Claude · Design
Scope one decision, then write what you will not claim
A Claude study is designed before anyone is recruited. You name the purchase, the roles who can move it, the access you actually have, and the sentences the final memo is forbidden to contain. Those forbidden sentences include any promise of return on investment and any claim that the people you reached represent a market, a segment, or a pipeline.
Start from ignorance, not from a campaign
Write the question as a gap: “We do not understand why security questionnaires stall pilots of data-access tools after the champion has already chosen a favorite.” That is researchable. “How do we prove ROI to the enterprise?” is a wish. If you cannot say what would surprise you, the study will only collect agreement. Tie the question to one buying job, not to your whole portfolio. The interview path that serves this question is documented in decision-maker interviews.
Access is the design, not a logistics footnote
A sample of current happy customers will describe the speech that won. You also need a loss, a long delay, and someone who kept the incumbent or the spreadsheet. Recruit the blocker directly when you can: security, legal, finance, the operator who inherits the Monday meeting. Tell them you are not there to reopen the sale. Pay or gift only in ways that do not ask for a testimonial. Record refusals. A memo that hides how many people declined is editing the vantage point.
Time-box the work. Four to six weeks is enough for a narrow decision if recruiting starts on day one. Observation of one live touchpoint, or a close reading of the artifact that touchpoint uses, should be on the calendar before synthesis. See touchpoint fieldwork for what “one live touchpoint” means, and market observation for public substitutes when you are not invited.
Enough pattern is not a census
Stop adding interviews when new ones repeat a friction you can already narrate, and when you can describe at least one decision that does not fit. Call that saturation of this design. Do not call it representativeness. Do not compute a share. Do not project the pattern onto “the mid-market.” The next honest step is a sentence with its evidence, which is positioning synthesis, fed by phrases you did not polish away in the language bank.
Ethics that keep the room real
Consent before recording. A plain account of who will read notes. A way to strike a sentence. No fake buyer identities, no quiet recording of a call you joined as a vendor, no scraping of a private channel because the posts are “already there.” If your presence changes the meeting, say so in the log. Confidentiality is part of the method: a blocker will not describe a scare if the note is headed for a public case study with their employer’s name. The way this site itself handles addresses and local preferences is in the privacy policy. Study data from a client project is a separate agreement; do not assume a website policy covers it.
Questions at the start of a study
How many interviews is “enough”?
There is no universal count. A narrow decision sometimes steadies around a handful of roles repeated across contrasting outcomes. If every conversation is still a new mechanism, you are not done, and you are also not allowed to publish a rate while you continue.
Can the sponsor sit in?
Rarely. A seller in the room turns the hour into account management. If the sponsor must hear it, use a later readout of patterns, with verbatim cleared for that audience, not a live demo of the source.
Where do findings go?
Into a memo with limits on page one, a friction list a salesperson can misuse if you let them, and positioning language tested by being read back. Not into a dashboard that implies precision the design never had.
Field notes by email
Occasional notes from stutklit
New Claude chapters, not a sequence timed to a fiscal quarter. Write to [email protected] to leave the list.
Design serves a conversation
When the limits paragraph is drafted, book the first person who can refuse the purchase.