Churn Probability
AnalyticsAlso: Churn Prediction · Churn Score · Propensity to Churn
Quick definition
Churn probability is a score assigned to each customer expressing how likely they are to stop purchasing or cancel within a defined future window. It is produced by a predictive model trained on past behaviour signals such as purchase frequency, recency, engagement and support contact history.
How it varies across Australia
Churn probability models vary widely in accuracy depending on the volume of historical data available and the consistency of the behavioural signals fed into them. Businesses with shorter customer histories or fewer repeat transactions tend to produce noisier scores that need wider confidence thresholds before acting on them.
See retention patterns across Australian industries →What it actually means
Churn probability is the answer to: which of our customers are about to leave, before they actually do?
Every customer leaves an evidence trail before they churn. They buy less frequently. They stop opening emails. They log fewer sessions. They contact support about a recurring problem. A churn probability model reads those signals and converts them into a number between zero and one, updated on a schedule that matches how fast your customer behaviour changes.
The score is only valuable if you do something different because of it. A business that calculates churn probability and then treats every customer the same has done the analytics work and skipped the operations work.
The right use of a high churn probability score is triage. It tells your retention team which customers to call first, which segment to re-engage with a specific offer, and which accounts to flag for an account manager before the cancellation request arrives. Paired with lifetime value, churn probability also helps prioritise where that retention investment is worth making. A high-probability churner with low LTV is a different decision from a high-probability churner who represents a significant revenue line.
Churn probability sits inside a broader retention analytics picture alongside churn rate, cohort analysis and net promoter score. The score is a forward-looking signal. The churn rate is the backward-looking result.
A churn probability score is only useful if it arrives early enough to act on and late enough to be accurate.
How to calculate it
Churn Probability = model output score (0-1) for a defined prediction window, typically 30, 60 or 90 days
Worked example. A subscription business trains a model on 24 months of customer data. Inputs include days since last login, number of support tickets in the past 60 days, change in monthly usage volume, and payment failure history. The model outputs a score for each customer daily. Customers scoring above 0.7 enter a high-risk retention queue and receive a proactive outreach within 48 hours.
The Australian context
Australian subscription businesses face a specific churn dynamic around cost-of-living sensitivity. Macro events like interest rate rises produce churn spikes that look like individual customer signals but are actually correlated across an entire cohort. A well-built churn probability model should be recalibrated after these events, otherwise the model reads normal post-spike behaviour as elevated risk and triggers unnecessary retention spend.
Where people get this wrong
Related terms
Common questions
What data do you need to build a churn probability model?
At minimum: purchase or login recency, frequency of engagement, and some signal of intent such as support contacts or feature usage changes. The more historical cohort data you have showing who actually churned and when, the more accurate the model becomes. Thin data produces noisy scores.
How is churn probability different from churn rate?
Churn rate measures the share of customers who already left over a period. Churn probability scores individual customers on how likely they are to leave in the future. One looks backward at the aggregate. The other looks forward at the individual.
At what score threshold should you intervene?
There is no universal threshold. Set the cutoff by working backwards from your retention capacity. If your team can handle 50 outreach contacts per week, set the threshold to capture roughly that many. Adjust based on how the intervention performs over time.
Can you build churn probability scoring without a data science team?
Yes. CRM platforms like HubSpot and Salesforce surface basic health scores. Email platforms track engagement decay. A simple rule-based model using recency and frequency tiers is often more actionable than a complex machine-learning model that nobody in the business understands well enough to act on.
Keep exploring
About New Rebellion
New Rebellion is a marketing intelligence consultancy. We build tools, score Australian businesses on how their marketing actually performs, and publish Debrief every day. This dictionary is part of how we work in the open.
How we think →