CS-16
APAC Telco Group — AI Churn Intelligence at Scale
A pan-APAC telecommunications group operating across Singapore, Malaysia, Australia, Thailand, and the Philippines — with 24M postpaid mobile subscribers — was experiencing accelerating churn driven by aggressive competitor pricing, 5G coverage gaps, and reactive customer management that only engaged at-risk customers after they had already initiated port-out. Annual postpaid churn of 2.8% was costing the group $248M in lost annual recurring revenue. Anicalls' AI Churn Intelligence Agent transformed the group's retention programme — reducing churn by 38% and retaining $94M in annual revenue.
38%Churn Reduction
$94MRevenue Retained
45 daysAvg Churn Prediction Lead
340%Retention ROI
Business Challenge
Why Reactive Retention Was Losing the Churn Battle
2.8% Annual Postpaid Churn
Annual postpaid churn of 2.8% across 24M subscribers equated to 672,000 customers leaving annually — at an average revenue per user of $369/year, a $248M annual revenue loss. Churn was accelerating quarter-on-quarter as new MVNO entrants and incumbent competitors launched aggressive SIM-only plans 30–40% cheaper than the group's bundled postpaid offerings. Retaining existing customers required targeted, personalised interventions — not the mass promotional campaigns that were the previous retention strategy.
Reactive Retention: Too Late to Save
The group's retention team only engaged customers after they had called to port-out their number — at which point the decision to leave was typically irreversible. Port-out call save rates were 18% — meaning 82% of customers who called to leave actually left despite the retention agent's best efforts. The average customer's churn decision was made 6–8 weeks before they called to port-out: without 45-day predictive capability, the retention programme was structurally too late.
Retention Spend Inefficiency
The group spent $42M annually on retention incentives — handset upgrade discounts, plan price reductions, loyalty points, and data bonus offers — distributed largely to customers who were not genuinely at-risk of churning. Without churn propensity scoring, offers were sent to the full postpaid base (mass SMS campaigns, in-app promotions) with <4% response rates. High-value at-risk customers received the same generic offers as low-risk customers — wasting $34M of the $42M retention budget on customers who would have stayed anyway.
5-Market Regulatory and Data Complexity
Operating across Singapore (PDPA), Malaysia (PDPA MY), Australia (Privacy Act + CDR), Thailand (PDPA TH), and the Philippines (DPA) meant five different regulatory frameworks for customer data use in marketing and retention programmes. The group's previous churn model was built on Australian customer data only — failing to account for the different usage patterns, price sensitivity, and competitor dynamics in each of the other four markets. A Singapore churn model built on Australian data produced 40% false positive rates.
Solution Delivered
Anicalls AI Churn Intelligence Agent — 5-Market APAC Deployment
Prediction
45-Day Churn Prediction Models
Market-specific churn prediction models — one per country, trained on local customer behaviour, competitor pricing history, and regulatory data constraints — predict individual churn probability with 87% accuracy and a 45-day warning lead time. Models incorporate 280+ signals per customer: usage trend changes, network complaint history, handset age, contract term, social network churn contagion (when a customer's contacts port-out, their own churn risk rises 3.4×), and price sensitivity scoring.
- 5 country-specific churn models (280+ signals)
- 87% prediction accuracy / 45-day lead time
- Social network churn contagion modelling
- Price sensitivity and competitor switch modelling
Retention AI
AI-Personalised Retention Intervention
AI recommends the optimal retention intervention for each at-risk customer — from the group's retention toolkit of 34 offer types (plan upgrades, handset offers, loyalty rewards, data bonuses, bill credits, and competitor price matches). Intervention timing, channel (SMS, app push, outbound call, in-store), and offer value are all AI-optimised per customer — maximising save rate while minimising retention cost per saved customer. Save rate improved from 18% (reactive port-out calls) to 64% (AI-predicted proactive outreach).
