Platform · Churn prediction
Customer churn prediction that flags who’s leaving while you can still win them back.
In ecommerce, customers don’t cancel. They just stop coming back, and by the time you notice, they’re gone. LiftSignals scores every customer’s risk of not buying in the next 90 days, measured against their own rhythm, so you can reach them while they still remember you.
No credit card · Read-only access · Works with Shopify, WooCommerce, BigCommerce, Amazon, Square or a CSV
Definition
Churn, when nobody cancels.
Customer churn is the share of customers who stop buying from you. Subscription businesses see churn as a cancellation. Ecommerce brands rarely get that signal: a customer simply doesn’t reorder. Churn prediction estimates, for each customer, how likely they are to never come back.
THE CHURN RATE FORMULA
Churn rate = customers lost in a period ÷ customers at the start of that period
Example: 1,000 active customers in January, 280 didn’t buy again within their usual window by June: 280 ÷ 1,000 = 28%. Useful for reporting, but it can’t tell you which customers to save.
| A churn report | LiftSignals churn prediction | |
|---|---|---|
| Tells you | How many left last quarter | Who is likely to leave next |
| Timing | After they’re gone | While they can still be won back |
| Threshold | One cut-off for everyone, like 90 days | Each customer’s own purchase rhythm |
| Output | A percentage | A ranked win-back audience |
Why churn is silent
The customers you lose never tell you they’re leaving.
01
A fixed cut-off misses the real signal.
“No order in 90 days” is late for someone who buys monthly and early for someone who buys twice a year. One threshold gets both wrong.
02
Win-back emails go out after they’ve moved on.
Most lapsed-customer flows trigger at 120 or 180 days. By then, the customer has found another brand.
03
You keep paying to retarget people who are gone.
Lost customers stay in your ad audiences for months, absorbing budget that could go to people still deciding.
How LiftSignals does it
A churn prediction model built on each customer’s own rhythm.
Signal
Each customer’s purchase rhythm: how often they usually buy and how far past that gap they are now. A monthly buyer at day 50 is a different story from a twice-a-year buyer at day 50.
Score
Probability of no purchase in the next 90 days, for every customer, refreshed every night.
Revenue at risk
Churn risk multiplied by predicted lifetime value, so you see dollars at stake, not just head counts.
Minimum data
About 500 customers and 90 days of orders, with repeat buyers in the mix. Below that, the app says so.
Accuracy
Checked on your own data against customers who did and didn’t come back, and shown in the app.
CHURN RISK · NEXT 90 DAYS
Illustrative read-out for a store with 8,400 customers. Value at risk is churn risk × predicted 12-month value. Your figures come from your own order history.
What to do with it
Four plays for customers who are drifting.
01
Win back high-value customers first
Start with customers who are both high risk and high value. A personal note beats a discount code for people who already love you.
02
Time the nudge to each customer
Trigger a reminder when a customer passes their own usual reorder gap, not a store-wide 90-day timer.
03
Stop retargeting customers who are gone
Suppress “likely lost” customers from paid retargeting and move that budget to people still in play.
04
Find out why they leave
Compare high-risk customers’ first products, discounts and channels with your loyal ones to see which entry paths lead to one-and-done.
Proof on every play
Then prove the win-back actually won them back.
Every audience you send carries an automatic holdout: a random slice of those customers who don’t get the message. 30 days later you see what the campaign caused, not what it claimed, in dollars, with the confidence interval printed next to it.
Is it right for you?
Who churn prediction is for.
A good fit if you
- Sell products people should reorder: consumables, apparel, beauty, food, pet
- Have at least about 500 customers and 90 days of orders
- Run win-back or lapsed-customer flows today
- Want to know revenue at risk, not just a churn rate
Not the right tool if you
- Run a contract or subscription business where churn is a cancellation
- Sell one-time purchases with no natural reorder
- Need churn models for SaaS product usage
- Have very few repeat customers yet
FAQ · churn prediction
Churn prediction questions, answered.
Want the math? Read the guide to how to calculate customer churn rate.
How do you predict customer churn?
By modeling each customer’s purchase rhythm and how far past it they are. LiftSignals estimates the probability that each customer won’t buy in the next 90 days and refreshes it nightly from your order history.
What is a churn prediction model?
A model that scores each customer’s likelihood of leaving. In ecommerce, where there’s no cancellation event, it learns from past customers who did and didn’t come back after similar gaps.
What is a good churn rate for ecommerce?
It depends heavily on category and how often people naturally reorder. That’s why LiftSignals measures risk against each customer’s own rhythm instead of comparing you to a benchmark.
Does this work for subscription brands?
It isn’t built for contractual churn, where a customer cancels a plan. It’s built for non-contractual buying, where customers simply stop reordering.
How early can you spot a customer who’s leaving?
As soon as they pass their own usual gap between orders. For a monthly buyer that can be weeks, not the months a fixed 90- or 180-day rule waits.
Can I send at-risk customers to Klaviyo?
Yes. Risk tiers and the high-risk, high-value list sync to Klaviyo, Attentive, Meta and Google, refreshed nightly.
How much order history does churn prediction need?
About 500 customers and 90 days of orders, and enough repeat buyers for the model to learn each customer’s normal gap between orders. Stores where most customers order once get blank churn scores rather than guesses.
RELATED CAPABILITIES
Customer lifetime value →
Predicted 12-month value for every customer.
Customer segmentation →
RFM segments and a combined value-and-risk grid.
Incrementality testing →
Holdouts that show what each campaign caused.
See who’s drifting away by tomorrow morning.
Connect your store this afternoon. Risk scores arrive overnight.