- A next best offer model ranks every eligible plan, device, add-on and loyalty offer for each subscriber by likelihood to accept and by value to you.
- Propensity, churn and lifetime-value scores work together, so a subscriber at risk of leaving gets a retention offer and a loyal one gets an upgrade.
- AT&T runs churn, propensity and next-best-offer models in production.
- Scores feed the campaign, care and retail systems you already run; none of them is replaced.
The same offer goes to everyone
Your subscribers see the same promotion in the app, in the store and on the care call. Some take it, many ignore it, and some take a discount they never needed. Your average revenue per user moves slowly, and margin leaks into offers that did not change anyone's decision.
You already hold the data to do better: usage, plan and device history, billing, network experience and every contact with care. It sits across billing, CRM and network systems, so each channel builds its own list and no one sees the whole subscriber. A score is only worth what the channel does with it: if the offer arrives after the contract renews, or goes to someone about to leave, the model was right and the revenue still did not come.
How telecom next best offer works
Step 1: Start from the subscriber record
Join usage, plan, device, billing, network quality and care contacts into one view per subscriber, with the offers each one is eligible for.
Step 2: Predict who will accept what
H2O Driverless AI trains a classification model per offer family, with automatic feature engineering on usage trends, contract dates and device age.
Step 3: Decide between growing and protecting
Combine each offer's acceptance score with the subscriber's churn risk and expected value, so the system does not upsell someone who is about to leave.
Step 4: Rank the next best offer
Apply your eligibility, margin and contact rules, then rank what remains; a recommendation picks one offer and the best time and channel to make it.
Step 5: Explain and write the message
Reason codes show why each offer was chosen, and a generative assistant drafts the channel message from that reason for your team to approve.
Step 6: Deploy, measure and monitor
The models serve your campaign and care systems and track model drift, and hold-out groups show the revenue each offer adds.
Offer models in production at AT&T
AT&T is an H2O customer and runs churn, propensity and next-best-offer models in production. AT&T has not published an acceptance-rate or revenue figure for these models, so your own result depends on your offers, your channels and your data, which is why the first step measures against a hold-out group.