What is Iterative Research?

Iterative research is a cycle of studying, changing, and studying again: a small round of research (often five to eight participants in a usability test, or a handful of interviews) surfaces problems and questions, the design or the hypothesis changes in response, and a further round tests whether the change worked and what it revealed next. It is the empirical engine inside user-centred design, the evaluative half of design thinking's prototype-and-test loop, and the practical form of a much older principle: knowledge advances by conjecture and correction, and the faster the corrections arrive, the further you get.

Why Iteration Beats Scale

The classic argument, made by Jakob Nielsen in the 1990s and still sound, rests on how usability problems are found. Each participant in a test reveals a substantial share of the problems present; additional participants reveal progressively fewer new ones; and, crucially, a study of twenty people finds the same big problems as a study of five, only with more repetition, whereas fixing the five-person study's problems and testing again finds a new layer of problems that were hidden behind the first. Three rounds of five therefore discover more than one round of fifteen, and cost the same. The argument applies to design problems (which live in the design and recur for everyone) and not to prevalence questions (how many users have which need), which still require representative samples; the logic of analytic generalisation underwrites the whole approach. A second benefit is corrective: a single large study freezes the questions on day one, while iteration lets each round's findings sharpen the next round's questions, the collect-analyse-redirect loop that qualitative methodology formalised long ago.

Running Cycles With Rigour

1. Keep the core constant, vary the fix.
Same tasks, same screener, same success rubric across rounds, so that improvement can be seen; change the design between rounds, not the measurement. The round-over-round metrics then form a genuine mini-trend.

2. Fix before you re-test.
A round whose findings weren't acted on is a repeat, not an iteration. The cadence should match the team's ability to change the artefact between rounds; for prototypes that can be days.

3. Log the changes and the reasons.
An audit trail of what each round found and what changed in response is the evidence that iteration converged rather than wandered, and the protection against fixing one problem by creating another unnoticed.

4. Watch for regression.
Each round should re-check the previously fixed problems, not only hunt new ones; fixes have side effects.

5. Make each round cheap enough to actually run.
Iteration dies of friction. Recruitment, fielding, and analysis that take a week per round produce a quarterly cycle labelled iterative; unmoderated prototype studies with a screened panel (a Ballpark study can turn a round in a day) are what let "test again tomorrow" be literal.

6. Know when to stop.
Rounds end when the measured goals are met or when new rounds stop finding problems worth fixing, the practical form of saturation. Iteration without a stopping criterion is a habit, not a method.

The Limits

Iteration optimises locally: a sequence of small improvements can polish a design that should have been rethought, which is why exploratory and generative research belongs before the loop and periodically alongside it (the parallel-design practice of testing several distinct concepts before iterating on one addresses the same risk). It also answers design questions, not prevalence or causal ones; small iterative rounds can't say how common a need is or whether a change moved a business metric, which remain jobs for surveys and experiments. And it depends on the artefact being changeable between rounds; where the thing tested is fixed (a shipped product with a long release cycle), iteration slows to the release cadence unless prototypes stand in.

The Takeaway

Iterative research trades one big study for many small ones, each testing the last round's fixes and sharpening the next round's questions. Keep the measurement constant, change the design, log everything, re-check the fixed, keep rounds cheap, and stop when the evidence says so. For design problems it finds more, faster, than scale ever will; for prevalence and causation, it hands off to the methods built for them.