Policybazaar moved 45% of insurance workflows to AI — issuance fell to 15 minutes
The insurance marketplace says AI cut policy issuance from four hours to 15 minutes for half its customers and lifted early fraud detection 14x.
What was changed
Policybazaar, India's largest insurance marketplace, has rewired its operations around artificial intelligence: by June 2025, 45% of its insurance workflows ran on AI, and policy issuance that once took up to four hours took 15 minutes for nearly half of all customers — attacking the biggest historical pain point in buying insurance. The numbers come from the company itself, laid out by CTO Saurabh Tiwari in an ETBFSI interview.
In claims, the most emotionally charged touchpoint, AI models drove what the company calls a 14x improvement in early claims detection, flagging 11% of term-plan cases and 16% of savings-plan cases as potentially fraudulent, which let the company triage faster and pay genuine claims quicker. Tiwari stressed the design keeps humans in the loop: models detect anomalies and outliers for investigation, advisors feed corrections back, and AI outputs on high-stakes decisions sit alongside other indicators — a structure he said mitigates the biases and blind spots of full automation.
The rest of the funnel changed too. Bots handled more than 30% of first-contact queries, up from 15% a year earlier; generative-AI trials lifted product discovery conversion 5–8%; service productivity rose 15–20% as resolution turnaround dropped 15%, with CSAT above 94%. The company sends 500,000+ personalized nudges daily, producing a 20% uptick in intent-to-buy. Support runs in 9+ Indian languages, and offline rural field agents use AI for eligibility checks and document verification. Roughly 10–15% of total tech investment went to AI, on a largely open-source LLM stack.
Why it worked
The four-hour issuance bottleneck was the sector's defining pain point, so cutting it to 15 minutes delivered a visible, marketable jump in experience.
Human-in-the-loop review was kept on sensitive calls like fraud, letting AI triage volume while advisors checked the flagged cases.
Distribution matched the market: multilingual support and offline-agent AI tools extended the model to rural customers digital-first rivals skip.
The stack leaned on open-source LLMs with internal governance, keeping the build-out inside 10–15% of tech spend rather than a bet-the-company outlay.
What can be applied
Automation pays fastest where the customer feels the wait: attacking a four-hour bottleneck beats shaving percentages off processes nobody complains about.
Aftermath
Policybazaar was piloting agentic voice-based outbound systems for premium reminders, renewals and documentation, and expected AI's share of its tech investment to keep rising, with the stated goal of staying 'ahead of the curve — responsibly, transparently, and always with the customer at the center.'