.
Impact — decisions made, actions taken, results that show up in the business.

Technology enables this chain. But it doesn't create it. People create it — operating models, and the way an organization decides. That's the part you can't buy.

And then AI arrives

This is where I want to stop, because it's the mistake I'll watch repeat over the next few years.

AI doesn't fix decisions. It accelerates whatever's underneath. If the chain below works, AI multiplies good decisions. If the chain is broken, AI takes you to the wrong answer faster, at greater scale, and with more confidence. Pilots don't fail for lack of a model; they fail because they rest on foundations that never held up a human decision, let alone an automated one.

And here's something almost nobody is saying: AI agents need exactly the same truth people do. The same governed metrics, the same semantic language, the same reliable platform. An agent wired to ambiguous data isn't autonomous — it's an intern in a great hurry with no judgment.

That's why “being AI-ready” isn't a product you install. It's an organizational capability you build — and it turns out to be the same chain above, actually working.

The shift that matters

For a decade we've framed data as a technology matter. It isn't. It's a matter of how an organization decides.

The real work isn't standing up platforms or deploying models. It's designing the organization — human and technical — that can turn strategy into decisions, and decisions into action. The platform and the AI are just two links in that chain.

The coming years won't be won by the companies with the most data, or the flashiest AI. They'll be won by the ones that can decide and act — with judgment, fast, and at scale.

That capability doesn't appear on its own. It's designed.

That's the work. And it's the best I know.