Start from the question, not the method
People decide to "do some research" and reach for a survey, because a survey feels like research. Method should follow the question, and most early questions are the kind surveys answer badly.
| Your question | Method |
|---|---|
| How does this work today? | Interviews — open, about past behaviour |
| Where exactly does it break? | Watch someone do it, in silence |
| How common is this problem? | Survey — but only after interviews name the options |
| Would they pay? | Try to take money; nothing else answers this |
| Which of two wordings works? | A test with real traffic |
| What do they call this? | Read their own writing — forums, reviews, tickets |
Finding people with no audience
- Communities where the work is discussed — participate first, then ask; a drive-by request gets ignored or removed
- Reviews of competing products, where people have already described their problem in public with their name attached
- Your own network's second degree — "who do you know who does X" outperforms asking your network directly
- Cold outreach to a named list, referencing something specific about their work
- Local trade groups and associations, which are unfashionable and answer the phone
On incentives: a voucher is fine and slightly changes who agrees. Unpaid conversations select for people who care about the problem, which is exactly the bias you want early on.
How many is enough
Five to eight conversations with people in the same role usually surfaces the dominant patterns. You'll know you're near the end when a conversation contains nothing you haven't already heard twice.
- Under five, you're generalising from individuals
- Five to eight, the pattern appears if there is one
- If nothing repeats by ten, that's a finding: you have several segments, not one market
- For each distinct role or segment, start the count again — an agency owner and an agency employee are different studies
From notes to decisions
The common failure isn't collecting too little. It's collecting a folder of transcripts nobody converts into a decision. The conversion step is mechanical and takes an hour.
- Pull out quotes, not summariesTheir words. Your paraphrase has already smuggled in your interpretation.
- Group quotes by the problem they describeNot by person. Groups that contain several people are patterns; groups of one are anecdotes.
- Count how many people mentioned each group unpromptedUnprompted mention is the strongest signal in a transcript.
- Write what each pattern would mean if true"If this is real, we build X first and drop Y." A finding with no consequence isn't a finding.
- Name what you're still guessingResearch narrows uncertainty; it doesn't remove it. Say what's left so the next round has a target.
Common mistakes
- Running a survey before you know what the options are
- Talking only to people who are easy to reach and therefore unrepresentative
- Stopping at three conversations because they agreed with you
- Recording paraphrases instead of quotes
- Finishing the research and never writing down what it changed
Key takeaways
- Pick the method after the question; most early questions need interviews, not surveys
- Five to eight per segment is usually enough for patterns to appear
- Unprompted mentions are the strongest signal you'll get
- Research that doesn't end in a written decision was a hobby
Try it yourself
Book three conversations this week with people who have the problem you think exists. Afterwards, write one sentence per conversation in their own words. If the three sentences don't overlap, you've learned something important about your market's shape.