- 34-offer-type AI recommendation engine
- Channel, timing, and offer value optimisation
- 18% → 64% save rate improvement
- Cost-per-save: $62 (vs $184 reactive)
Compliance
5-Market Regulatory Compliance Layer
Built-in regulatory compliance enforcement for all 5 markets — PDPA Singapore, PDPA Malaysia, Australian Privacy Act (APP 7 direct marketing), Thailand PDPA, and Philippines Data Privacy Act. Marketing consent status checked in real time before every customer contact; DNC register compliance enforced; data retention periods enforced per market. The compliance layer reduced regulatory marketing complaints by 94% across all 5 markets in Year 1.
- Real-time consent status check before every contact
- 5-market DNC register integration
- PDPA / Privacy Act / DPA compliant data use
- 94% reduction in marketing regulatory complaints
AI Workforce Deployment
The Anicalls AI Retention Operations Team
AI Churn Scoring Agents
15 AI Churn Scoring Agents process daily churn scores for all 24M postpaid subscribers across 5 markets — refreshing scores nightly with new usage data, complaint events, and competitor activity signals. Each agent monitors 1.6M subscribers, generating daily ranked at-risk lists for the retention teams with predicted churn probability, key churn drivers, and recommended intervention type. The daily ranked lists replaced the group's previous monthly batch churn model updates — enabling 30× more responsive intervention timing.
AI Retention Outreach Agents
AI-powered outbound retention calls — using conversational AI for initial customer outreach in English, Mandarin, Malay, Thai, and Filipino — handle Tier 1 retention conversations for medium-risk customers, transferring to human retention agents only when the customer requests a human or when the AI detects a complex negotiation scenario. The AI handles 8,400 retention conversations daily, freeing human retention specialists to focus exclusively on the highest-value, most complex at-risk customers.
AI Retention Optimisation Agents
Dedicated AI Optimisation Agents continuously A/B test retention interventions — varying offer type, channel, timing, and message framing — and update intervention recommendations in real time based on observed save outcomes. The feedback loop reduces average time-to-learning from quarterly (previous model refresh cycle) to daily, enabling the retention programme to adapt to competitor pricing changes within 48 hours of detection — versus the previous 3-month cycle for model retraining.
Technologies Used
The AI Technology Stack Deployed
Platform
Telco Churn Intelligence Platform™
Purpose-built telecom churn prediction platform — pre-integrated with major APAC BSS/OSS systems (Amdocs, Ericsson BSCS, Huawei BSS) and telecom-specific data sources (CDRs, network quality events, handset inventory, competitor tariff intelligence). Data processed within each country's borders in compliance with local data sovereignty requirements. Nightly model refresh pipeline processes 340M CDR events daily across 5 markets.
- Amdocs / Ericsson BSCS / Huawei BSS integration
- 340M CDR events processed daily
- Country-level data sovereignty compliance
- Competitor tariff intelligence integration
ML Models
Gradient Boosting Churn Ensemble
XGBoost and LightGBM ensemble models per market — trained on 36 months of churn history, incorporating 280+ features per customer including usage trends, network experience quality scores, competitor switch events, social network churn contagion, and macroeconomic signals (unemployment trends correlated with price sensitivity in Malaysia and Philippines markets). SHAP explainability outputs provide retention agents with the top 3 churn drivers for each customer conversation.
- XGBoost + LightGBM ensemble per market (5 models)
- 280+ features per customer per daily refresh
- Social network graph churn contagion signals
- SHAP churn driver explainability for agents
Conversational AI
5-Language Retention Conversational AI
Multilingual conversational AI for outbound retention calls and inbound port-out intercept — trained on successful and unsuccessful retention conversations from the group's own historical call recordings. Supports English, Mandarin (Singapore/Malaysia variants), Malay, Thai, and Filipino. Real-time offer authorisation: the AI can authorise offers within pre-approved value thresholds without human approval, enabling same-call resolution for 74% of AI-handled retention conversations.
