IndiaFinance and insurance2025 – ongoing

PB Fintech (Policybazaar)

Finova AI text and voice insurance assistant

Policybazaar's Finova AI chatbot held quality conversations with 73% of the users who opened it

WorkingVendor case studyVendor ↗

Problem

Policybazaar serves a large online customer base that asks insurance questions outside business hours and in several Indian languages.

Approach

The company built Finova AI, a text and voice assistant that answers insurance questions in English and some Indic languages. Voice input, language detection, translation and a toxicity check run in the browser. A server-side model trained on Policybazaar data writes the answers. The team moved the toxicity model from a WebGL to a WebGPU backend where supported.

Results

73%share of users who opened the chatbot and had quality conversations, per GoogleVendor case study[1]

2xclick-through rate of the new call-to-action versus the previous one, per GoogleVendor case study[1]

10xfaster toxicity-model inference after the switch to WebGPU, per GoogleVendor case study[1]

Takeaway

Running language and safety checks on the user's device kept the server model focused on insurance answers. The figures come from Google's case study.

Sources

Policybazaar extended after-hours customer assistance with Web AI, and 73% users had quality interactions
web.dev (Google)21 May 2025Vendor publication

Source summaries

[1]web.dev (Google)21 May 2025

Google's case study describes Policybazaar's Finova AI chatbot and its in-browser pipeline for speech, translation and toxicity checks. It reports 73% of users who opened the chatbot had quality conversations, a 2x higher click-through rate on the new call-to-action and 10x faster inference with WebGPU.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026