Ringg Says Its AI Agents Resolve Up to 65% of Routine Customer Inquiries

Ringg says its AI agents handle more than 7 million connected calls a month and resolve up to 65% of routine customer inquiries without a human. A new OpenAI customer case study puts those results alongside the machinery behind them: models that interpret requests, software that carries out tasks, and a route to a person when needed.

A call has to become a completed task

Ringg’s platform works across voice, chat, WhatsApp and the web. Answering a question may require finding a policy or account record. Finishing the request may mean booking an appointment, updating a customer-management system or moving the conversation to a specialist. Ringg’s software connects the agent to those systems and can pass a summary to a human when a case needs escalation.

That distinction matters for the headline figures. Connected calls measure volume; the reported resolution rate applies to routine inquiries, not every call. The case study does not explain how Ringg classifies an inquiry as routine. Ringg also reports an average customer-satisfaction score of 4.8, but the case study does not supply enough detail to compare that score with the resolution rate.

The model depends on the job

Ringg does not send every request to its newest model. GPT-4.1 handles most live voice and chat traffic; GPT-5.6 Luna handles requests where its performance, speed or cost is a better fit. GPT-5.6 Terra analyzes calls afterward, including summaries and sentiment classification. GPT-5.6 Sol supports testing and prompt improvement. Ringg says moving selected live workloads from GPT-4.1 to Luna cut model costs by approximately 90% while meeting its quality and response-speed needs. That is a result for selected workloads, not a platform-wide savings figure.

For a live request, Ringg brings together the customer’s message, conversation history, account information, relevant documents and the tools the agent can use. A routing layer chooses the model and settings; Ringg’s software then carries out the chosen actions and returns a response through the customer’s channel. For a long exchange, the system creates a structured summary as the conversation approaches roughly 80,000 tokens, preserving key context without repeatedly sending the full history.

What customers are reporting

The case study gives three customer examples, each using a different yardstick. At insurance platform Policybazaar, Ringg says agents handle 67% of calls without a person, while average response time fell from 8–12 minutes to under a minute. At healthcare platform Practo, it reports 85% first-call resolution and response times below three seconds. Practo’s reported operating costs fell 70% against its previous human-led workflow, and Ringg says it now completes more than 1,000 appointment bookings there each day.

Ringg says investment platform Groww resolves 72% of incoming questions about IPOs, futures and options through self-service, with an average handling time of two minutes. Those figures describe different requests and different measures: a call handled without a person, a request resolved on the first call, and a query completed through self-service are not interchangeable outcomes.

Testing before a wider rollout

Ringg tests models on past conversations and simulated customer requests before introducing them to a small share of live traffic. Its tests include conversations that switch languages or mix English with local phrases. In one evaluation, Ringg says GPT-5.6 Terra beat Gemini 2.5 Flash and reached up to 97% accuracy on common regional languages. The case study does not give enough detail to treat that result as a general measure of multilingual customer-service performance.

Once a model is live, Ringg’s router watches response times and service availability across regions and shifts traffic when an endpoint slows or goes down. Its next proposed step goes beyond calls and chats: Ringg is developing browser agents for tasks such as onboarding and claims processing. It is also working on a way to preserve a customer’s context across channels. Those projects are in development, not reported customer results from the current platform.