ChurnMiserPredictive churn scoring explained: how AI flags at-risk customers
A churn score is a probability that a customer is about to stop buying. Here's how predictive scoring works in plain English, what signals it uses, and how to turn a score into a recovered customer.
What is predictive churn scoring?
Predictive churn scoring assigns every customer a number — usually a probability from 0 to 1, or a 0–100 risk score — representing how likely they are to stop buying. It answers a sharper question than "who has already lapsed?": it asks "who is about to?"
That shift matters because the cheapest customer to keep is one who hasn't left yet. Once a customer is fully dormant, winning them back is harder and more expensive than nudging an at-risk customer who still remembers you.
Why it beats a fixed inactivity window
The simplest churn definition is a single window: "anyone who hasn't bought in 90 days is churned." It's easy, but blunt. A loyal weekly buyer who goes quiet for 30 days may be a bigger red flag than an occasional buyer at day 89. A fixed window treats both the same.
A churn score is personalized: it compares each customer to their own normal pattern, not a one-size-fits-all cutoff.
This is why predictive scoring pairs well with — and improves on — the basic churn rate calculation: the rate tells you how much you're losing; the score tells you who, and how urgently.
The signals a churn model uses
Most ecommerce churn models start from RFM and add behavioral data:
| Signal | What it captures | Why it predicts churn |
|---|---|---|
| Recency | Time since last purchase | The single strongest signal — risk climbs as a customer drifts past their normal reorder gap |
| Frequency | How often they buy | Frequent buyers slowing down is an early warning |
| Monetary | How much they spend | Weights the cost of losing them; high-value churn hurts most |
| Behavioral | Site visits, email opens, product views | Disengagement often precedes a missed reorder |
| Lifecycle | Tenure, first vs repeat, category | First-time buyers churn differently than established customers |
For more on how RFM groups customers into actionable segments, see RFM segmentation for Shopify.
How a score is produced
At a high level, the process is:
- Collect history. Pull each customer's order history and behavioral events.
- Engineer features. Turn raw events into the signals above (e.g. days since last order, rolling purchase frequency).
- Learn the baseline. Calibrate to your store's normal patterns rather than a generic benchmark — a 45-day gap means different things for coffee and for furniture.
- Score & refresh. Output a per-customer risk score and update it regularly (ideally daily) as new behavior arrives.
ChurnMiser runs this loop nightly on Amazon Bedrock AI, scoring every customer against your store's own baseline using RFM plus Shopify Pixel behavioral signals.
Turning scores into action
A score is only valuable if it drives a decision. The practical playbook:
- Triage by risk × value. Prioritize high-risk, high-value customers first.
- Match the message to the stage. A gentle reminder for early-risk; a stronger incentive for high-risk; a true win-back campaign for dormant.
- Automate it. Scores change daily; manual exports can't keep up. Automated campaigns triggered by risk thresholds do.
- Measure and learn. Track who came back, and feed the outcome back into your targeting.
Frequently asked questions
What is predictive churn scoring?
It assigns each customer a probability that they'll stop buying, based on their behavior. Instead of waiting for a customer to lapse, a model reads signals like recency, frequency, spend, and on-site behavior and outputs a risk score so you can intervene early.
What signals does a churn model use?
Recency (time since last purchase), frequency (how often they buy), monetary value (how much they spend), and behavioral data such as site visits, email engagement, and product views — often summarized as RFM plus behavioral signals.
Is predictive churn scoring accurate?
No model is perfect, but one calibrated to your store's own baseline and refreshed frequently is far more useful than a single fixed inactivity window. Accuracy improves with per-store calibration and by measuring outcomes and feeding them back.
See your store's churn risk in 2 minutes
Take the free ChurnMiser retention audit — answer 7 questions and get a personalized revenue-leak report showing exactly where you're losing customers. No credit card.
Take the free retention audit →Or install ChurnMiser from Shopify and start scoring customers today.