Multichannel Customer Service Automation
Ringg has developed a multichannel customer service agent platform utilizing GPT-5.6 and other OpenAI models to automate interactions across voice, chat, WhatsApp, and web channels. The platform manages over 7 million connected calls each month, achieving an impressive average customer satisfaction score of 4.8.
Architecture and Model Routing
To facilitate complex multi-step tasks such as policy checks, scheduling, CRM updates, and escalations, Ringg integrates model inference, a tool orchestration layer, and semantic knowledge retrieval. The platform intelligently routes requests to specific OpenAI models based on the requirements of latency, instruction following, and cost efficiency. For instance, GPT-4.1 is employed for most real-time voice and chat traffic, while GPT-5.6 Luna is utilized for selected real-time workloads. Additionally, GPT-5.6 Terra is designated for post-call analysis, and GPT-5.6 Sol is used for evaluation and model-as-judge workflows. When conversation context becomes extensive, the system generates a structured summary at approximately 80,000 tokens, allowing interactions to continue without needing to resend the entire history.
Cost Reduction and Model Evaluation
Ringg reports significant cost savings, with a migration of selected real-time workloads from GPT-4.1 to GPT-5.6 Luna resulting in a reduction of model costs by approximately 90%. The company employs offline tests and simulated customer flows to evaluate model performance and optimize prompts and configurations. In their assessments, GPT-5.6 Terra has outperformed alternatives like Gemini 2.5 Flash in tasks related to summarization and sentiment analysis, achieving up to 97% accuracy on common regional languages.
Customer Outcomes
Ringg provides several customer success stories to highlight the effectiveness of its platform. For example, Policybazaar has connected over 57,000 requests through Ringg, with 67% of calls resolved without human intervention, and average response times reduced from 8–12 minutes to under 60 seconds. Practo has achieved an 85% first-call resolution rate with response times below three seconds, while also reducing operating costs by about 70%. Similarly, Groww resolves 72% of certain inbound queries with an average handling time of just two minutes.
Development of Browser Agents
In addition to its existing capabilities, Ringg is advancing the development of browser agents that integrate on-screen activity with conversational context for various workflows, including onboarding, Know Your Customer (KYC) processes, IT troubleshooting, and claims processing. Co-founder Siddharth Tripathi noted that OpenAI’s computer-use capabilities have accelerated the development of this browser-agent roadmap, enabling the team to maintain tool usage, low latency, and effective instruction following during model migrations.
Operational Monitoring and Deployment
Ringg stages its model deployments carefully, monitoring latency and endpoint health to ensure optimal performance. The system is designed to shift traffic or isolate versions to mitigate operational impacts while expanding production use.
Original source: OpenAI News