Circles with OpenAI reports 22% ARPU lift and 65% autonomous telco support

Singapore-based Circles published an OpenAI case study: an AI Concierge on the OpenAI API is said to raise ARPU by 22%, cut churn by 9%, and resolve 65% of support interactions autonomously via multi-agent CareX at Circles.Life Singapore. Internally, Codex is credited with a 29% gain in development efficiency.

Circles sells a SaaS platform for digital telco operators and also runs its consumer brand Circles.Life. The post says it works with operators across 14 countries on six continents. The story is not a new model. It is support, personalization, and monetization wired into one conversational layer.

CareX stops chatbotting and starts closing tickets through specialists

Classic telco support is reactive: menus, repeated account details, queues. CareX is described as an orchestrator that holds customer history and live context, then hands only the needed slice to specialist agents for billing, subscriptions, network, and account work.

In an early deployment one operator reportedly hit 55% autonomous resolution in the first week; Circles now claims 65% across supported workflows without a human. With real-time voice, the stated target is 95% autonomous resolution of the full workflow. That is a vendor ambition, not an independent audit.

Xplore IQ turns campaigns into a next best action on a single account

Xplore IQ is meant to pick the next action from behavior, account history, and real-time signals: a roaming pack, a plan upgrade, a churn intervention. Circles compared customers with AI recommendations to those without; the AI group shows 22% ARPU via upgrades and add-ons and 9% lower churn.

For telco product teams the point is larger than a smarter chatbot. Operators sit on usage, billing, network, and location signals, yet rarely convert them into a timely offer without a manual campaign. If those metrics hold outside the lab, the AI P&L story shifts from support cost-cutting to a revenue engine.

Singapore vendor numbers are not yet a global stack

Every hard figure comes from Circles and OpenAI materials, mainly Singapore. Baseline, experiment length, and whether the control group sat on the same product remain opaque. The 65% autonomous rate applies to supported workflows, not necessarily the whole contact center.

Governance matters more to enterprise buyers than the ARPU headline: PII should be identified and encrypted before the model layer, agents get scoped access per task, plus human escalation, rate limits, and rollback. Without that, a multi-agent telco stack is a compliance nightmare dressed as a product.

What must show up outside the Circles and OpenAI press kit

Watch three signals. First: whether independent operators outside Circles.Life print similar ARPU and churn lines. Second: whether 65% autonomy survives voice, not only a narrow text happy path. Third: whether scoped PII and audit trails clear a European buying committee, not just an APAC demo day.

Lilith's verdict

OpenAI is not selling a smarter FAQ. It is selling the telco a lever on the customer account while the phone is still in their hand.