lonstar Qualitative B2B research

claude research

A pattern is repetition with a boundary, not a proof

After the interviews and the traces, the notes look like weather. Pattern finding is the pass that groups them without pretending the group is a sample. You are looking for the same friction, the same phrase, or the same absence across people who do not share a script.

Blank cream index cards linked by olive string on a cork wall, with no writing visible

Terms that get confused on the wall

Anecdote

One vivid story. It can teach a mechanism. It cannot travel into a headline as if others lived it. Keep it in the appendix with the role attached.

Theme

A label you invented to hold quotes: “trust,” “speed,” “risk.” Themes are buckets. They are not yet findings. If the label could fit any B2B study, it is too soft.

Pattern

The same mechanism in at least two roles or two separate deals: a missing diagram restarts review; a champion deletes the word “platform”; procurement signs the vendor already in the portal. State the boundary beside it.

Proof

A claim that a defined population behaves this way at a stated rate. Claude research does not produce proof of that kind. Do not borrow the word because a cluster feels sturdy. The glossary keeps the distinction in one place.

How notes get clustered

Work from quotes and events, not from the theme you hope to sell. Put each note on its own card: speaker role, deal outcome if you know it, and the sentence. Group by mechanism. “Security” is a weak group. “The review stopped until someone redrew the data flow” is a mechanism. Split cards that only share a noun.

Then test the group. Does it appear in a champion and in someone who can refuse, or only inside one enthusiastic team? Does market observation show the same artifact? If the cluster lives entirely inside people your sales team primed with the same deck, mark it as possible echo, not as a pattern. Echoes still matter. They tell you which language is already in the account. They should not be laundered into “customer truth.”

A worked pass

Suppose six conversations mention time. Three of them, from a user, a champion, and a reviewer, describe the same week: the pilot worked locally, then the security appendix asked for a diagram, and the champion waited rather than draw it. Two other notes say the product “saves time” with no episode attached. One economic buyer says the renewal was late because budget froze. You do not have a pattern called time. You have a candidate pattern about a missing diagram, an anecdote-shaped slogan, and a separate budget stall. Write them apart. That discipline is what makes the later purchase friction page usable.

Stop when new conversations repeat mechanisms you already named, or when they clearly open a different one you cannot staff. Stopping is a judgment. Say so. “We stopped after nine conversations across four roles because the diagram stall had shown up in three deals and the newest calls were not adding a mechanism” is a complete sentence. “We reached significance” is not available to this method.

Notes on clustering without fake certainty

lonstar sends occasional Claude research notes. The list is a notebook, not a nurture sequence aimed at a quota.

When the clusters are named, write the cut. Leave the rest in the folder.