Hazard windows are not “the next 30 days”
Why a rolling calendar month hides the week a gym member always skips, and how we mark it instead.
Studio · Llanrhidian
Retention has a weather pattern. We teach people to read it.
Workspace Fieldcore is a small teaching desk for Churn Detection Analytics. We work from the moment an account goes quiet, not from the dashboard that only lights up after the card is declined.
Heard in the last three sittings
The Early Warning Cohort made me stop treating a 30-day inactivity flag as a strategy. We now score silence against the person’s usual cadence, which is slower work and far less noisy.
Priya H., retention lead, Bristol membership club
I still argue with the survival module — our billing cycle is messy — but the interview protocol for quiet accounts is the piece I actually run every Friday.
James, Cardiff
4 / 5 after the Feature Hygiene clinic. Useful, slightly severe on leakage.
Platform review · learner in insurance ops
Flagship
Early Warning Cohort
Seven weeks with a live desk, a shared hazard notebook, and a final review of one product you actually ship. Informational fee: £1,180 for the full sitting.
You leave with a detection window you can defend to a sceptical finance partner, not a generic “at risk” label.
See modules and limitsWhat the desk actually changes
01
Cadence before labels
Most churn scores flatten people into weekly bins. We reconstruct the account’s own rhythm first, then mark departures from that rhythm. It is slower to set up and kinder to seasonal members.
02
Intervention as a contaminating event
A save offer is not a neutral observation. The method page walks through how outreach biases the very signals you hoped to measure, and what to log so the next cohort is still honest.
03
GB membership quirks
Direct debit timing, cooling-off windows, and mid-month pause culture show up differently than US card-on-file churn. We teach with those clocks in view.
From the journal
Short pieces written after classes, not marketing copy dressed as insight.
Accuracy is a flattering metric for rare cancels
A 97% model can still miss the accounts that matter. Calibration notes from a sitting in March.
18
2,140
11-day
7 weeks