Article · Decision Intelligence series
Decision Intelligence — why most organisations have data, not decisions
Every organisation can produce a dashboard. Almost none can produce the reasoning behind their last big call.
Ask a team to justify a decision made six months ago — a pricing change, a market entry, a hire, a cancelled project — and watch what happens. The outcome is remembered. The result is on a slide somewhere. But the actual reasoning — what alternatives were on the table, what was assumed, what would have changed the answer — has usually evaporated. What’s left is a story reconstructed backwards to fit whatever happened, which is a different thing entirely, and a worse one.
That gap is not a data problem. Most organisations have more data than they know what to do with. It is a decision problem — the space between having information and having chosen something on purpose, for reasons anyone could inspect later.
An organisation can be information-rich and decision-poor at the same time. Most are.
§ 1
Information is not a decision
A dashboard tells you what happened. A forecast tells you what might happen. Neither one, by itself, chooses anything. A decision is the moment someone takes a position under uncertainty — commits to one path over the others that were genuinely available — and that moment is where most organisations quietly go informal.
The analysis is rigorous. The choice is a feeling, arrived at in a room, dressed up afterwards in the language of the analysis that preceded it. Nobody is being dishonest. It’s simply that the discipline most companies apply to gathering information has no equivalent for reasoning with it. Decision Intelligence is the name for building that missing half.
§ 2
What a decision worth trusting actually requires
Four properties, and they have to hold together — a decision can satisfy three of these and still not be one anyone should trust.
01
Framed before it's argued
The problem is stated in its own right, before anyone proposes an answer. What is actually being decided, what would count as success, what constraints are real versus assumed. Skip this step and the debate is really about which pre-formed answer wins, not about which problem is genuinely on the table.
02
Reasoned in the open
The logic is visible, not just the conclusion. Which alternatives were considered and rejected, and why. Which assumptions the answer depends on. A recommendation without visible reasoning is not a decision — it's an instruction, and instructions don't improve with use.
03
Owned by a name
Someone specific is accountable for the call, and that is recorded at the time, not inferred afterwards from who happens to be in the room when it goes wrong. Diffuse ownership is the most common failure mode in organisational decision-making, and the easiest one to fix.
04
Checked against what actually happened
The reasoning gets revisited against the outcome — not to assign blame, but to find out whether the thinking was sound, independent of whether the result was good. Good decisions produce bad outcomes. Bad decisions produce good ones. An organisation that only tracks outcomes is training itself on the wrong signal.
§ 3
Where AI actually earns its place
The honest answer is: not at the point of choosing. The value AI brings to a decision is upstream of that — in widening what a person can see before they commit to anything.
Most decisions aren’t wrong because someone reasoned badly with the information in front of them. They’re wrong because the information in front of them was incomplete, the framing was too narrow, or the obvious few ways of thinking about the problem were the only ones anyone had time to reach for. That’s the part AI is genuinely good at closing: pulling in the context a person didn’t have time to assemble, surfacing a way of framing the problem nobody in the room had thought to try, running the second-order consequences of an option before anyone commits to it.
What it should not do is supply the choice. A system that hands over a recommendation with the reasoning hidden has just automated the exact failure mode Decision Intelligence exists to fix — a confident answer with no visible thinking behind it, except now it arrives faster and looks more authoritative for having come from a machine.
The measure of good AI-assisted decision-making isn’t how much of the decision it took off your hands. It’s how much better-reasoned the decision was when it stayed in them.
§ 4
Where this is weakest
This framing has a limit, and it’s worth stating plainly: not every decision is worth this much structure. A five-minute call with low stakes and easy reversal doesn’t need a framed problem statement and a named owner — applying that much ceremony everywhere is how good frameworks die, smothered by their own overhead.
The judgment call is knowing which decisions are which. That itself deserves the same rigour it’s describing, and I don’t think there’s a clean formula for it — only a rough rule: the more expensive a decision is to reverse, the more this discipline earns its cost.
§ 5
A test worth running on your last big decision
Four questions, and most organisations find at least one produces a shrug.
Find the problem statement. Not the answer — the problem, written down, before anyone proposed a fix. Does it exist anywhere, or did the conversation start with a recommendation?
Find the alternatives. What else was genuinely considered, and why was it set aside? If the honest answer is “nothing else was seriously on the table,” the decision was a formality, not a choice.
Find the owner. Whose name was on the call at the time — not who gets asked about it now that the outcome is known?
Find the reasoning, a year later. Could someone who wasn't in the room reconstruct why this made sense at the time, using only what was written down?
If more than one of these comes up empty, the organisation isn’t short on data. It’s short on decisions — on the discipline of turning information into a choice someone can stand behind, and that someone else can learn from.