When Research Decides What Gets Built
The cliché is that UX research polishes the screens after the decisions are made. The most useful study I ever ran did the opposite — it decided what to build. Here's the line from a 2018 thesis to a product's core navigation model.
The cliché about UX research is that it happens at the end. The product is decided, the screens are built, and research shows up to sand the corners — move a button, rename a label, confirm that people can complete the task someone already chose for them. Useful. Also the least interesting thing research can do.
The most useful study I ever ran did the opposite. It decided what to build.
I came to research the long way: physics, then communication design, then a consulting venture that failed the way things fail when you have the right ideas and no ability to make anyone act on them. That failure taught me the thing no method could: a finding is worthless unless it answers a decision someone actually has to make. So I went back for a master's in usability engineering, then worked at research agencies — eresult, then field research at trivago — before going independent. Somewhere in there I learned the difference between research that informs a decision and research that merely decorates one.
A lot of that difference is method. The user-requirements approach I trained and certified in — — moves through four information types in order: context of use, user needs, user requirements, and only then solution ideas. The discipline isn't the vocabulary. It's the wall between the problem space and the solution space. The most common way research gets wasted is the solution-space leak: the study is nominally about understanding a need, but everyone in the room is already three features deep, and the "findings" just ratify a solution chosen before anyone talked to a user. Keep the problem space clean — refuse to name a feature while you're still describing a need — and research stays capable of surprising you. That capacity to surprise is the value. Research that can only confirm was never research.
The other thing that separates real research from theater is knowing how much is enough, and here the field trips over its own folklore. gets quoted like scripture and misapplied constantly. It comes from a specific model of usability testing and says nothing about interviews. Qualitative interviews have no golden number; you recruit toward saturation — usually starting around five or six and continuing until new sessions stop telling you anything new. are a different regime entirely — roughly forty participants before your confidence intervals are worth quoting. And incentives aren't just grease to fill a schedule: pay too little and you bias the sample toward people with time to spare, which makes participant pay an equity lever as much as a recruiting one. None of this is exotic. It's the difference between a number you can defend and a number you borrowed from a headline.
Which is the setup for the study that decided what to build.
In 2018, for my master's thesis, I ran a small refinding study — watching people retrieve things from their own digital lives — around a prototype called . It scored a 77.5 on the System Usability Scale, above the 68 average. But the score wasn't the finding. The finding was structural: people don't have a single preferred way back to their own data. The same person re-finds one thing by time (scrolling back to roughly when), another by topic (the project, the folder, the kind of thing it was), and another by people (who was there, who sent it). No route dominated. Ask which is best and the honest answer is that the question is wrong.
That is not a usability finding. It's a product decision waiting to happen.
Because every tool that privileges one route makes the others a workaround — folders privilege topic, camera rolls privilege time, chat apps privilege people — and the study said, with data, that privileging any single route is a mistake. So when that research turned into a product, the finding became the architecture. Timeline OS is built around three first-class axes — time, topic, people — because a study said people need all three, not because three axes seemed like a tidy idea. The research didn't polish the screens. It chose the model the screens are built on.
That's the version of research worth defending. Not the box you check before launch — the study you run early enough, and clean enough, that it can still change the answer. Keep the problem space separate from the solution space so a finding can surprise you. Recruit to the standard the claim actually requires, not the one a headline made famous. Do those two things and research stops being the function that confirms decisions and becomes the one that makes them. The screens come later. What gets built is the part research was always for.