Survival Analysis

CRM & Retention

Also: Time-to-Event Analysis · Hazard Modelling

Core ideaModels when churn happens, not just if
Key outputHazard rate over customer lifetime
Common pitfallIgnoring customers who haven't churned yet
vs cohort analysisTime-to-event, not fixed-interval snapshots

Quick definition

Survival Analysis is a statistical method that estimates how long a customer relationship is likely to last before an event such as churn occurs. Borrowed from medical research, where it originally tracked patient survival time, marketers use it to predict when customers are likely to leave, not just whether they will.

How it varies across Australia

Survival curves vary sharply by business model across Australia. Subscription telcos and streaming services see risk cluster tightly around contract or trial end dates, while usage-based SaaS businesses tend to show risk spread more evenly across the customer lifetime. The shape of the curve matters more than any single churn percentage.

See retention patterns across Australian industries

What it actually means

Picture two hospital wards tracking patient survival. One ward reports that sixty per cent of patients survive year one. The other tracks exactly when each patient's condition changes, week by week. Survival analysis is the second ward's approach applied to customers instead of patients.

Most churn rate reporting only tells you the average proportion of customers lost in a fixed period. It flattens the timeline. Survival analysis keeps the timeline intact and produces a hazard rate, the changing probability of churn at any given point in a customer's life. A telco might learn risk peaks at month three then drops. A SaaS company might find risk stays flat until a renewal date approaches then spikes.

The technique also handles a problem plain churn rate cannot. Some customers in your dataset haven't churned yet, they just haven't reached the end of the observation window. Survival analysis includes them properly rather than discarding them or wrongly counting them as retained.

Retention rate, cohort analysis and lifetime value all describe pieces of the same puzzle. Survival analysis is the model that ties the timing together and tells you where to intervene, not just how much churn happened.

Two businesses can share the same annual churn rate and still be fighting completely different fires.

How it shows up

Survival analysis shows up as a survival curve, a line that starts at one hundred per cent and slopes downward as customers leave over time. Steep early drops signal an onboarding problem. A slow steady decline late in the curve usually points to competitive or pricing pressure. Analysts also produce a hazard function, showing exactly which months or weeks carry the highest churn risk, which is what feeds proactive retention campaigns and segmentation of at-risk customers.

The Australian context

Subscription businesses in Australia, telcos, media streaming and SaaS in particular, rely on survival analysis to time retention campaigns around contract anniversaries and free trial expiries. Australian telcos have used the technique for years to flag customers approaching the end of a twenty-four month contract, since churn risk spikes sharply in that window. Smaller Australian SaaS companies are slower adopters, mostly because the analysis needs a reasonable volume of churn events to produce a reliable curve. A business with only a handful of cancellations a month won't get much signal from a formal survival model and is better served watching retention rate by cohort until volume grows.

Where people get this wrong

Treating churn rate and survival analysis as interchangeable.Churn rate gives one static number, survival analysis reveals how risk changes over the customer lifecycle, which is what tells you when to intervene.
Ignoring censored customers in the dataset.Customers who haven't churned yet still carry information. Dropping them or treating them as retained forever biases the curve.
Running the analysis on too small a sample.Survival curves need enough churn events across enough time periods to be statistically stable. A handful of cancellations won't produce a reliable hazard rate.

Related terms

Common questions

Do I need a data scientist to run survival analysis?

Not necessarily for a basic survival curve, which some analytics tools calculate automatically. Building a full hazard model with multiple variables, like plan type or acquisition channel, usually needs someone comfortable with statistical software and enough churn events to trust the output.

How is survival analysis different from a churn prediction model?

Churn prediction models usually output a single probability, will this customer churn in the next thirty days. Survival analysis outputs a full curve showing the changing probability of churn across the entire customer lifetime, which is more useful for deciding when to act.

What sample size do I need?

There's no fixed number, but analysts generally want at least a few dozen churn events spread across the observation window before a curve becomes reliable. Businesses with very low churn volume are better off tracking retention rate by cohort until enough events accumulate.

Can survival analysis predict lifetime value?

Indirectly. A survival curve gives you the expected time a customer will remain active, which feeds directly into a lifetime value calculation. Combine expected tenure with average revenue per period and you get a more accurate estimate than assuming a flat average lifespan.

Debrief

Get the next one

No spam. No fluff. Just the next article, straight to your inbox.

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 →