Research that ends in a decision
There's a graveyard inside most organisations, and it's full of good research. Beautifully run studies, clean samples, genuine findings — all of it embalmed in a slide deck that someone presented once, everyone nodded at, and nobody ever acted on. The research wasn't wrong. It just never connected to a decision. And insight that doesn't change a decision is, economically, indistinguishable from no insight at all.
If a study can't name the decision it changes, it isn't research. It's expensive curiosity.
I've spent seventeen years on the insight side of this — running forty-plus research sprints to surface demand and PMF signals, building customer-understanding engines, leading insight functions from instinct-led to data-driven. The single biggest lever I've found has nothing to do with method sophistication. It's a sequencing discipline: design the research backwards from the decision, never forwards from the question.
Forwards research vs backwards research
Most research is designed forwards. Someone has a question — "what do customers think of our brand?" — and commissions a study to answer it. The study answers it. And then everyone discovers the answer doesn't map to anything they can do, because "what do customers think" was never a decision. It was curiosity wearing a lab coat.
Backwards research starts at the other end. You name the decision first — "should we reposition the brand for a younger segment, yes or no, by Q3" — and then ask what you'd need to know to make that call with confidence. Every question in the study now earns its place by mapping to the decision. Anything that doesn't, gets cut. The output isn't a description of reality; it's the specific evidence required to choose.
Designing backwards, in practice
1. Write the decision and its owner first
Before any fieldwork, get one sentence on paper: the decision, the person who'll make it, and the date. "Marketing will decide whether to launch the value tier by March." Now every finding has a home. A study without a named decision-owner is a study nobody is obligated to act on — which is why nobody does.
2. Pre-commit to what each result would mean
This is the discipline most teams skip. Before you see the data, agree: "if it comes back above X, we do this; below X, we do that." Pre-committing to the action for each outcome does two things. It exposes the studies that wouldn't change your behaviour either way — kill those, they're theatre. And it removes the wiggle room that lets uncomfortable findings get reinterpreted into inaction after the fact.
3. Build for speed where speed changes the decision
Not everything needs a flagship study. Much of the value comes from rapid sprints — fast, focused, good-enough reads that arrive while the decision is still live. I've pioneered quick-turn solutions in high-regulation categories precisely because a perfect answer delivered after the decision is made is worth nothing, and a 90%-confident answer delivered in time is worth everything. Match the rigour to the stakes, and the speed to the decision window.
A perfect answer after the decision is made is worth exactly nothing.
From study to engine
One-off studies, however well-designed, still depend on someone remembering to commission them. The bigger unlock is turning insight from a project into an engine — a standing capability that feeds decisions continuously. When I built customer-understanding engines, the point wasn't any single sprint. It was the operating cadence: a rhythm where insight flowed into commercial and product decisions as a matter of routine, not a special occasion.
An engine has three parts. A standing question bank tied to live decisions. A fast pipeline to answer them. And — the part that's usually missing — a forcing function that puts the answer in front of the decision-owner before they decide, not after. Build that loop and you stop asking "should we do research on this?" The research is already happening, pointed permanently at the choices that matter.
The measure of good research
Here's the only test I trust. Six months after a study, ask one question: what decision did this change? If there's a clear answer — we priced differently, we repositioned, we killed the feature — the research did its job. If the answer is "it gave us useful context," the research died in the deck, and you paid full price for a funeral.
Insight isn't valuable because it's true. Plenty of true things change nothing. Insight is valuable when it's load-bearing — when a real decision rests on it and would have gone differently without it. Design every study to be load-bearing, or don't run it. The graveyard is full enough.