This Indonesian telecom provider served over 42 million mobile subscribers across the archipelago and was losing approximately 2.8 million subscribers per quarter to competitors — a churn rate of 6.7% that was 1.5 percentage points above the industry average. The cost of acquiring a new subscriber was approximately IDR 185,000, while retaining an existing one cost only IDR 45,000, making churn the company's most expensive operational problem at an estimated USD 48 million annually.
The existing churn management approach was reactive and generic. The retention team could only identify churning customers after they had already ported their number or stopped recharging, at which point win-back success rates were below 8%. The company had a basic churn model using logistic regression on billing data, but it had an accuracy of only 52% — barely better than a coin flip — because it could not capture the complex, multi-signal nature of churn behavior in Indonesia's prepaid-dominant market.
Indonesia's telecom market presented unique challenges: 78% of subscribers were prepaid, meaning they could switch operators without any formal cancellation process. Churn signals were subtle — slight changes in recharge frequency, data usage patterns, and call-to-network ratios that preceded a switch by 30 to 60 days. The sheer scale of the subscriber base made it impossible for the retention team to intervene proactively without a reliable prediction system.
Pertama Partners' AI Readiness Audit analyzed three years of subscriber behavioral data encompassing over 8 billion CDR (call detail records), data usage records, recharge transactions, customer service interactions, and network quality measurements. We identified 127 behavioral features that correlated with churn risk, including subtle indicators like the ratio of on-network to off-network calls (subscribers testing a competitor's network), declining data usage on specific cell towers (indicating geographic coverage dissatisfaction), and recharge amount changes.
Our AI Pilot Program built an ensemble churn prediction model combining gradient-boosted trees, recurrent neural networks for temporal pattern recognition, and a separate model for geographic churn hotspot detection. The system generated daily churn risk scores for every subscriber with a 30-day prediction horizon and automatically triggered personalized retention interventions — tailored data bonuses, loyalty rewards, or customer service outreach — based on the predicted churn driver for each subscriber.
The AI Transformation Program scaled the system nationally and integrated it with the company's campaign management platform for automated, personalized retention interventions at scale. Executive Training sessions helped the CMO and CFO establish a dynamic retention budget allocation model that directed spending to the highest-risk, highest-value subscribers. Team Training equipped the retention team to interpret churn drivers and design effective interventions.
"In Indonesia, a subscriber can switch operators in five minutes at any warung. Pertama Partners gave us the intelligence to understand why they want to leave 30 days before they do, and the tools to give them a reason to stay."— Rizal Hakim, Chief Marketing Officer
Every transformation starts with a conversation. Let's discuss your challenges and opportunities.
Discuss Your Challenge