I have sat in two kinds of design review. In one, a screen is defended by how it feels. In the other, a screen is defended by a number. Both rooms produce weak work, for the same reason: one input is being asked to carry a decision it cannot carry alone.
The split is old and the usual names for it are tired, so here is the version I use. Research is the part of the job that can be wrong in public. Taste is the part that can only be argued. A designer with the first and not the second ships screens nobody objects to and nobody remembers. One with the second and not the first ships beautiful answers to a problem the business did not have.
What research can settle.
Research answers questions of the form does this happen. Where do people stop. Which word in a label reads as a warning. Those questions have answers, the answers repeat across people, and you can be shown to be wrong about them, which is what makes them worth the fortnight.
On ADACAP the finding that mattered was not a preference. Everyone I studied opened the app and then checked it against their bank. Nobody said they distrusted the figures. They simply did not act until the two agreed. That is behaviour, it repeated, and it moved reconciliation from a settings page into the centre of the product. Taste would never have found it, because taste works on what you can see and this was something people did with their hands.
What it cannot settle.
Research will not choose between three designs that all work. It will not tell you a screen is one section too long, or that a figure wants to be larger than the word beside it, or that a page has said the same thing twice in two different fonts. Put five workable layouts in front of a study and you will learn that all five are workable.
Ask people which they prefer and they will answer, politely and at length, and the answer is worth very little. Preference given on the spot is a guess about a future self. Behaviour under a real task is evidence. The skill is knowing which one you just collected.
The gap left over is where taste lives. Taste is an accumulated sense of how a thing should feel to use, built out of thousands of hours of looking, and calling it subjective is mostly a way of dodging the argument.
Research narrows the field to the designs that could work. It never picks the one you ship.
Keeping both in the room.
Two habits do most of the work. First, write every finding as a sentence someone could disagree with. Users want clarity cannot be disagreed with, so it decides nothing. Managers checked the balance against their bank before acting on it can be, and it has consequences you can build from.
Second, whenever you make a call on taste, say what it serves. Not this feels cleaner, but this puts the figure people came for above the fold and pushes the breakdown down a screen. Now it is a claim about the product rather than about you, and a colleague can tell you it is wrong.
Both halves on one project.
SPARQ prepared bank executives for board meetings, which was costing them hours every month. Research gave the shape of the problem and one hard constraint: nobody would repeat a number to a board without seeing where it came from. That constraint was not a preference and it was not negotiable.
What followed was a judgement. One tap from any sentence the AI wrote to the passage it came from, always, on every screen. It cost layout space and engineering time, no one asked for it in those words, and it is the reason the thing got used. Pilot managers reported preparation down 72%. The research made the constraint visible. Taste decided what to build in answer to it.
When there is no time for either.
Most briefs arrive without two weeks for interviews. The fallback is to write down what you are assuming, in that same disagreeable form, where the team can see it. On a small feature I will list four assumptions before opening Figma. Two are usually safe. One is normally the entire risk of the project, and naming it out loud is often enough to buy half a day of somebody's time to check it.
Neither half is the senior part of the job. What changes with experience is how fast you notice which kind of question is on the table, and how quickly you stop the room from answering it with the wrong one.
