> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kaireonai.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Telecom
> Reduce churn, time upgrades, and keep contact under control across app, SMS, and the call center — one coordinated next-best-action per subscriber.
## The problem in this industry
Telecom has the churn problem in its purest form. Contracts end, competitors advertise a better deal, a bill-shock month sours the relationship — and a subscriber who has been with you for years walks in a week. The retention window is narrow and the signals that predict a departure (declining usage, a support escalation, a contract nearing its end) are known but rarely acted on in time. By the time a customer calls to cancel, you're negotiating from behind.
At the same time telecoms have real upside to offer: data add-ons, plan upgrades, device refreshes, family-plan bundles, fiber cross-sell. The challenge is that these live in different teams and different channels — the app, SMS, email, the call-center agent's screen, the IVR — and without coordination a subscriber can get an upgrade SMS on Monday, a retention email on Tuesday, and an upsell prompt from an agent on Wednesday. That incoherence is what makes customers feel like a target rather than a valued account.
KaireonAI unifies the decision. For each subscriber it weighs churn risk against upgrade opportunity, respects a single cross-channel contact budget so the app and the call center aren't working at cross purposes, and surfaces the one next-best-action — whether that's a retention credit, an upgrade, or nothing at all this week.
## What you build in KaireonAI
You express your plans, devices, retention plays, and channel mix as a small set of platform building blocks. Here is how a typical carrier setup maps on:
| Platform concept | How you use it in telecom |
| ------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Offers** | Data add-on pack, unlimited-plan upgrade, device upgrade, loyalty retention credit, family-plan bundle, international/roaming pack, contract-renewal incentive, fiber cross-sell |
| **Channels** | Mobile app, SMS, email, call-center agent desktop, IVR |
| **Decisioning gates** (Eligibility → Fit → Match) | *Eligibility*: `attribute_condition` on `contract_end_days <= 60` for the renewal incentive; `device_upgrade_eligible == true` for a handset offer. *Fit*: `tenure_months >= 12` for the loyalty credit. *Match*: `propensity_threshold` in fit mode so a weak-fit offer is softened rather than dropped |
| **Contact policies** | `customer_total_cap` — the single most important control in telecom: cap *total* marketing touches per subscriber per week across every offer and channel; `cross_channel_cap` so one offer isn't repeated on app + SMS; `cooldown` between retention pitches; `do_not_contact` for opt-outs |
| **Scoring approach** | A [`gradient_boosted`](/ai-ml/algorithms/gradient-boosted) model is the workhorse for churn — usage, tenure, tickets, and billing interact in ways linear models miss. Use a [`scorecard`](/ai-ml/algorithms/scorecard) for a transparent day-one baseline, and a [`thompson_bandit`](/ai-ml/algorithms/thompson-bandit) for fast-learning promotional offers |
Model the retention program as a [journey](/studio/journeys) so a multi-touch sequence (nudge → offer → agent follow-up) is orchestrated coherently, and weight ranking by [customer lifetime value](/studio/clv) so you spend your richest retention credits on your highest-value subscribers.
## A worked example
**Sofia** has been a subscriber for 3 years on a mid-tier plan. Her `contract_end_days` is 45, her data usage has dropped 30% over two months, and she opened a billing support ticket last week. She's device-upgrade eligible. She has already received two marketing SMS messages this week.
Data add-on, unlimited upgrade, device upgrade, loyalty retention credit, family bundle, roaming pack, renewal incentive, fiber cross-sell.
The roaming pack drops (no travel signal). The data add-on drops (her usage is *falling*, not capped). The family bundle drops (single line, no fit). Five survive — including the renewal incentive, which her 45-day contract window unlocks.
A `customer_total_cap` of three marketing touches per week is close to its limit — she's had two SMS already — so lower-priority offers are held back to protect the budget for the one that matters. The recent billing ticket also triggers a short `cooldown` on hard-sell upsells.
The gradient-boosted churn model scores her as high-risk (falling usage + support friction + contract ending). Ranking blends [PRIE](/decisioning/prie-formula-design); the loyalty retention credit ranks first because its impact term is amplified by her churn risk and lifetime value.
Sofia gets a single, well-timed retention offer through her preferred channel — not another upsell SMS that would have pushed her toward the exit.
**A growing subscriber flips the decision.** Run the same flow for a low-churn-risk subscriber whose usage is *climbing* and whose contract has 18 months left, and the retention credit ranks near the bottom — the device-upgrade or unlimited-plan offer wins instead. The churn model and PRIE weighting react to each subscriber's trajectory, so the platform defends the at-risk and grows the healthy from the same configuration.
## Measuring success
* **[Business Dashboard](/operations-reporting/dashboards#business-dashboard)** — acceptance and revenue per offer, and the daily trend of impressions vs. conversions.
* **[Model Health Dashboard](/operations-reporting/dashboards#model-health-dashboard)** — churn-model AUC trend and feature importance; a sudden AUC drop flags feature drift worth a retrain.
* **Retention uplift** — run an [experiment](/ai-ml/uplift-modeling) with a holdout of at-risk subscribers who get no retention contact, and use the built-in z-test to prove the program saves accounts that would otherwise have churned.
* **Contact-budget adherence** — the suppression rate tells you whether the `customer_total_cap` is doing its job of keeping total touches under control.
## Where the agentic layer helps
[Decisioning Autopilot](/ai-ml/autopilot) watches churn-model drift and retention experiments. When usage patterns shift and the live churn model's accuracy degrades, Autopilot proposes a retrain; when a challenger retention offer wins its experiment, it proposes the promotion — reviewable in the inbox or auto-applied through the audited path, your call per tenant.
[Decision Sentinel](/ai-ml/sentinel) is critical when many teams touch config. A retention manager tightening a cap, a campaign manager adding a suppression, and an ops change to an offer schedule can interact to silently zero out sends. Sentinel watches the decision stream and alerts on a suppression spike or empty-decision surge — with optional auto-pause — before an entire subscriber segment goes uncontacted. The [governed AI assistant](/ai-ml/ai-assistant) lets teams make changes in plain language, routed through approvals, so speed doesn't cost you control.
## Try it
Build a churn flow from scratch — risk scoring, retention offers, and a contact cooldown.
Apply the Telecom starter kit for a ready-made setup.
Orchestrate multi-touch retention sequences with a visual flow editor.
Register and build a telecom decision flow end to end.