A Malaysian telco just proved that the most lucrative AI use case isn't chatbots or search — it's making your phone bill feel less like extortion.
The Summary
- Circles, a digital telco operator, integrated OpenAI's API and Codex to personalize customer experiences, driving 22% higher average revenue per user and cutting churn by 9%
- The AI handles everything from plan recommendations to proactive issue resolution, running in production across Southeast Asian markets
- Development velocity increased dramatically — features that took weeks now ship in days, suggesting the real AI moat is execution speed, not model access
The Signal
Circles isn't a household name, but it's exactly the kind of company that telegraphs where AI value actually accrues. They're a digital-first mobile network operator serving markets across Southeast Asia, and they just posted numbers that should make every SaaS executive nervous. Twenty-two percent ARPU lift. Nine percent churn reduction. Both metrics moving in the right direction, simultaneously, which basically never happens in telco.
The mechanism matters more than the results. Circles built what they're calling an "AI-native" customer experience layer on top of OpenAI's API. Every customer interaction — plan selection, data usage alerts, billing inquiries, network troubleshooting — routes through models that understand context, predict needs, and take action without human routing. The system uses Codex for internal tooling, letting engineers ship customer-facing features by describing them in natural language rather than writing traditional code.
"Features that took weeks now ship in days, and the AI handles customer personalization at a scale human agents never could."
Here's what's actually happening under the hood:
- Models analyze usage patterns to surface plan upgrades before customers hit overage charges
- Proactive alerts about network issues in a user's area, with automatic service credits applied
- Natural language billing explanations that actually make sense, reducing support volume by addressing confusion before it becomes a complaint
This isn't a chatbot bolted onto legacy infrastructure. Circles rebuilt their customer stack assuming AI could handle the cognitive load that telcos traditionally throw bodies at. The ARPU increase comes from better-timed upsells that feel helpful rather than predatory. The churn reduction comes from catching problems before they metastasize into cancellation decisions.
The development velocity piece is the sleeper story. Circles reports that using Codex collapsed their feature development cycle, which means they're now competing on iteration speed rather than initial infrastructure investment. In a commoditized market like telco, whoever can test and ship personalized experiences fastest wins. Model access is table stakes. Execution velocity is the moat.
The Implication
If you're running a business where customer retention depends on feeling understood, you're now competing against companies that can personalize at Circles' scale and speed. The telco industry was supposed to be too regulated, too infrastructure-heavy, too boring for AI to matter much. Instead it's becoming a case study for how quickly margin and churn metrics move when you rebuild customer experience around language models. Watch for this pattern to spread into insurance, banking, and any other sector where customer service is currently a cost center staffed by humans reading scripts.