FuseChurn Propensity
Churn Propensity

A score you can read

No black box. The risk score is the sum of a handful of named signals, each with a plain-English reason. Some push risk up; breadth, tenure and a climbing habit protect against it.

The signals

±35
Recency drift
How far past their normal rhythm they have drifted
±20
Frequency trend
Whether their activity rate is fading, holding or climbing
±15
Engagement depth
A shallow, single-product footprint lapses more easily
±10
Tenure
New relationships lapse more easily than established ones
±10
Early-lapse pattern
A one-and-done start rarely self-corrects without a nudge
±10
Payment fragility
Subscribers churn silently on a failed payment, not a decision

Points sum to a 0 to 100 score. Protective signals (a broad footprint, a long tenure, a climbing frequency) subtract, so a healthy, engaged relationship scores low. Illustrative weights; the production model calibrates them on your data.

The bands

Healthy
score 0 to 24
Watch
score 25 to 49
At risk
score 50 to 74
Critical
score 75+
What feeds it in your world

No new tracking to invent. These are events you already capture; the model reads them as they are.

  • ·Timestamped purchases, per customer (recency and personal cadence)
  • ·Purchase frequency over time, so a fading habit is visible
  • ·Products used, the breadth that protects against lapse
  • ·Subscription status and payment events, the involuntary-churn signal
How it sharpens over time

Models are tuned, not installed. Each step is earned on evidence, and the previous step keeps running until the next one beats it.

  1. v0
    Heuristic, day one
    Transparent rules on the data you already have. Live in Braze in weeks, and everyone can see why it fired.
  2. v1
    Calibrated to your history
    Thresholds and weights re-fit to your own base and vertical, backtested against what actually happened.
  3. v2
    ML where it earns it
    A trained model replaces the rules only when it beats them on holdout data. Explainability stays a requirement.
  4. v3
    Monitored and re-tuned
    Drift watched, thresholds reviewed on a cadence, and every change proven with holdouts before it ships.

Churn starts as these transparent signals, gets its weights re-fit to your own lapse history, and graduates to a trained model only when it beats the rules on holdout data. The at-risk threshold is tuned per vertical, because a quiet fortnight means different things in different businesses.