Generative vs evaluative research
- Generative research discovers problems worth solving; evaluative research tests whether a solution solves one.
- Running an evaluative method on a generative question (or vice versa) is the most common research mistake.
- Discovery is continuous, not a phase — the cheapest insight is the one that arrives before the roadmap is set.
The most useful split in research is not by method but by intent.
Generative (discovery) research happens before solutions exist. Its output is problems, needs, opportunities and mental models — the raw material for strategy. Methods: deep interviews, field observation, diary studies, competitive analysis, secondary research. The deliverable is a clearer problem, not a validated screen.
Evaluative research happens once something exists — a sketch, prototype or live feature. Its output is evidence about whether the thing works: usability tests, first-click tests, surveys against benchmarks, A/B tests.
The classic failure mode is answering a generative question with an evaluative method: “we tested the prototype and people liked it” says nothing about whether the underlying problem matters to anyone. The reverse failure — endless discovery with no evaluation — burns quarters without shipping learning.
Teresa Torres’s framing is useful: discovery is a continuous habit, not a project phase. Small weekly touchpoints with users (one interview, one test) beat a heroic research sprint twice a year, because the cost of changing course is lowest before decisions harden.
Try it: classify the last three research activities your team did (or plans). Label each generative or evaluative, and check the label matches what the team actually wanted to learn.
- Hall — Just Enough Research, ch. 2
- Torres — Continuous Discovery Habits
- NN/g — Discovery: Definition & Methods