Computing churn rate

Churn rate = customers lost in the period / customers at the start × 100%

An example: 10,000 active customers at the start of a quarter, 1,200 of whom made no purchase within
90 days → churn rate = 12%.

Choosing the churn window

The window depends on the category — on the typical gap between purchases:

Category Typical purchase frequency Churn window
FMCG, groceries Weekly to monthly 30–60 days
Fashion, apparel Two to four times a year 90–120 days
Electronics Once or twice a year 180–365 days
Furniture, renovation Less than once a year 365+ days

Churn prediction: how the model works

The model studies the features that precede churn:

  • Falling visit frequency: three times a week before, once a month now
  • Shallower sessions: fewer pages per visit
  • RFM movement: recency rising, frequency falling
  • No reaction to campaigns: open rate declining

From those features it assigns each customer a churn score and enrols the high-risk ones into a
retention segment.

Retention scenarios at high churn risk

Churn score above the threshold → retention trigger:
├── Email with a personal selection
├── Push notification with a personal offer
├── An onsite banner on the next visit
└── Personalized recommendations on the homepage

Reactivating in time works several times better than working with customers who have already gone.