How telecom operators can reduce churn with next-best-offer models
Uplift-ranked offers beat blanket discounts on both churn and margin.
Churn is the defining economic problem of telecom. Acquisition is expensive, networks are largely fixed cost, and margin lives in keeping the subscribers you already have. So operators pour money into retention — and much of it is wasted, because it is spent as a blanket discount rather than a targeted decision.
The blanket-discount trap
A retention team sees churn rising and reaches for a broad offer: a discount, bonus data, a free month. It reduces churn on paper, but it also hands margin to two groups who cost you money — subscribers who were never going to leave, and subscribers who will leave anyway. The net effect is often a smaller loss disguised as a win.
The question is never “who will churn?” It is “whose churn can we actually change, profitably?”
Next best offer is a decision, not a promotion
A next-best-offer model ranks, for each subscriber, the specific action most likely to change their behavior in a way that pays off. It combines three signals:
- —Churn propensity from usage, recharge, network-experience, and billing data.
- —Value — ARPU and expected tenure — so effort concentrates where it matters.
- —Uplift — the causal effect of each offer, so you spend only where the offer moves the outcome.
The output is not “discount everyone at risk.” It is “for this subscriber, the plan upgrade changes the outcome; for that one, no offer is the right call.”
Beyond retention
The same engine grows ARPU. Once you can rank offers by expected value, up-sell and cross-sell stop being campaigns and become per-subscriber decisions — recharge nudges timed to behavior, bundles matched to usage, device offers where the economics work. Retention and growth run on one decision layer.
Why it works in practice
Telecom is unusually rich in behavioral data and unusually well-suited to controlled experiments — you can hold out a group and measure real lift. That means an operator can prove the value of next-best-offer decisioning quickly and defensibly, then scale it. It is one of the clearest, fastest AI wins in the sector.
Founder of OKEMA. PhD in Systems & Information Engineering (University of Virginia) and 8+ years in applied ML, causal inference, and reinforcement learning — with production work across Microsoft, Netflix, and DoorDash.