Circles cuts churn 9% and lifts ARPU 22% with OpenAI tech
Circles, a Singapore-founded technology company whose SaaS platform helps digital telco operators run AI-powered customer experiences, published a case study describing how it built an AI-native telco stack on OpenAI's API. Founded in 2014, Circles works with telco operators across 14 countries and six continents, and it also runs its own consumer telco business, Circles.Life, in Singapore.
The centerpiece is Circles' AI Concierge, a conversational interface powered by OpenAI's API platform that brings search, account management, support, and personalized recommendations into one experience. Behind it sits CareX, a proprietary multi-agent architecture combining an orchestration agent, which reads a customer's history and current activity, with specialist agents for billing, subscriptions, network management, and account services. When a customer raises an issue, the orchestration agent gathers the app context, account history, and request, then routes it to the specialist agent, which receives only the information needed to resolve that issue; if the case requires it, the AI agent escalates to a human with the full context attached.
CareX now autonomously resolves 65% of customer service interactions across supported workflows, without human intervention. In an early deployment, one operator reached 55% autonomous resolution within its first week. Circles is expanding CareX into real-time voice support and is targeting 95% autonomous resolution across the entire workflow, text and voice combined.
A separate engine, Xplore IQ, also built on the OpenAI API, combines customer behavior, account history, and real-time signals to recommend the next best action, such as a roaming pack, a plan upgrade, or a step to prevent churn. In Singapore, Circles compared customers who received these AI-powered recommendations against those who did not: the AI-powered group showed a 22% increase in ARPU, driven by higher plan upgrades and add-on subscriptions, and a 9% reduction in churn.
An unnamed Circles representative is quoted in the case study: "AI should empower users, not force-fit them into outdated journeys. OpenAI's role has been critical in enabling Circles to scale this vision globally. With the AI concierge, we are moving beyond providing simple answers to delivering real-world outcomes, along with balancing cost and latency to maximize value for operators and customers alike."
Beyond customer-facing products, Circles also uses OpenAI's Codex internally across engineering, infosec, and incident-management teams, applying it to design, coding assistance, and unit testing while keeping specialists responsible for validation and final review. Circles says this has delivered a 29% increase in development efficiency.
On safeguards, Circles says it applies automated controls before customer information reaches the model layer, including identifying and encrypting personally identifiable information before it is passed to LLMs, and that specialist agents get scoped access to only the information needed for a given task rather than a full customer profile. The company also cites human escalation paths, phased rollouts, rate limits, rollback mechanisms, and commercial safeguards as part of keeping operational control as the AI systems scale.
Looking ahead, Circles says it is enhancing CareX's observability and building out real-time voice support, with a longer-term goal of a conversational, context-aware telco experience that moves past traditional app navigation.
Key facts
- In Singapore, Circles' AI-powered personalization (Xplore IQ) raised ARPU by 22% and cut churn by 9%, compared to customers who did not receive AI-driven recommendations.
- CareX, Circles' multi-agent architecture behind its AI Concierge, autonomously resolves 65% of customer service interactions without human intervention; one operator's early deployment hit 55% resolution in its first week.
- Circles is targeting 95% autonomous resolution across the whole workflow, including voice, as CareX expands into real-time voice support.
- Internally, Circles uses OpenAI's Codex for design, coding assistance, and unit testing, reporting a 29% increase in development efficiency.
- Founded in 2014 and based in Singapore, Circles works with telco operators in 14 countries across six continents and also runs its own consumer telco, Circles.Life.
Why it matters
Telcos sit on rich customer signals, usage, billing, network activity, location, but have historically struggled to turn that data into timely, personalized action; support tends to be reactive, with customers repeating themselves across menus. Circles' case study is a concrete, numbers-backed example of what an AI-native stack built on OpenAI's API and Codex can move: ARPU, churn, support resolution rates, and internal engineering speed, all in one deployment.
Who it affects
Directly, it affects Circles' telco-operator customers across the 14 countries it serves and Circles' own consumer telco, Circles.Life, in Singapore, plus the end customers who now interact with an AI Concierge instead of static support menus. Internally, it affects Circles' engineering, infosec, and incident-management teams, who use Codex for design, coding assistance, and unit testing while retaining responsibility for validation and final review.
How to use it
Circles offers the AI Concierge, CareX, and Xplore IQ as part of its SaaS platform for digital telco operators; the case study does not disclose pricing or licensing terms for the platform. Operators adopting it get a conversational interface for search, account management, support, and personalized recommendations, built on OpenAI's API, plus Codex for their own engineering workflows.
How solid is it
This is a case study published on OpenAI's own site, built on figures and a quote supplied by Circles rather than independently audited numbers. No individual is named as the source of the quoted statement, and the article gives no publication date. The 22% ARPU and 9% churn figures are reported for Singapore specifically, not for all 14 countries Circles operates in, and no absolute revenue or customer-count figures accompany any of the percentages.
Risks and caveats
All of the headline figures, ARPU, churn, autonomous resolution, and development efficiency, are self-reported by Circles rather than externally verified. The relationship between the 55% resolution rate from one operator's first week and the platform-wide 65% figure is not spelled out, so it is unclear whether they describe the same deployment over time or different operators entirely. The privacy and access safeguards Circles describes, PII encryption before data reaches the model and scoped agent access, are likewise the company's own description of its controls, not an independent audit.
“AI should empower users, not force-fit them into outdated journeys. OpenAI's role has been critical in enabling Circles to scale this vision globally. With the AI concierge, we are moving beyond providing simple answers to delivering real-world outcomes, along with balancing cost and latency to maximize value for operators and customers alike.”
— unnamed Circles representative, quoted in the case study