Start with the decision, not the data
Before commissioning research, write down the decision it needs to inform. “Understand our customers better” is too broad to guide a useful search. “Decide which onboarding step to simplify first” points to specific evidence: where people stop, what they say is confusing, and which issues affect the audience you serve. A clear decision also helps you avoid collecting information that is interesting but unlikely to change what you do.
Next, list the questions you must answer and the level of confidence the decision requires. A small, reversible change may only need enough evidence to choose a sensible first test. A costly or difficult-to-reverse change warrants closer scrutiny. This distinction keeps research effort proportionate to the stakes.
Look for evidence that already exists
Useful material may be scattered across an organization: past surveys, customer interview notes, support requests, product analytics, sales conversations, search queries, returns, or feedback forms. External material, such as publicly available reports, may provide context, though its categories and methods may not match yours. Make a short inventory showing what each source covers, when it was collected, who it represents, and what question it might help answer.
Existing data can be especially valuable when the question concerns observed behavior. If you want to know which part of a sign-up flow loses the most visitors, available usage data may be a better starting point than asking people to recall the process. Conversely, behavior alone may not explain why someone left. Reviewing what you have can reveal both the pattern and the point where a targeted interview would add value.
Check whether the evidence fits the question
Data is not automatically relevant because it is available. Ask whether it describes the right people, product, market, and time period. Feedback from frequent customers may not represent first-time buyers. A survey about general satisfaction may not answer a question about one checkout step. Older findings can still be useful for understanding persistent needs, but they deserve caution when the product or audience has changed.
Also examine how the information was produced. Who had a chance to respond, and who did not? Were survey questions leading or answer choices incomplete? Do analytics record the action you care about consistently? Are interview notes detailed enough to distinguish a participant’s words from an interviewer’s interpretation? You do not need perfect data to learn something, but you do need to know its limits before treating a pattern as a conclusion.
Compare sources when possible. Repeated complaints in support records, a drop-off at the corresponding product step, and similar comments in interviews offer a stronger basis for investigation than any one source alone. Disagreement is useful too: it may point to different audience groups, a measurement problem, or a question that existing evidence cannot settle.
Decide whether a gap justifies new research
Create a simple evidence map with three labels for each decision question: sufficiently answered, tentatively answered, or unanswered. Note the reason for each label. For example, you might know that new users abandon a setup step but have only tentative ideas about why. That makes the next task clear: investigate the cause, rather than launch a broad survey that repeats what the usage data already shows.
New research is most useful when an important question remains unanswered, existing sources exclude a crucial audience, or available evidence is too old or weak for the decision at hand. Choose a method that addresses the gap. Interviews can explore motivations and language; a carefully designed survey can help assess how widely a clearly defined view is held among the people reached; usability sessions can reveal where a task becomes difficult. No single method resolves every uncertainty.
Turn the review into a practical next step
End the review with a brief decision note. State what you know, what you are assuming, what remains uncertain, and what action follows. That action might be to proceed, test a modest change, analyze an existing dataset more carefully, or commission a focused piece of research. Include the limitations so a later reader does not mistake a working assumption for an established finding.
Use existing records responsibly. Limit access to customer information, avoid sharing identifying details unnecessarily, and follow the permissions and retention rules that apply to the data. An evidence review should make decisions clearer, not create a reason to reuse personal information without care. The goal is not to avoid new research at all costs; it is to ask new questions only after you understand which answers are already within reach.
