Retention that reaches subscribers before the churn call.
Churn-prediction-driven loyalty for operators — targeted offers per subscriber, measured against holdouts, integrated with charging and the BSS estate you already run.
- Model the programme
- Connect the estate
- Shadow and calibrate
- First campaign with a holdout
- Operate and optimise
Operators learn about churn too late.
An operator learns about churn when the port-out request arrives, and by then the only tool left is a discount. The signals were there weeks earlier — a complaint, a dropped-call cluster on the subscriber’s home cell, a smaller top-up, an app that stopped opening. They lived in four systems that do not talk to each other, and the retention team was working from last month’s list.
Loyalty & Retention for telco closes that loop. Usage, complaints, network events and billing signals feed a member-level risk score. The score drives an offer decision — a data bonus, a device upgrade, a plan move — matched to the subscriber’s value and budgeted in advance. The offer is delivered through charging in real time, and every campaign carries a holdout so the uplift is measured rather than assumed. Prediction to prevention: risk scores alone save nobody; the engine runs the full retention machine, operated.
It fits the BSS estate rather than replacing it. Integration runs through governed contracts with billing, charging and CRM, designed to add retention capability without destabilising the stack underneath. And it is sized and rehearsed for the days that hit hardest — season launches and reload days.
Five modules, configured for operator economics.
Every module reads and writes the same subscriber ledger, so a risk score, the offer it triggered, the data it awarded and the uplift it produced are one record — not four reports that disagree.
Churn-signal ingestion
Usage, complaints, network events and billing signals feed member-level risk scores. Signals arrive as events, are joined on the subscriber and scored regularly with model versions and evaluation metrics on the record.
Targeted retention offers
Data bonuses, device upgrades, plan moves — matched to member value and risk, with budgets. Offer rules are configuration: which score, which segment, which offer, how much budget, and what share goes to the holdout.
Engagement loyalty
Points on tenure, top-ups and app engagement — reasons to stay that are not price cuts. Tenure tiers, birthday and anniversary rewards and app-engagement earn are settings with a version history and simulation before publish.
Dealer alignment
Pairs with Sales Performance so channel incentives and retention pull together. A dealer paid to sell a plan and a retention offer that moves the subscriber off it should not fight; both read the same subscriber record.
Surge rehearsal
Throughput sized to the latency budgets self-care demands. Offer decisions are engineered to a p95 budget agreed with you, and the availability target under campaign surges is agreed with you according to your needs. The surge is rehearsed, not hoped for.
What it does, in detail.
Four systems’ worth of signals, one risk score.
Churn rarely announces itself in one place. Billing sees the late payment. The network sees the dropped-call cluster. CRM sees the complaint. The app sees the silence. The engine consumes each as an event, joins them on the subscriber and scores the probability of leaving in the next window. Scores become segments; segments drive offers. The features behind each score are readable, the model version is on the record, and evaluation metrics are published every scoring run so your team knows when a model has drifted rather than discovering it in the churn number.
- Usage, complaint, network-quality and billing events on one subscriber timeline
- Regular scoring with model versions and evaluation metrics on the record
- Scores exposed as segments to the offer engine, CRM and your warehouse
Offers matched to value, capped by budget, measured by holdout.
A blanket discount to everyone who looks risky spends the retention budget on subscribers who were never leaving. The offer rules read both risk and value: a high-value subscriber at high risk earns a device upgrade, a low-value one a data bonus, and a subscriber who has already had two offers this quarter gets none. Every rule carries a budget line and a holdout share. When the campaign closes, uplift is measured against the holdout — not against last quarter — so the retention number the board sees is the one the programme actually caused. The rules shown here are examples.
- Offer eligibility by score, value band, tenure, plan and recent offer history
- Budgets and per-subscriber caps enforced at decision time, not reconciled afterwards
- Uplift versus holdout on every campaign, on one screen for retention and finance
Example rules. Yours are configured with you.
Accepted in the app, live on the account in the same session.
A retention offer that arrives as a voucher to be redeemed later is an offer half made. The engine integrates with the charging system so that data, minutes or a plan move accepted in self-care are provisioned on the subscriber’s account in the same session, and the confirmation event returns to the ledger. Offer decisions are held to a p95 latency budget agreed with you. The contract with charging is governed and versioned, with retries, idempotency keys and a dead-letter path — so a charging timeout does not award the same bonus twice or lose it.
