Research creates value when it changes the quality of a decision. A large report can fail this test. A small, well-designed investigation can pass it.
Before commissioning a survey or asking AI to summarize a market, write the decision in one sentence. “Should we enter regional food manufacturing with a standardized maintenance offer?” is useful. “Understand the market better” leaves the team without boundaries.
Define what would change your mind
List the assumptions behind the proposed action. There must be a meaningful problem, an accessible buyer, a credible offer, sufficient demand and acceptable economics.
Then identify the assumption most capable of reversing the decision. If service travel costs make small accounts unprofitable, a general brand-awareness study is secondary. Research route density, service frequency and willingness to accept remote support.
Set decision criteria before seeing the findings. Otherwise, teams can reinterpret weak evidence to defend an attractive idea.
Combine sources deliberately
- Internal information reveals existing behavior: enquiries, returns, lost orders, service calls and margin by customer type.
- External secondary sources describe market structure and broader change.
- Primary research investigates questions your existing data cannot answer.
Each source has limits. CRM data reflects whom you already pursue. Industry reports may use incompatible category definitions. Interviews explain experiences but rarely establish market prevalence. Website behavior shows what visitors did, with limited insight into who never visited.
Triangulate rather than searching for one authoritative number. If sources disagree, investigate the definition, geography, period and sample before averaging them.
Use interviews to understand mechanisms
Ask customers to reconstruct a recent decision. What triggered the search? Which alternatives were considered? Who objected? What evidence mattered? What happened after implementation?
Avoid asking only whether people like your concept. Approval is inexpensive. A detailed account of past behavior is usually more informative than a general statement of future intention.
Recruit beyond enthusiastic customers. Include lost accounts, non-buyers, frontline users and people who struggle with your current service. Document how recruitment limits the conclusions.
Treat AI as an assistant to evidence
AI can help organize interview notes, compare themes and draft questions. It can also invent references, flatten contradictions and reproduce assumptions embedded in its input.
Verify quotations against original records and retain the connection between a finding and its source. Use approved handling for confidential customer material. An invented customer persona or synthetic interview does not establish that real customers hold a particular view.
In an AI-assisted research process, record where AI contributed and where a human checked the output. Efficiency should preserve traceability.
Convert learning into a decision memo
A useful memo distinguishes observations, interpretation and recommendation. Include the strongest contrary evidence, the uncertainty that remains and the smallest next test.
Consider a hypothetical service business finding strong interest in a premium response package. Interviews reveal that buyers value it, but operational analysis shows insufficient local coverage. The correct response may be a restricted regional pilot, not a national launch.
Assign an owner and a decision date. Store the result where future teams can find it, including the assumptions that later proved wrong. Research then becomes organizational learning instead of a sequence of disconnected projects.