When this Agent fits
Use it when acquisition continues but activation or retention remains unclear, when lifecycle messaging has become a calendar of disconnected sends, or when cancellation survey answers are being treated as complete root causes.
Inputs it needs
- Activation and retention definitions with eligible cohorts
- Product events, lifecycle messages and experiment history
- Optional interviews, support signals and cancellation evidence
- Consent, channel policy, audience exclusions and experiment owner
Operating loop
01
Define
Confirm the cohort, behavior, baseline and result window before diagnosing leakage.
02
Diagnose
Compare behavioral evidence, stated feedback and lifecycle touchpoints without merging them.
03
Experiment
Prepare one bounded product or message test with a clear owner and stopping rule.
04
Learn
Read the result after the agreed window and preserve only scoped, sourced learning.
Outputs
- Activation or retention constraint map
- Cohort and evidence-gap diagnosis
- Experiment or lifecycle-message brief
- Outcome readout and learning proposal
Boundaries and approvals
- It does not infer motivation from one metric or one survey response.
- It does not send messages or change product flows without permission and review.
- Retention populations, download totals and revenue cohorts are kept separate.
- A test result is not generalized beyond its cohort and time window.
How the result is verified
- Metric definitions, cohort filters and data coverage are visible.
- Every experiment states success, guardrail and stopping conditions.
- The final readout records verified movement, no impact or insufficient evidence.