- Real-time provisioning of data and minutes through charging-system patterns
- Idempotent award with confirmation back to the ledger and to CRM
- Real-time offer decisions inside the self-care flow
- 01Score
- 02Decide
- 03Present
- 04Accept
- 05Provision
- 06Measure
Retention offer lifecycle
Reasons to stay that are not a price cut.
Discounts retain until the next discount. Tenure loyalty gives subscribers something that compounds: points on every month of tenure and every top-up, engagement earn for using the app, and tiers that arrive with benefits — priority care, an annual data gift, partner rewards — rather than another percentage off. The tier ladder and earn rules are versioned configuration, simulated against last year’s subscriber base before publication, so you know what a change costs and who it moves before a single subscriber sees it.
- Points on tenure, top-ups and app engagement, with expiry policy as configuration
- Tenure tiers with benefits catalogue and partner earn-and-burn
- Simulation of ladder changes on historical data before go-live
One retention machine, four seats.
Change the offer without opening a ticket.
Offer rules, budgets, holdout shares and tenure tiers are edited, simulated and scheduled in the console. Every change is versioned and attributable, so the programme moves at campaign speed and stays auditable.
- Simulate an offer rule on last quarter’s scored base
- Schedule campaigns to go live at a date and time
- Uplift, spend and churn on one screen
Retention spend you can account for.
Every offer carries its budget line, cost basis and the campaign that caused it. Spend, awarded value and measured uplift against holdouts are reports over the ledger, so retention is a measured line rather than a cost centre with a story.
- Spend by campaign, segment and offer type
- Awarded data and minutes valued at your cost basis
- Uplift versus holdout, exportable to your planning tools
Governed contracts with billing, charging and CRM.
Your estate publishes subscriber, usage and billing events; the engine consumes them idempotently and publishes decisions and awards back. Kafka topics with versioned schemas, an outbox on every write, replay tooling and OpenTelemetry traces end to end — the integration is built to be predictable.
- REST and event APIs with versioned schemas
- Charging and CRM contracts with retries, idempotency and dead-letter handling
- Sandbox tenant with replayable sample traffic
Incentives and retention that pull the same way.
Paired with Sales Performance, dealer incentives read the same subscriber record as retention. A plan sold on commission and a retention offer that moves the subscriber off it are visible on one timeline, so channel policy and retention policy stop working against each other.
- Dealer incentive and retention offer on one subscriber timeline
- Clawback and retention rules that reference each other
- Dealer portal view of retained subscribers in their base
Where it sits in an operator estate.
Monthly churn list versus Loyalty & Retention.
Patterns for BSS estates.
Billing
Invoice, payment and top-up events for scoring and tenure earn
Charging (OCS)
Real-time award of data, minutes and plan moves with idempotent confirmation
CRM & care
Complaints, interactions and case outcomes as signals; offers surfaced to agents
Network analytics
Quality-of-experience and dropped-call signals per subscriber
Self-care app & web
Offer presentation, acceptance and balance through REST and GraphQL
Sales Performance
Dealer incentives and clawbacks on the same subscriber record
Identity
Subscriber identity across app, web, care and dealer
Data warehouse & BI
Scores, decisions and awards exported to Snowflake, BigQuery or your lakehouse; Power BI models
From first signal to measured uplift.
Model the programme
Risk signals, value bands, offer catalogue, budgets and tenure tiers configured against your current retention design. Scoring models are trained on your subscriber history.
Connect the estate
Billing, charging, CRM and network feeds flow into a sandbox tenant through governed contracts. Your real event shape runs before any subscriber is touched.
Shadow and calibrate
Models score live traffic in shadow; decisions are logged but not delivered until evaluation metrics meet the bar you set. Existing points balances migrate with reconciliation.
First campaign with a holdout
One segment, one offer, a holdout share, and a surge rehearsal before the first reload day. Every step reversible.
Operate and optimise
Model versions, campaign uplift and spend reviewed regularly, at a rhythm agreed with your CVM team; the platform operated to SLOs and incident cover agreed with you according to your needs.
Frequently asked questions
One engine, five verticals.
See it on your own data.
A guided demo with an engineer who operates the platform — configured toward your programme, not a canned deck.