Amplitude investigation

When this applies

Use this after an Amplitude chart, funnel, retention curve, or cohort is already computed. The connector is https://mcp.usercall.co. The entry tool is create_study. The shared loop is on Ask users why.

Signal

Keep Amplitude's definition of the cohort: the event, the window, and the counts. A correlation or a predicted driver is a hypothesis you can take into the interview. It is not a sentence from the user.

Example: Week-4 retention for accounts that used Export in week 1 fell from 46% to 28% after the March export redesign. The cohort is 860 accounts. User streams show Export, then no session in the following three weeks.

Unknown

The open question is why accounts that successfully exported in week 1 stopped coming back. A broken file, a missing destination, and a sense that the export was not worth the trip are different causes. The retention curve has not separated them.

Interview affected users

Interview people in that cohort: Export in week 1, no return by week 4. Tell the study what Amplitude already proved so the conversation starts at the cause.

create_study({
  key_research_goal: "Why did accounts that exported in their first week stop returning by week 4? Ask what they needed the file to do, what happened when they opened it, and what they use now.",
  business_context: "Analytics product. Amplitude retention: week-4 return for accounts that used Export in week 1 fell from 46% to 28% after the March export redesign. Cohort size 860. Streams show Export, then no session for three weeks. Interview people in that cohort.",
  target_interviews: 8,
  duration_minutes: 12
})

Share interview_link with that cohort. One active agent study is allowed per account. A short credit balance returns checkout_url for a person to open.

Evidence

Poll get_study_status until status is complete, then read the summary.

get_study_results({
  study_id: "<study_id>",
  format: "summary"
})

Report the themes next to the retention drop. Leave the 46% to 28% figure as Amplitude's measurement. The interviews explain it. They do not replace it.

Stop here

If the chart is a tracking gap — the return event stopped firing, or the cohort definition changed in the same release — fix the definition before you ask users why they left.