How much UX research is enough?

Worry about research leads teams to two common failures, and both are expensive. The first team skips research entirely, because the roadmap is urgent and everyone already "knows the user". They build the wrong thing with impressive efficiency. The second team uses research to delay: another survey, another round, another report, because research feels like progress and postpones the difficult moment of committing. They build the right thing a year late, or never.

Both teams asked "how much research is normal". The useful question is "what decision is this research for", because the decision sets the budget.

Size research to the decision

Take the decisions in front of you and sort them by the cost of being wrong.

Cheap to reverse: label wording, the order of menu items, a color, an empty-state message. Release your best guess or run the fastest possible check. Anything more is delay that looks like work.

Expensive to reverse: the core workflow, the product's mental model, the pricing screen, the onboarding your activation numbers depend on. These decisions become fixed in code, integrations, and user habits, and reversing them later costs sprints. Here real research pays: structured interviews with the people who will use it, observation of how they do the job today, and a prototype tested before the build starts.

The awkward middle: most decisions are between these two ends, and the honest step is to ask what evidence would actually change your mind. If no realistic finding would alter the plan, the research is only for show. Skip it and spend the time on a decision where evidence could still change the result.

This risk-scaled approach is not ours alone. The GOV.UK Service Manual, one of the more disciplined public documents on building services, treats user research as a continuous activity that runs through all of delivery rather than a phase with a completion date. The same approach works for commercial products.

The five-user rule, and what it does not cover

For usability testing, the classic finding is Jakob Nielsen's: about five users find roughly 85% of the usability problems in a design. The implication is that testing is cheap enough to repeat: run five, fix what they found, run five more on the fixed version. Three small rounds work better than one large study at the same cost, because each round tests a better design.

Know what the rule does not cover. It applies to finding usability problems: where people get stuck, confused, or lost. It does not size surveys, validate market demand, or measure whether a redesign changed a metric; those are measurement questions and need real sample sizes. The mistake to avoid is using five enthusiastic reactions as proof anyone wants the product. Five users can tell you your checkout is confusing. They cannot tell you whether anyone wants to check out.

Discovery research vs validation research

Two different activities are both called "research", and teams that confuse them buy the wrong one.

Discovery research asks what should exist: who has the problem, how they solve it today, what they would change. Its methods are interviews and observation, and its output is direction. Skipping it is how well-built products end up with no users. Our post on knowing what to build argues this is now the decision with the greatest effect in software; discovery research is that decision's evidence.

Validation research asks whether this design works: can the intended people complete the intended task. Its method is watching users attempt tasks with a prototype, and its output is a fix list. Skipping it is how correct ideas fail on execution.

The order matters. Validating a design for a product nobody wants is improving the wrong thing. On a custom clothing configurator, the discovery insight, that customers start from the day they are picturing rather than from fabric codes, has to come first. Testing screens built around fabric codes would optimize a flow that should not exist.

What to skip

Some research is reliably not worth buying. Surveys asking users what features they want: people are poor predictors of their own behavior, and feature votes measure which words people know rather than what they need. Focus groups for interface decisions: people in groups perform for each other and do not show their real behavior. Large studies of things a prototype would settle faster. And any research whose findings arrive after the decision has already been released, which in fast-moving teams is more research than anyone admits.

The test for all of it: name the decision, name the evidence that would change it, and check the timing. Research that cannot change a decision is a cost that only makes the team feel better.

Who should watch the sessions

One practical rule increases the value of whatever research you buy: the people who make product decisions attend the sessions. The live sessions, at least a few of them, rather than a summary or a highlights video. A founder who has watched three users fail the same task stops needing to be persuaded by the report, and the dozens of small product decisions they make afterward are better informed. Research that only the researcher experiences rarely changes decisions, because the organization receives conclusions without the belief that comes from watching. If you pay someone else for the research, write attendance into the plan.

After launch: small amounts, all the time

The research that gets skipped most is the cheapest: staying in contact after launch. A short watched session on a live feature each month. Reading the how-do-I tickets as design data rather than support volume. A quick round of five users before each significant release. This is how you catch the problems no upfront plan could predict, while they are still small: the same signals catalogued in signs your product has a UX problem, caught before they grow.

If you are hiring the research rather than running it, how a partner sizes it is itself a signal, and it is one of the tests in how to choose a product design agency: a partner who proposes the same research phase regardless of your decisions is selling a fixed package instead of answering your question.

Enough research is the amount that makes your next expensive decision an informed one. It is almost always less than the anxious teams buy, and more than the confident teams do.