Protocols
A protocol keeps interviews from turning into friendly chats
The interview guide says what a good trace contains. This page says how a study is staffed, consented, and written down so the claude standard can be checked by someone who was not on the call.
A study week, in order
Day 0
Write the range before you write the script
Name the situations you need: a win, a loss, a stall, a renewal, a non-buyer. Add the roles from the committee map. This grid is a coverage tool. It is not a sampling frame and it does not authorize a confidence interval. Say that out loud in the kickoff so nobody later calls twelve conversations “representative.”
Day 1
Consent and the boundary of the record
Tell the participant who will read the notes, whether you record, and how you will strip account names. B2B buyers often speak about internal politics they would not put in a testimonial. The protocol should promise a level of disguise you can actually keep. If legal or procurement at your own company must review the guide, do that before the first call, not after a candid recording exists.
During the call
One deal, then the touchpoint
Stay on a single decision until the sequence is visible. Then ask them to reopen one artifact. That bridge is what lets observation start before the meeting ends. If time collapses, cut feature-ranking questions first. They are the part most likely to manufacture fake precision.
Same day
Separate utterance, inference, and absence
Write the trace while the call is still a scene and not a moral. Mark what you inferred. List who was missing. Those three layers are what pattern synthesis is allowed to code. A cleaned-up paragraph that blends them cannot be audited, and the evidence standard will send it back.
After the range
Stop when coverage stalls, not when a quota feels scientific
Add conversations while a new role or a new stall is still appearing. Pause when additional calls repeat scenes you already understand and the empty cells in the grid are empty because those people will not talk, not because you forgot them. Document the pause. That note is part of claude discipline. It prevents a small set of willing champions from being narrated as a market.
What belongs on the one-page guide
Keep the guide short enough to use while listening. A script that fills four pages becomes a questionnaire, and questionnaires pull B2B research back toward the fake census this hub refuses.
- The situation you are there to reconstruct, in one sentence.
- Five sequence questions, written in the past tense, about a real deal. Future-tense “would you” questions stay off the page.
- A prompt to open one touchpoint and narrate the last use.
- A prompt for personal downside, aimed at the role in front of you.
- A closing question on who else had to agree, so the committee does not collapse into one voice. Terms are standardized in the glossary.
Field notes
Occasional claude notes, not a sequence of promises
titklifu emails when a guide is revised. The notes stay methodological. They do not forecast pipeline or attach a performance claim to a method.
Run the call, then code what repeated
Protocols without coding become a stack of transcripts. Coding without protocols becomes a mood board. Use both.