claude research
Market observation records what people do before they narrate it
Interviews ask for a story. Observation asks what the story left behind. In Claude research, market observation means watching buying behavior in places you are allowed to be: a consented sales call, a ride-along, a trade-show aisle, a procurement portal your team operates, a public pricing page, a security packet that was actually sent. It is not covert monitoring of private people, and it is not a scrape of inboxes you do not own.
A week of watching
Keep the week small enough to write up. The point is a trace you can put next to Claude customer interviews, not a warehouse of screenshots.
-
Day 1
List the traces you already have
Call recordings with notice, lost-deal notes, the last twenty security questionnaires, the pages a reviewer is sent. If you would need a hidden tool to see it, it is out of scope.
-
Day 2
Sit in on a live path
With permission, listen to one full sales conversation or shadow one implementation call. Note who speaks, who is absent, and which question restarts the meeting.
-
Day 3
Walk a public market surface
Compare how three alternatives explain the same job in public. Record the nouns. Do not score them. Language comparison is not market share.
-
Day 4
Write only what repeated
A single awkward call is an anecdote. A question that restarts three calls is a candidate for pattern finding.
Touchpoints that leave a trace
Prefer artifacts over impressions. A forwarded email with the champion’s own subject line is worth more than a memory of “good energy in the room.” A rescheduled security review is a behavioral mark. A badge scan at a booth is weak unless you also know which conversation followed it.
Build the inventory on the touchpoint mapping page, then come back and watch two of them properly. Breadth without a trace becomes the journey poster this method is trying to retire.
Observation also catches absence. The economic buyer who never appears on the call, the legal reviewer who replies only by redline, the pilot team that stops logging in after week two: those are events. They belong in the same notebook as quotes, and they often explain purchase friction better than a satisfaction score.
Boundary
Consent, access, and the line you do not cross
Record calls only where notice is real and the law of that place allows it. Do not impersonate a buyer to extract a competitor’s process. Do not quietly join customer communities under a false name. Do not treat an employee’s private messages as a field site. Claude research can be sharp without being covert. If access is unclear, ask, or stay with materials the company already has a right to review.
When you write the note, separate observation from inference. “The reviewer opened the security appendix and not the story page” is an observation if your analytics or a shared screen showed it. “They do not care about narrative” is an inference and should be marked as one. Inferences are allowed in positioning synthesis. They are not allowed to masquerade as counts. Nothing in a week of watching is a representative statistic, and nothing in it is a promise that revenue will move.
Put the week next to the interviews
A trace that contradicts a confident champion is a finding. It is still not a market share. Carry both into the glossary when the team starts calling the work “data” without saying what kind.