Ask a business what it's building and you'll hear the middles: a data lake, a reporting layer, an integration programme, an AI initiative. Ask why often enough and every one of those answers ends in the same place — so that, someday, people can make better decisions.
The decision is the point. And yet it's the one thing that's always left for the end — after the cleanup, after the migration, after the rollout. It gets treated as the reward for finishing everything else.
We think that's exactly backwards. The decision isn't the reward at the end of the work. It's the specification for the work.
A decision is a specification
Pick one real decision. Not a theme, not a "use case" — a decision: a critical supplier raises prices; what do we do? The moment you name it, it starts issuing instructions.
- It names its information. Spend exposure, the contract clause, the qualified alternates, inventory cover. Six numbers — not six systems.
- It names its choices. Negotiate, move volume, qualify an alternative, accept. Not "insights" — options someone can actually take.
- It names its rules. The judgment your best people already apply, finally said out loud.
- It names its actions. Who sends the counter. Who places the orders. By when.
- It names its outcome. Landed cost against plan, by quarter end. Success has a definition before the work starts.
Nothing on that list is speculative infrastructure. Every piece has a job on the day it ships, because the decision demanded it into existence.
Start anywhere else, and watch what happens
Start with the data, and you'll clean columns no decision will ever read. Start with the platform, and you'll integrate systems no choice ever consults. Start with a "use case" — the sneakiest one, because it sounds decision-shaped — and you'll usually find it's a description of activity: the reporting around the decision gets automated, while the decision itself stays right where it was. In the meeting.
One at a time is the feature
Decision-first also changes the size of the first step, and this matters more than it sounds. You don't need estate-wide readiness to take one decision end to end. One decision is small enough to be live in weeks and consequential enough to prove something real.
And the second decision builds faster than the first — because the information plumbing, the rule-writing habit, and the outcome loop already exist. That compounding is the quiet payoff of starting narrow. (More on that in How Business Judgment Becomes Reusable.)
So the useful question isn't "what's our data strategy?" It's smaller, sharper, and harder to hide from: what decision would you start with?