A 22% jump in average revenue per user isn’t something telcos stumble into. That’s the kind of number that gets executives rehired and strategies reprinted. Circles, the Singapore-headquartered digital telco operator, just published exactly that result — crediting a deep integration with OpenAI’s API and Codex platform for numbers that most carriers would frame and hang on a wall. Churn dropped 9%. Developer efficiency improved meaningfully. And this wasn’t a proof-of-concept — it’s live, in production, serving real customers across multiple markets.
The telco industry has been trying to “go digital” for roughly fifteen years. Most of those efforts produced slightly better apps and chatbots that couldn’t understand the question being asked. What Circles appears to have done differently is build AI into the product layer itself, not bolt it on as a feature afterward. That distinction matters more than it sounds.
Why Circles and Why Now
Circles launched in 2014 as a mobile virtual network operator (MVNO) with an unusual thesis: run a telco with almost no physical retail, no call centers in the traditional sense, and a software-first stack. They operate across Singapore, Australia, and Japan under various brand partnerships, and they’ve always positioned themselves as the operator for people who’d rather sort out their phone plan through an app than stand in a queue.
That model works fine in normal times. It works exceptionally well when your AI can actually personalize that app experience at scale. The problem Circles faced — the same problem every MVNO faces — is that customer lifetime value is fragile. Price competition is brutal. Churn is the enemy. And upselling feels gross when it’s not relevant.
So the question they were trying to answer wasn’t “how do we add an AI chatbot?” It was closer to: how do we make every customer interaction feel like it was designed specifically for that person, without hiring thousands of people to do it manually?
That’s where OpenAI’s API came in. And specifically, where OpenAI Codex changed the internal development story at the same time.
What Circles Actually Built
The Circles implementation has two distinct tracks that are worth separating, because they solve different problems.
Track One: Customer-Facing Personalization
Using the OpenAI API, Circles built what they’re calling an AI-native telco experience. In practice, this means the platform can analyze a customer’s usage patterns, predict what they’ll need next, and surface relevant plan upgrades, add-ons, or support — proactively, not reactively.
Think about what that actually looks like. A customer who consistently burns through their data allowance in week three of a billing cycle gets a nudge about a higher-data plan before they hit the wall. A customer who calls internationally twice a month sees international add-on options at the right moment. A customer who hasn’t used roaming but is about to travel gets a relevant offer, not a generic one.
This is the kind of personalization that, done badly, feels like surveillance. Done well — and the 22% ARPU increase suggests Circles is doing it well — it feels like a service that actually pays attention.
The churn reduction piece follows the same logic. At 9% reduction, that’s customers who would have left staying put, presumably because the experience improved enough that switching felt unnecessary. In a market where telco churn rates typically run anywhere from 15% to 30% annually depending on the segment, a 9% reduction is genuinely significant.
Track Two: Internal Development with Codex
The second track is about how Circles builds software internally. They’ve integrated OpenAI Codex into their development workflow, and the efficiency gains here are the part that should interest other engineering teams the most.
Codex, for context, is OpenAI’s AI system for code generation and understanding. It’s the engine that originally powered GitHub Copilot before Microsoft and OpenAI diverged their implementations. At Circles, developers are using it to accelerate feature development, reduce time spent on boilerplate, and move faster from idea to deployed feature.
The specific efficiency numbers Circles cited are meaningful because faster internal development creates a compounding advantage. If your team ships personalization features faster than a competitor’s team, you get more data, faster iteration, and better models sooner. In a market where differentiation is hard to maintain, development velocity is a real moat.
Key Outcomes From the Circles Deployment
- +22% ARPU — average revenue per user increase, attributed to AI-driven personalization and timely upsell offers
- -9% churn — customer retention improved across active markets
- Improved dev efficiency — internal teams using Codex for accelerated feature development
- AI-native architecture — AI embedded in the product layer, not added as a surface feature
- Multi-market deployment — live across Circles’ operational markets, not a single-market pilot
What This Actually Means for Telcos — and the Broader Enterprise AI Picture
Here’s the thing: telcos are an interesting test case for enterprise AI precisely because they’re so data-rich and so commoditized at the same time. Every major carrier has years of usage data, billing history, network behavior, support interactions. That data has always theoretically been valuable. The gap has always been in turning it into something actionable at the individual customer level, in real time, without a team of analysts in the loop.