- 5-language conversational AI (APAC dialects)
- Real-time offer authorisation (pre-approved thresholds)
- Trained on group's own retention call recordings
- 74% same-call resolution rate
Quantified ROI
The Revenue Impact at 18 Months
$94M Revenue Retained Annually
The 38% churn reduction — from 2.8% to 1.73% annual postpaid churn — retained 255,000 additional subscribers annually at $369 average ARR: $94M in revenue retained. The retention programme cost $22M annually (Anicalls platform, AI operations, retention offer budget at reduced cost-per-save), producing a 340% ROI. Importantly, retention offer spend fell from $42M to $28M despite saving more customers — because AI targeting eliminated waste on non-at-risk customers.
Retention Cost-per-Save: $62 vs. $184
The AI-driven proactive retention programme saved customers at $62 per save — versus $184 per save in the previous reactive port-out intervention programme (higher offer values required to save customers who had already decided to leave). The $122 cost-per-save reduction across 255,000 additional saved customers produced $31M in retention efficiency savings — reducing the total cost of the retention programme by 34% while saving 3× more customers.
ARPU Uplift from Personalised Offers
AI-personalised retention offers — recommending the right plan upgrade rather than reactive bill credits — produced an average 8.4% ARPU uplift among saved customers versus their pre-retention-intervention plan. Customers offered personalised plan upgrades (AI-predicted as high-value retention drivers) accepted 3.1× more frequently than those receiving generic price discounts. The ARPU uplift generated an additional $12M in annual revenue from the retained customer base — partially funded by the reduction in discount-based offer costs.
Business Outcomes
Strategic Telecom Transformation
Market Share
Postpaid Market Share Recovered
The churn reduction reversed the group's postpaid subscriber decline — moving from a net subscriber loss of 180,000 per quarter to a net gain of 42,000 per quarter within 12 months. Postpaid market share (which had been declining for 6 consecutive quarters) stabilised and recovered 1.4 percentage points across the 5-market footprint. The turnaround was cited in the group's annual report as a key strategic achievement, with the board attributing the change to the AI-driven churn prevention programme.
Network
Network Investment Prioritisation
Churn analysis revealed that 34% of high-risk churners in specific postcodes were motivated by network quality complaints — enabling the network team to prioritise 5G rollout in the 12 highest-churn-risk coverage areas. Targeted 5G upgrades in those areas reduced churn by 51% among customers with network-quality churn drivers — demonstrating that the AI churn data was also a valuable network planning tool. Network investment ROI improved 28% through AI-prioritised capital allocation.
Regulatory
Marketing Compliance Leadership
The 94% reduction in regulatory marketing complaints — driven by real-time consent enforcement and DNC register compliance — positioned the group as a regulatory leader in APAC telecom marketing practices. IMDA (Singapore) and MCMC (Malaysia) cited the group's consent management framework as a model for industry adoption. The improved regulatory standing contributed to a favourable spectrum allocation decision in Malaysia — commercially valued at $85M in network capacity advantage over the 10-year licence period.
Executive Testimonial
"Our retention programme was fighting the wrong battle — trying to save customers who had already left emotionally. Anicalls gave us the ability to have the right conversation 45 days before a customer decided to leave — when there was still time to change the outcome. The 38% churn reduction speaks for itself. What surprised us was the ARPU improvement: customers we retained with personalised plan upgrades spent more with us, not less. That was not in the business case — it was a bonus."
Metrics Dashboard
18-Month Performance Scorecard
1.73%Annual Churn (was 2.8%)
$94MRevenue Retained Annually
87%Churn Prediction Accuracy
45 daysPrediction Lead Time
64%Save Rate (was 18%)
$62Cost per Save (was $184)
+8.4%ARPU Uplift on Saved Customers
340%Programme ROI
Predict Churn 45 Days Before It Happens
See how the Anicalls Telco Churn Intelligence Platform can reduce your postpaid churn, lower your cost-per-save, and grow ARPU among retained customers across your APAC markets.