Ringg has developed an enterprise agent platform that operates across voice, chat, WhatsApp, and web channels to address call volume and fragmented systems in the customer service operations of major consumer companies in India. Moving suitable real-time workloads from GPT‑4.1 to GPT‑5.6 reduced model costs by approximately 90% while maintaining the required quality and latency levels. Ringg agents manage more than 7 million connected calls per month; the average customer satisfaction score among its customers is 4.8.
The platform can perform actions in CRM, ticketing, payment, and scheduling systems, as well as internal APIs, and transfers conversations to a specialist with a contextual summary when necessary. Its knowledge system combines semantic search with structured filtering across datasets, PDFs, CSVs, and business documents. Tasks can be divided among specialized sub-agents for qualification, support, verification, scheduling, and escalation. Ringg evaluated OpenAI models based on quality, latency, instruction following, tool use, multilingual performance, reliability, and cost; it stated that OpenAI delivered the most balanced result in production workloads.
- GPT‑4.1 handles most real-time voice and chat traffic.
- GPT‑5.6 Luna is used for requests where it is suitable in terms of performance, latency, or price-performance.
- GPT‑5.6 Terra handles post-conversation summaries and sentiment classification.
- GPT‑5.6 Sol is used for evaluation, prompt improvement, and model-as-judge processes.
The routing system combines customer input with instructions, history, customer data, the knowledge base, and tools before sending it to the model. When the context reaches approximately 80,000 tokens, a structured summary is generated. The models are tested with past conversations and simulated flows; GPT‑5.6 Terra achieved up to 97% accuracy in regional languages in a comparison with Gemini 2.5 Flash. The models are first tested on a small portion of production traffic; traffic is adjusted based on latency and endpoint status.
Ringg agents resolve up to 65% of routine requests without human support. At Policybazaar, more than 57,000 requests were connected, 67% of calls were handled without human intervention, and response times fell from 8–12 minutes to under 60 seconds. At Practo, the first-call resolution rate reached 85%, response times fell below three seconds, costs decreased by 70%, and more than 1,000 appointments were completed per day. At Groww, 72% of IPO, futures, and options inquiries were resolved through self-service. Ringg is also developing a system that preserves context across browser agents and channels.
Why it matters
This development shows that, in customer service, a model can be implemented that brings together different channels on a shared transaction and information layer instead of relying on structures tied to a single communication channel. Resolving a significant portion of routine requests without human support creates a division of labor that could enable employees to focus on more complex conversations and cases that require intervention. The fact that model selection is based on the type of request, latency, multilingual performance, and cost highlights the importance of managing workloads in a segmented manner in AI use, rather than simply choosing the most powerful model. However, whether high automation rates can be maintained across different customer profiles and in more complex processes remains an open question, particularly in workflows involving access to systems such as payment and scheduling.
Background
OpenAI is not a new name in the FikirPilot archive: over the past 90 days, we have published 74 news articles mentioning this name; the latest was dated October 3, 2026.
Term: agent
An AI agent is software that calls tools and carries out multi-step tasks to achieve a goal instead of producing a single response.