Churn Prediction Explained
Understand how AskBiz predicts which customers are likely to churn and how to use churn risk scores to take action early.
What churn prediction does
AskBiz's churn prediction model analyses your customer purchase history and assigns each customer a churn risk score — a probability (0-100%) that they will not purchase again within their expected repurchase window. Customers with high churn risk scores are the ones most worth targeting with retention interventions before they are lost.
How the model works
The model is based on RFM analysis — Recency, Frequency, and Monetary value — combined with your specific customer cohort's natural repurchase rhythm.
The key signals are:
- Recency — how long since the customer last purchased, relative to their own historical purchase frequency
- Purchase frequency trend — is the customer buying less often than they used to?
- Spend trend — is the customer spending less per order than their historical average?
- Category drift — has the customer stopped buying from a category they previously purchased regularly?
Customers who have gone longer than 1.5x their normal repurchase interval without buying receive an elevated churn risk score.
Reading the churn risk dashboard
The churn risk dashboard (Customer Intelligence → Churn Prediction) shows:
- High risk (70-100%): customers who are very likely already churned or about to — act immediately
- Medium risk (40-69%): customers showing warning signs — prioritise in your next retention campaign
- Low risk (0-39%): customers whose behaviour is within their normal pattern — monitor but no urgent action needed
You can filter the list by segment, acquisition channel, or product category to identify whether churn risk is concentrated in a particular customer type.
Acting on churn predictions
For each high-risk customer segment, AskBiz suggests a retention action based on their purchase history:
- Customers who lapsed after a single purchase → welcome back offer with first-repeat incentive
- Customers whose purchase frequency has declined → re-engagement campaign highlighting new products
- Customers who stopped buying a specific category → targeted reactivation for that category
These suggestions are starting points — your knowledge of your customers and business will inform the best approach.
Improving prediction accuracy
Churn prediction accuracy improves with more data. The model needs at least 6 months of order history and at least 100 customers who have made more than one purchase to generate reliable predictions. If you have less than this, the churn dashboard will show indicative scores with a low-confidence flag. Accuracy typically improves significantly after 12 months of data.