Customer Churn Predictive Analytics

Know who is leaving 68 days before they go.

A risk score for every customer, the early-warning signals unique to your business, the revenue at stake quantified, and a retention playbook that tells your team exactly who to call first.

Request a proposal Book a call

0.87

Typical model AUC

68days

Median warning window

5–7wk

To first scored list

5×

Lift in top decile

Why it matters

Losing a customer is expensive. Not knowing why is worse.

By the time someone formally cancels, the decision was made weeks earlier. The signals were in your own data the whole time — declining usage, a complaint that went cold, a missed payment.

Churn modelling learns the specific pattern that precedes departure in your business, then scores every active customer against it. You get a ranked list, the reasons behind each score, and the revenue at risk — which turns retention from a vague worry into a weekly call list.

Request a proposal Book a call

Churn Risk Register $506k at risk

MODEL ACCURACY

AUC 0.87

WARNING WINDOW

68 days

Critical — act this week

84 customers $196k

High risk

217 customers $310k

Watch

596 customers $402k

Healthy

3,412 customers —

TOP SIGNAL THIS MONTH

Two consecutive months of declining usage plus one unresolved complaint — 4.2× more likely to churn.

Method

How we run it

Define

Agree exactly what churn means in your business and assemble the historical data.

Engineer

Build behavioural features from transactions, usage, complaints and payments.

Model

Train and validate against a held-out period. Accuracy and lift reported honestly.

Deploy

Scored risk register, driver explanations and a tiered retention playbook.

Deliverables

What you receive

Risk score per customer

Every active customer scored 0–100 with a risk tier, refreshed on the cycle you choose.

Why each one is at risk

The specific factors driving each score, so the retention call is informed rather than generic.

Revenue at risk

The value attached to each tier, so you can size the retention budget against real exposure.

Retention playbook

Different intervention per tier, with scripts, offers and escalation rules.

Model performance report

AUC, lift curve, feature importance and limitations — documented for your analysts.

Holdout measurement

A control group design so you can prove the retention programme actually paid back.

Typical investment: US$5,500–16,000 for the initial model depending on data condition and number of products, plus optional monthly rescoring from US$540.

Where it works

Sectors we have modelled

Financial services

Account dormancy, loan non-renewal and deposit flight.

Telecoms & ISPs

Prepaid usage decline, plan downgrades and port-out risk.

Retail & FMCG

Purchase-cycle gaps and basket shrinkage in loyalty programmes.

Membership & insurance

Lapse prediction, renewal risk and premium sensitivity.

Book a free consultation

Also in Insights & Diagnostics

Pair it with another lens

Stakeholder Satisfaction Surveys

Measure perceptions, expectations and satisfaction across every stakeholder group.

Explore

Corporate Feedback Analytics

Turn customer, employee and partner feedback into live dashboards and decisions.

Explore

Sentiment Analysis

AI and NLP across surveys, reviews and social media to read true public perception.

Explore

FAQ

Questions we get asked

At least 18–24 months of transaction or usage history and a record of who left and when. Without known churners the model has nothing to learn from — if that is your situation we start with a definition exercise and basic segmentation instead.

We report AUC and lift honestly against a held-out test period. Typical performance is AUC 0.80–0.90, which in practice means the top risk decile contains four to six times more churners than random selection. If a model underperforms we say so rather than shipping it.

That is the first question we settle, and it matters more than the algorithm. For subscriptions it is cancellation; for retail it is a gap longer than the customer's normal cycle; for lending it is non-renewal. We define it with you in writing before modelling.

Not necessarily. Many clients start with a monthly scored list delivered as a file or dashboard. Where you want live scores inside your CRM we can deploy the model there, which is a larger integration.

The playbook is part of the deliverable. Different tiers get different interventions — critical accounts get a human call, high risk gets a targeted offer, watch tier gets a service fix. We also design a holdout group so you can prove the retention effort worked.

Get started

Ready to find out what they really think?

A 30-minute call is enough to scope the study, agree the sample and give you a fixed price. No obligation, and we will tell you if a cheaper method would do.

Request a proposal Talk to Sales

No obligation · Response within one business day · NDA on request