The most dangerous customer in ecommerce is not the one who complains — it is the one who quietly stops. No return request. No unsubscribe. Just silence. Churn prediction is the discipline of identifying these customers before they leave, while there is still time to act. And you do not need a machine learning model to do it.
Why Churn Happens Before You See It
Customers do not churn when they cancel a subscription or request a refund. They churn weeks or months earlier, when their engagement starts to decay. They stop opening emails. They visit less frequently. Their time on site shortens. They stop clicking on product recommendations.
By the time a customer's last purchase date is 90 days old, the churn decision has already been made — usually at day 30 or 40. The brands that win at retention are the ones who identify that early signal and respond to it, not the ones who send a win-back email at day 90.
Three Signals That Predict Churn
- Engagement Decay: A customer who used to open every email but has not opened one in 30 days is showing you something. Email engagement is one of the earliest leading indicators of churn, often preceding actual purchase lapse by 4-8 weeks.
- Purchase Gap Widening: Compare each customer's current days-since-last-order against their historical average. When the gap exceeds 150% of their normal cycle, the risk is elevated. This is the signal RFM recency captures — use it proactively, not retroactively.
- Browsing Without Buying: A customer who visits your site but does not add to cart, or adds to cart but abandons, is signaling friction or reduced motivation. Tracking this behavior against a baseline can reveal declining purchase intent weeks before the purchase actually stops.
Building a Simple Churn Score Without Any ML
You do not need a predictive model. You need a scoring system. Assign points for each risk signal:
- No email open in 21+ days: +2 points
- Days since last order exceeds 130% of personal average: +2 points
- 3+ abandoned carts in last 30 days: +1 point
- No site visit in 14+ days: +1 point
- Return or refund in last 60 days: +1 point
A customer scoring 4+ points is at elevated churn risk. A customer scoring 6+ should trigger an immediate intervention. This is not science — it is a structured way of surfacing what the data is already telling you.
What to Do When You Identify a At-Risk Customer
Speed matters. The earlier in the decay curve you intervene, the higher the recovery rate. A customer at 30 days past their normal cycle will respond to a relevance-based message. A customer at 90 days needs a discount and still might not respond.
The intervention should match what triggered the churn signal. Engagement decay → send content or a product recommendation, not a promotion. Purchase gap → a targeted reminder based on what they previously bought. Browsing without buying → address the friction (a question, a review prompt, or a small incentive to complete the purchase).
The goal of churn prediction is not to predict churn. It is to intervene early enough that you never have to run a win-back campaign.
The One Number to Track
Your 30-day churn rate — the percentage of customers who were active in the previous period and made no purchase in the current period. Track this monthly. If it is rising, your retention strategy needs to shift upstream. If it is falling, your early interventions are working.
Most brands track last-purchase date as a lagging indicator. Shift your attention to engagement metrics as leading indicators. The customers who are churning next month are showing you signals this month — if you know where to look.
Tools You Already Have
You do not need a new tool. Your email platform already has open rates. Your ecommerce platform already has purchase dates and session data. Your CRM already has customer history. The data exists — what most brands are missing is the process of combining it into a churn risk view and acting on it weekly.
Start with one signal: email engagement. Set up an automated segment of customers who have not opened an email in 21 days but are still within their normal purchase cycle. Send them a relevance-based email. Measure recovery rate. Improve. That is churn prediction in practice.