Postpaid Churn Prediction at an APAC Telco — 38% Reduction in High-Value Churn

MindMap deployed a retention platform composed of Churn Predictor (Cp - from the CX category) as the churn-prediction engine, Customer 360 (C3) as the unified subscriber-profile layer, Customer 360 (C3) as the retention-campaign-optimisation layer, and Churn Predictor (Ch) as the subscriber-value-modelling layer.

Phase one was the unified-subscriber-profile build. The previous churn-prediction work had been hampered by data fragmentation — the subscriber's usage data sat in the BSS, the service-interaction data sat in the CRM, the billing-and-payment data sat in the billing system, the network-experience data sat in a separate analytics warehouse. Customer 360 unified these into a single subscriber profile refreshed near-real-time, with the feature-engineering layer producing the structured features the churn-prediction model required.

Phase two was the churn-prediction model build. The model is a multi-horizon ensemble — a 7-day-horizon model for the imminent-churn cases requiring immediate retention intervention, a 30-day-horizon model for the planning-horizon retention activity, and a 90-day-horizon model for the strategic-retention investments. Features include usage-pattern features (declining usage as a churn precursor), service-interaction features (specific interaction patterns as churn precursors), network-experience features (degraded experience as a churn driver), competitive-pressure features (signals of comparison-shopping behaviour), and lifecycle features (contract-renewal proximity).

Phase three was the retention-campaign-design build. Campaign Analyzer takes the model's predictions and produces per-subscriber recommended retention actions — the specific offer (rate-plan change, device subsidy, value-added-service inclusion), the optimal channel for the offer (outbound call, SMS, in-app notification, retail-store invitation), and the optimal timing. The recommendations are A/B-tested continuously through the campaign-management layer.

Phase four was the operational-workflow integration. The retention team's daily workflow now starts with the platform's prioritised retention-action list for the day — the subscribers in highest churn risk with the highest expected retention return on intervention. The previous segment-campaign workflow has been replaced by a per-subscriber action-based workflow.

The platform runs on the operator's private cloud with full local data-residency. Subscriber data — substantial PII exposure — stays inside the operator's perimeter.

The unified-subscriber-profile layer uses Customer 360 with streaming data integration from the operator's BSS, CRM, billing platform and network-experience analytics. The profile schema includes approximately 1,200 distinct features per subscriber, refreshed near-real-time for the high-velocity features (usage, network-experience) and on a daily cadence for the slower-velocity features (billing-and-payment, account-status).

Churn Predictor's multi-horizon model uses an ensemble of gradient-boosted trees (for the cross-sectional churn-driver features) and a temporal-fusion-transformer (for the trajectory features — declining usage trends, deteriorating network-experience trends). The training corpus is the operator's full historical subscriber lifecycle data — approximately 28 million subscriber-month observations.

Campaign Analyzer handles the retention-action optimisation. For each at-risk subscriber, the recommended retention action is selected from the operator's action library based on the subscriber's predicted churn drivers, predicted offer-acceptance probability, predicted retention impact and the operator's working-capital constraints on retention spend. The recommendations are A/B-tested in production to drive continuous improvement.

LTV Calculator provides the per-subscriber lifetime-value modelling that prioritises retention spend. The operator's retention-spend budget is finite; LTV Calculator's per-subscriber value scoring drives the spend prioritisation toward the subscribers where retention investment has the highest expected return.

Integration with the operator's existing campaign-management and customer-interaction tools (the outbound-dialler, the SMS-gateway, the in-app-notification platform, the retail-store CRM) is via each tool's standard inbound API.