Ringg's AI Agents Now Resolve Up to 65% of Calls on Their Own!

Hi, it's me, Shiichan! Today I've got a story from OpenAI News that could really change customer support in India.

OpenAI News openai.com

What was announced?

OpenAI News published a customer story about Ringg, a voice and chat AI agent platform. Ringg serves large consumer businesses mostly in India, and it uses OpenAI models including GPT-5.6 to automate customer interactions across phone, chat, WhatsApp, and the web.

The story so far

Ringg was tackling the classic scaling wall in customer support: every time inquiries grow, hiring more people to keep up eventually hits limits on speed and cost. Instead of adding more staff, Ringg set out to make its agents smarter.

What changes

Here's how Ringg's platform is built.

• It covers four channels: voice, chat, WhatsApp, and web
• An orchestration layer that takes actions across CRMs, ticketing systems, payment platforms, scheduling tools, and internal APIs
• A knowledge system that combines structured filtering with semantic retrieval across PDFs, CSVs, and business documents
• Specialized subagents for qualification, support, verification, scheduling, and escalation

Ringg also routes tasks to different OpenAI models depending on the job.

• GPT-4.1: handles most real-time voice and chat traffic
• GPT-5.6 Luna: used when performance, latency, or price-performance make it the better fit
• GPT-5.6 Terra: handles post-call summaries and sentiment analysis
• GPT-5.6 Sol: used for evaluation and model-as-judge quality checks

As a result, Ringg now connects more than 7 million calls a month, resolving up to 65% of routine inquiries without a human agent. Average CSAT sits at 4.8, and moving workloads from GPT-4.1 to GPT-5.6 Luna cut model costs by about 90%. The system also handles regional language variation well, reaching up to 97% accuracy, and keeps conversations coherent by summarizing context at around 80,000 tokens.

Dive Deep

OpenAI News also shares concrete numbers from companies using Ringg.

• Policybazaar (insurance): handles 57,000+ requests a month, resolving 67% without a human, and cut response time from 8-12 minutes to under 60 seconds (about an 88% improvement)
• Practo (healthcare): 85% first-call resolution for appointment bookings, response times under 3 seconds, a 70% cut in operating cost, and 1,000+ bookings processed daily
• Groww (investing): 72% self-service resolution for IPO, futures, and options queries, with an average handling time of 2 minutes

Ringg is also building browser agents using OpenAI's computer-use capabilities, aiming to apply them to onboarding, KYC verification, IT troubleshooting, and claims processing.

Co-founder Siddharth Tripathi shared this comment.

Model quality is only part of the equation. We also need low latency, reliable tool use, strong instruction following, and economics that work at scale.

In other words, model smarts alone aren't enough - low latency, reliable tool use, strong instruction following, and economics that hold up at scale all have to come together before it works in production.

Wrap-up

• OpenAI News published a case study on how Ringg runs multichannel AI customer support for large businesses in India
• By routing tasks across GPT-4.1 and GPT-5.6 (Luna / Terra / Sol), Ringg resolves up to 65% of over 7 million monthly calls automatically while cutting model costs by about 90%
• With real improvement numbers from Policybazaar, Practo, and Groww, this is a especially useful read for companies and support teams dealing with high volumes of customer contact!