Investigation · RESEARCH PRACTICE
I Stopped Trying to Be Right and Started Trying to Be Early. The Work Started Landing.
5 min read · June 2026
Thesis
Research does not change decisions by being right. It changes them by arriving before the decision is locked, and by saying one clear thing.
The Question
I will say the thing analysts are not supposed to say out loud. Most of the research I have produced or watched get produced changed no decision at all. It was commissioned, delivered, complimented, and shelved, often by people who had already decided and wanted cover. For a long time my response was to make the next study more rigorous, which did nothing, because rigour was never the problem. So I went back through my own work and sorted it into the pile that changed a decision and the pile that did not, and looked for what the first pile actually had in common.
The Hypothesis
My comfortable assumption was that the studies that landed were the best ones. I no longer believe that. The hypothesis I ended up with: decision-changing research has almost nothing to do with how clever the method was, and almost everything to do with two things, getting there before the decision set, and having the nerve to say one unambiguous thing instead of a balanced menu.
How I Looked At It
Own work, sanitised, sorted into two piles by a single test: did someone demonstrably do something different because of it. Small sample, my own memory, all the bias that comes with grading your own homework. I am treating this as an observation, not a proof.
What I Found
A supplier-risk deep dive I did years ago is the one I keep coming back to. The client was leaning toward a single source in a region I had found reasons to worry about. The analysis was not exotic. What mattered was that it arrived two weeks before the contract decision, not two weeks after, and that it ended with a position, dual-source and accept the cost, rather than "here are the trade-offs." It changed the call. When that region wobbled later, the client had a second source and kept shipping, and the cost of dual-sourcing turned out to be the cheapest insurance they bought that year. That study also won the firm's top internal review that cycle, which I mention only because the recognition came from the recommendation being usable, not from the depth of the work.
The competitive positioning work I did later sat at the other end and still landed, for a different reason. It fed how a set of capabilities got represented in an external analyst evaluation. It worked because someone senior already half-suspected what the data confirmed, and the research gave them the spine to act on a suspicion they could not previously defend in a room.
The pile that changed nothing is the instructive one, and it cost more than the wins earned. The common feature was not weak analysis. It was good analysis that arrived after the decision had set, or that ended in a careful three-option summary that let everyone read their existing preference into it. A balanced deck is a comfortable deck and a useless one. I produced plenty before I understood the bill.
What It Means For Business
If you are buying research, rigour is table stakes and timing is the actual variable, so the first question on any brief should not be "how deep" but "what date is the decision, and can we beat it." If your researcher keeps handing you even-handed menus, you are paying for the appearance of insight and making the call yourself anyway. The way I run it now: I find out when the decision lands before I start, I deliver before it, and I take a position I am willing to be wrong about out loud. That last part loses you the occasional argument and wins you the only thing that matters, which is a decision that moved because you were in the room in time.
The Decision
What I’d Test With More Time
I would want to track research projects from the start with one field filled in, the date the decision will be made. My quiet suspicion is that delivery date minus decision date predicts impact better than any quality score we put on the work itself. I have started treating that as a small model of its own. I have not proven it. I would like to.