Rules
Part of Startup Market Research: A Practical Guide
Market research turns solid once a reproducible public-data file anchors it
A reproducible public-data file, skeptic review, and recent-experience interviews turn startup market research into decisions a team can defend.
What to take away
- Research is a file you build, a set of people you recruit, and a skeptic who tries to break the conclusion before the team spends.
- Anchor every claim to a file someone else can reopensource, period, geography, units, and revision date.
- Recruit people who met the problem in the last ninety days, and ask what they did, not what they would do.
- Watch the work happen. What a buyer says in an interview and what the team does on a Tuesday rarely match.
- Run a skeptic pass before the memo circulates, not after the budget is committed.
How to research startup market research is not a reading exercise.
Build the public-data file first
Start with the Census Bureau's Census Business Builder, which maps demographic, socioeconomic, and business data by geography and industry. Pull the establishment counts for your metro and NAICS code.
What each figure carries
- Collection period
- Geography
- Classification
- Revision history
Then write down what the file is not. Every figure carries a collection period, a geography, a classification, and a revision history. A count from a five-year-old survey is a different object than a count from last quarter, even when both sit in the same dashboard.
Five columns and one row per figure keep the file reopenable.
Column / What goes in the cell
- Source
- The dataset and release, for example Census Business Builder.
- Period
- The collection window the figure covers.
- Geography
- The metro, county, or ZIP code described.
- Units
- Establishments, employees, dollars, or households.
- Revision date
- The day you last checked the source.
Recruit recent experience, not opinions
Screen for people who hit the problem in the last ninety days. Buyers, users, people who chose a competitor, and people who looked and walked away all belong in the sample.
Screening criteria:
Recruit recent experience
- Used or evaluated an alternative in the last ninety days.
- Can name the last time the problem cost them time or money.
- Will take a thirty-minute call and allow notes.
Recruit through industry groups, LinkedIn, panel vendors, or referrals from people who declined. Eight to twelve interviews per segment is a typical range. Stop when two calls in a row add no new blocker.
Script skeleton:
Recruit recent experience
- Walk me through the last time this came up.
- What did you use instead, and what did it cost?
- Who else was involved, and who signed?
- What almost stopped you?
Ask what happened last time. Which alternative did they use, what did it cost, who signed, what broke. Do not describe your product before they finish answering.
Observe the workflow with permission
Sit with the team and watch the actual sequence: the tools open, the handoffs, the waiting, the workaround someone built in a spreadsheet. Compare that against what the interviews claimed.
Collect only what the decision needs. Personal and confidential material stays out of the file unless the decision genuinely turns on it.
Test a real commitment
Offer something truthful and proportionate: a qualified meeting, a manual assessment, a paid pilot, a deposit. Predefine the success threshold and the stop rule before you open the offer.
Record why people refused. Segment, timing, channel, and trust each produce a different refusal. Offer and price also produce a different refusal. Each points at a different fix. Startup customer acquisition cost by channel is the variable this test measures before budget moves.
Worked example: a $40,000 channel decision
A seed-stage team selling scheduling software to dental groups wanted to spend $40,000 on paid search. The public-data file showed 1,100 target practices in two metros.
Interviews with eight practice managers put the real blocker on front-desk turnover, not price. A paid pilot with three practices produced two activations and one refund request.
The team redirected the budget to a partner channel through dental staffing firms. Cost per qualified meeting fell from a modeled $900 to an observed $310. The arithmetic is theirs; the file is what made it checkable.
Write the memo, then let a skeptic try to break it
The memo states the question, the methods, the sample, and the sources. It also states the contradictions, the limits, the conclusion, and the review trigger. Cite primary sources and keep raw notes somewhere secure.
A worked excerpt from the scheduling study:
| Element | Line from the memo |
|---|---|
| Question | Can we reach 1,100 practices without paid search? |
| Methods | Eight interviews, three paid pilots, one public-data file. |
| Sample | Managers who hired or replaced a scheduler in ninety days. |
| Sources | Census Business Builder, the AAPOR guide, pilot logs. |
| Evidence | Two of three pilots activated; one asked for a refund. |
| Contradictions | Interviews blamed price; pilots blamed front-desk turnover. |
| Limits | Two metros, eight interviews, no group practices. |
| Conclusion | The partner channel beats paid search here. |
| Next action | Move the $40,000 to staffing-firm partnerships. |
| Review trigger | Cost per meeting drifting toward the modeled $900. |
Hand it to someone who did not run the study. Their job is to trace each conclusion back to a source, a participant, or an observed behavior. They must also name the missing alternative, the leading question, or the generalization the sample will not carry.
Review the survey before it ships
AAPOR's best practices for survey research cover objectives, population, sampling, questionnaire design, pretesting, respondent protection, and disclosure.
Treat the survey tool as software, not as a method.
Borrow the procurement questions
These questions belong to market research, not procurement: the vendor's terms decide what your file can claim, and a file you cannot reopen is not evidence.
The GOV.UK technology selection guidance asks whether a choice is adaptable, who controls the data, what security review it needs, and what it costs to own over time. Those questions apply to a research vendor as much as to a platform.
The CISA software acquisition fact sheet adds development practice, supply-chain exposure, deployment, and vulnerability management. Use them to screen a data vendor; they are not an approval.
Common questions
What should be researched first?
Research the highest-risk assumption, the one that would change the next decision if it turned out false. Size the study to that decision, not to the topic.
How should interviews be recorded?
Give notice, get permission, collect only what the decision needs, and set a retention date. Keep the raw notes restricted and preserve the context around any quote you plan to use.
What belongs in the final memo?
The worked excerpt above lists every element: question, methods, sample, sources, evidence, contradictions, limits, conclusion, next action, and the review trigger. If the result kills the idea, the memo still did its job.
