/// TELECOM · CHURNCustomer Churn Prediction for Telecom
Churn prediction for operators, ISPs and MVNOs that spots customers about to leave — including the prepaid ones who never say goodbye — and aims retention offers where they change the outcome.
Message on WhatsApp
TRUSTED BY TEAMS
Telecom churn has causes other industries do not: a network that drops calls where a customer lives, a bill that jumps without warning, a contract quietly ending, or a prepaid SIM that simply stops being topped up.
Sodio builds customer churn prediction for operators, ISPs and MVNOs that reads those signals, scores every customer, and — just as importantly — works out which customers a retention offer would actually keep, so the budget is not spent on people who were never going to leave.
/// IS IT WORTH BUILDING YET?
When Churn Prediction Is Worth Building
A churn score is only useful if someone acts on it, and only accurate if the model can see the signals that matter.
Worth building when
- A large customer base, where small improvements in churn are worth a lot
- A retention team or offers exist to act on the scores
- Network, billing and support data can be brought together
- Retention campaigns can be tested against a control group
Not yet, when
- A small base where the team already knows its at-risk customers
- There is no way to act on a warning
- Each sits in a system nobody can reach
- Every at-risk customer must receive the same treatment
/// WHAT IT COVERS
What Telecom Churn Prediction Covers
Postpaid churn scoring
Prepaid churn detection
Network experience signals
Bill shock alerts
Contract and device timing
Switching signals
Uplift targeting
Next best action
Campaign measurement
Churn reasons
/// THE SIGNALS
The Signals That Come Before a Customer Leaves
///WHO TO SAVE
Who to Save, and Who to Leave Alone
Knowing who is likely to churn is only half the problem. The other half is knowing whose decision an offer would change. Reducing churn in telecom cheaply means aiming retention at the right people, and uplift modelling sorts at-risk customers into four groups.
Persuadable — will stay only if you act
Target them. This is where retention spend pays off.
Sure things — will stay anyway
Leave them. A discount here is money given away.
Lost causes — will leave whatever you do
Do not spend on them; learn why they left.
Sleeping dogs — a retention call would prompt them to leave
Do not contact them at all.
Retention ROI Focus
The goal is not to predict churn. It is to spend the retention budget where it changes the outcome.
Target them. This is where retention spend pays off.
Leave them. A discount here is money given away.
Do not spend on them; learn why they left.
Do not contact them at all.
The goal is not to predict churn. It is to spend the retention budget where it changes the outcome.
/// FAQ
Frequently Asked Questions
It is using a customer's history — usage, bills, network experience, support contacts, contract dates — to estimate how likely they are to leave, early enough to do something about it. A churn prediction model produces a score for each customer and the reasons behind it.
Postpaid customers leave at a visible moment: the end of a contract, or a request to move their number. Prepaid customers usually leave silently — they stop topping up and the SIM goes quiet. So prepaid churn has to be defined by inactivity and detected from falling usage, well before the account lapses.
Network experience is often the one other industries do not have: customers who suffer dropped calls or slow data where they live and work are far more likely to leave. After that come bill shocks, contracts ending, devices due for upgrade, repeated support contacts and — strongest of all — a request to switch provider or port a number.
They will, if they are aimed at everyone likely to churn. Uplift targeting avoids that by estimating who will change their mind because of an offer. Customers who would stay regardless, and those who will leave whatever happens, are left out — and so are the ones a retention call might actually prompt to leave.
With control groups. A share of at-risk customers receives no retention action, and churn in that group is compared with churn among customers who were contacted. The difference is the real effect of the programme.
No. The same method works for any subscription business. Telecom has particular signals — network experience, prepaid inactivity, switching codes — which is what this page covers.
/// RELATED SERVICES
Related Industries and Services
Routes where scope overlaps with telecom churn prediction.
/// GET STARTED
Start With Last Quarter’s Leavers
Share a quarter of customer data, including who left. We will assess, free, how many of those customers could have been flagged in time, and how many a well-aimed offer might have kept.