OpenAI’s API gives companies like Circles the natural language understanding layer to bridge that gap. The model can interpret ambiguous customer intent, generate relevant recommendations, and adjust tone based on context — things that rule-based systems and older ML approaches handled clumsily at best.
I wouldn’t be surprised if this case study triggers a fairly serious conversation at every tier-two and tier-three carrier globally. The tier-one carriers — your AT&Ts, your Vodafones — have been running AI personalization experiments for years with mixed results and massive internal inertia. But smaller, more agile operators like Circles can move faster, integrate more cleanly, and actually measure outcomes without six months of governance overhead.
The competitive comparison worth making here is against what carriers typically do: they buy expensive CRM platforms, they license marketing automation tools, they hire consultants to build recommendation engines that take two years to deploy and immediately start going stale. The OpenAI API approach is faster to integrate, faster to update, and arguably more capable out of the box than anything a mid-market telco would build internally.
It also fits a pattern we’ve been seeing across OpenAI’s enterprise push. How Univé built an AI-ready workforce with ChatGPT Enterprise showed a similar dynamic — an organization that moved fast by partnering with OpenAI rather than building from scratch. And avatarin’s 24/7 AI retail agent for Yamada Denki demonstrated how AI-native customer interaction could replace traditional staffing models in retail. Circles is the telco version of that story.
For developers watching from the outside, the Codex angle is worth taking seriously. OpenAI has been pushing Codex hard as a developer productivity tool, and real-world numbers from a production deployment carry more weight than benchmark claims. If Circles is shipping features meaningfully faster as a result, that’s a data point that engineering leaders at other companies can actually use in budget conversations.
As for who loses here — the traditional telco software vendors and systems integrators have reason to be nervous. If a company can build a differentiated customer experience on top of OpenAI’s API in a fraction of the time it would take to customize an enterprise platform, the case for those expensive bespoke solutions gets weaker. As OpenAI continues driving down cost per intelligence unit, the math keeps shifting further in favor of API-first approaches.
Key Takeaways
- Circles achieved a 22% ARPU increase and 9% churn reduction using OpenAI’s API for customer personalization — results that are production-validated, not projected
- The deployment covers both customer-facing AI and internal development tooling via Codex, making it a full-stack AI integration rather than a single feature
- This model — API-first, fast to deploy, continuously improvable — is a direct challenge to legacy telco software vendors who sell slower, more expensive alternatives
- Smaller, agile telcos have a structural advantage in adopting this approach over large incumbents bogged down by internal process
- OpenAI’s enterprise case study portfolio is building a coherent narrative: fast integration, measurable ROI, industry-specific outcomes
Frequently Asked Questions
What did Circles use OpenAI for specifically?
Circles used the OpenAI API to power personalized customer experiences — surfacing relevant plan upgrades, add-ons, and support interactions based on individual usage patterns. They also integrated OpenAI Codex into their internal development workflow to improve engineering efficiency.
Are these results typical for AI deployments in telco?
Honestly, a 22% ARPU lift is on the high end of what’s been publicly reported in the telco AI space. Most implementations report more modest gains or don’t publish numbers at all. That said, Circles’ MVNO model and digital-first architecture likely made them better positioned than a traditional carrier to execute cleanly and measure accurately.
How does this compare to what larger telcos are doing with AI?
Larger carriers like T-Mobile and Vodafone have their own AI personalization programs, often built on a mix of proprietary models and third-party platforms. The difference is speed and flexibility — Circles can update its AI layer as OpenAI improves the underlying models, without a full re-implementation cycle.
Is OpenAI Codex still a separate product?
OpenAI has evolved the Codex offering over time — it’s currently available as part of the OpenAI API with specific code-focused capabilities, and it powers various developer tools. The version Circles is using is the API-accessible Codex that enterprise developers can integrate directly into their build pipelines.
The Circles story will likely look like an early example of something that becomes standard practice across the telco vertical over the next few years. The question isn’t whether AI-driven personalization works at scale — Circles just answered that. The question is which operators move fast enough to build a lead before the approach becomes table stakes.
