Week two annoyed me. Drawing eight memberships by hand felt archaic until I noticed three of them only ever visited on bank holidays. That pattern was invisible in our weekly active tile.
S. Okafor · product ops, LondonFlagship course
Early Warning Cohort
A seven-week sitting for operators who already have a churn model, a CRM flag, or a weekly export — and who suspect the flag is late. Informational fee £1,180. No checkout on this site.
What you leave able to do
- Rebuild an account’s typical cadence before applying any risk label.
- Mark a hazard window that a finance partner can audit, including who was censored.
- Log save offers as events that contaminate later features.
- Run a 20-minute silence interview without defaulting to a discount.
- Write a one-page detection brief for a single product, not a platform slogan.
Modules
Week 1 — Inventory without theatre
You bring the current rule (SQL, Looker tile, or a spreadsheet). We catalogue what it actually measures versus what the team believes it measures.
Week 2 — Cadence portraits
Each student reconstructs eight accounts by hand. The point is friction: you feel which bins were a convenience.
Week 3 — Hazard windows
Discrete-time hazards, competing pauses, and the difference between “did not open the app” and “did not belong that week”.
Week 4 — Feature hygiene
Leakage clinic. We keep a public ledger of illegal features discovered in the room.
Week 5 — Silence interviews
Live practice with a script, then a critique of recordings you make at work (no customer audio is stored by us).
Week 6 — Intervention bias
How last quarter’s win-back campaign made this quarter’s model look clever.
Week 7 — Detection brief
A reviewed artefact: window, exclusions, owner, and the first metric you will stop reporting.
Notes from past sitters
★★★★☆
Strong on leakage. The intervention-bias week assumed more campaign logging than we have. I still use the detection brief template.
Anonymous client in regional broadband