Circles, the Singapore-based telco platform company, has published results from the AI concierge it built on the OpenAI API[1]. The system resolves 65% of customer support interactions without human involvement, and at Circles.Life in Singapore it has lifted average revenue per user (ARPU) by 22% while cutting churn by 9%. The goal is to replace the familiar mobile-subscriber experience of navigating menus and repeating the same explanation with a single conversation.
CareX resolves 65% of queries without a human
At the center of the AI concierge is CareX, Circles' proprietary multi-agent architecture[1]. An orchestration agent tracks each customer's history and current activity, with specialist agents underneath it covering billing, subscriptions, network management, and account services.
What separates it from a standard chatbot is that it does not stop at producing an answer. When a billing question arrives, the orchestration agent combines the customer's in-app context, account history, and the request itself, then routes it to the right specialist agent. That agent receives only the information needed to handle the case. When the request cannot be completed, it escalates to a human operator with the full context attached.
The trajectory of the numbers has also been made public. In an early deployment, one operator reached a 55% autonomous resolution rate within the first week[1]. CareX today closes 65% of interactions across supported workflows without human intervention.
Two-thirds of inquiries handled without a person is enough to reshape call center staffing from the ground up. That said, the figure is qualified as applying to supported workflows, so it should not be read as a share of all inquiries.
Xplore IQ pushed ARPU up 22%
The second pillar is Xplore IQ, which combines customer behavior, account history, and real-time signals to identify the most relevant next action[1]. That covers roaming pack suggestions, plan upgrades, and heading off churn before it happens.
The way the effect was measured is one of the more notable parts of this announcement. In Singapore, Circles compared customers who received AI-driven recommendations against those who did not. The group receiving recommendations took up more plan upgrades and add-on subscriptions, producing a 22% increase in ARPU. Timely, relevant recommendations also brought churn down by 9%[1].
Telecom operators hold large volumes of customer data across usage, billing, network activity, and location. Circles' view is that the hard part has always been converting that data into timely action[1]. Replacing broad campaigns aimed at everyone with recommendations tailored to each customer is what produced the gap.
Codex improved development efficiency by 29%
Circles also uses OpenAI's Codex internally, not just in its customer-facing products[1]. Engineering, infosec, and incident management teams applied Codex to design, coding assistance, and unit testing, and report a 29% increase in development efficiency. Validation and final review remain the responsibility of human specialists.
Personal information is encrypted before it reaches the model
Building AI for telecom operators means handling sensitive information. Circles says it applies automated controls before customer data reaches the model layer[1]. Specifically, personally identifiable information is detected and encrypted before it is passed to large language models (LLMs).
Access granted to specialist agents is also scoped. Rather than exposing an entire customer profile, the design passes only the range required to resolve the case at hand. Circles combines this with human escalation paths, phased rollouts, rate limits, rollback mechanisms, and commercial safeguards to maintain operational control.
Japan's povo already runs on the Circles platform
Founded in 2014, Circles provides a SaaS platform that telecom operators use to launch and run digital brands[1]. It also operates Circles.Life, its own consumer telco business in Singapore, and works with partners across 14 countries and six continents. Publicly named partners include KDDI, Etisalat Group (e&), AT&T, and Telkomsel[2].
The KDDI relationship is the one closest to home for readers in Japan. KDDI Digital Life, a KDDI subsidiary, adopted the Circles cloud platform Circles-X to run povo, the brand built around a pricing structure that starts at zero yen for the base plan[3]. Consolidating functions that were traditionally built separately, such as the BSS handling customer management and billing and the OSS handling network operations, into a single cloud platform with data in one place was cited as the reason for the choice.
The AI concierge is a new layer sitting on top of that foundation. There has been no announcement that povo will adopt the feature directly, but with the same company's platform already running in the Japanese market, this is not entirely a story about somewhere else.
Voice is next, with a 95% target
Circles plans to strengthen observability across CareX and extend the platform into real-time voice support[1]. Across the full workflow spanning text and voice, it is targeting a 95% autonomous resolution rate. Many customers still reach for the phone first when something goes wrong, so whether that entry point can be automated is the next question.
The AI concierge itself was announced jointly by Circles and OpenAI on April 30, 2026, and both companies describe it as a multi-year collaboration[2]. What OpenAI has now published is a summary of the measured results that came out of running it since then.
Summary
Circles built its AI concierge on the OpenAI API and has published two sets of results: CareX, which resolves 65% of customer support interactions without human involvement, and Xplore IQ, which lifted ARPU by 22%. Personal information is encrypted before it reaches the model, and specialist agents receive only the scope they need. The company's platform is already used by povo in Japan, and the next focus is real-time voice support and the 95% autonomous resolution target.
出典:https://openai.com/index/circles
出典:https://circles.co/in-the-loop/ai-native-telco-stack-carex-xplore-iq-openai-partnership
