Churn prediction is a model that estimates which customers are at risk of lapsing. In retail and hospitality, churn usually means a regular who has not visited or ordered for longer than their normal pattern suggests.
The model compares each customer's recent behaviour with the patterns of people who previously stopped buying. Warning signs include longer gaps between visits, falling spend, fewer app opens, ignored emails or a recent complaint. The output is a risk score that can feed a win-back campaign.
Retailers with loyalty programs use it to trigger personalised offers, reminders or a call from a store manager. Subscription and meal-plan businesses and venues with memberships, such as gyms and clubs, use it to protect recurring revenue. It also helps focus discounts on those who need them rather than everyone.
You need enough identified customers and history: cash-only, anonymous trade gives little to learn from. Test whether outreach actually changes behaviour by holding back a control group, because some customers would have returned anyway. Avoid spamming at-risk customers, and respect opt-outs.
Keeping an existing customer is generally cheaper than finding a new one, so early warning has a clear payoff. Combine the score with a sensible offer budget, so that the cost of winning someone back never exceeds what that customer is worth.