Niva Bupa fired up a live pilot for an AI co-pilot to its 5,200 direct selling (DS) agents in Mumbai on April 1, 2026.
Director of Technology and Business Enablement Raja Rai confirmed the roll-out in a Thursday evening mail: “The AI helper answers policy questions within two seconds. Earlier agents waited up to 40 seconds.”
And agents are seeing a sharp drop in lead bounce.
Raja shared Friday’s early signal: 1,800 daily policy queries from the field were answered by the co-pilot, cutting manual look-ups from 35 down to 9 per agent per shift. The system draws policy, claims, and endorsement data from Niva Bupa’s 250,000 anonymised PDF files.
In Gurgaon, the insurer’s technology team built the assistant on a fine-tuned Hindi version of Mistral-7B. Raja said it runs on four NVIDIA T4 GPUs leased from AWS Mumbai (cost ₹8 lakh since November 2025), not on home-brewed infra.
“The model is not replacing agents,” Raja added. “It’s cutting the grunt work so they can spend more minutes on closing.”
Niva Bupa tracks lead time per stage in its CRM: quotation → acceptance → payment. Before the co-pilot, the average quotation-to-payment cycle stood at 7 days. After the March pilot in Bengaluru, that figure fell to 4.6 days among the 800 agents.
And during a 12-day stress test ending March 24, agents using the system converted 19.4% more leads versus the control group.
The Mumbai roll-out is restricted to offline agents across Malad and Andheri branches. The co-pilot only answers these eight question types:
1. Policy payout timeline
2. Exclusion list (health covers)
3. Renewal grace period
4. Endorsement charges
5. Hospital cash benefits
6. Ayush treatment limits
7. Tax benefit under 80D
8. Pre-existing disease clause
Raja declined to say how much the company spent on the pilot, citing pending IRDAI approval for any expense disclosure. IRDAI would need to see actuarial justification for any cost charged to policyholders, he added.
Niva Bupa Finance Control bought four thousand Le ChatPro licences on April 1 at ₹499 agent/year to let agents keep asking follow-ups while still on the field. That’s ₹20 lakh at list, but Raja hinted a 25% discount after negotiations with Mistral’s local reseller.
The co-pilot is plugged into Niva Bupa’s Salesforce CRM where it shows a green “AI Generated Answer” flag on the agent’s tablet. Tickets automatically close when policy numbers match.
Field agents trained on two weekday sessions. Priya Pawar, senior DS agent at Andheri Juhu branch, said on April 2: “I saved 32 minutes yesterday. Now I can visit another society in the same time-slot.”
And in Malad’s Lokhandwala Complex, agent Ajay Bhat reported a 23% rise in quotation acceptance on the first three days.
But rivalry with larger rivals is visible in Mumbai itself.
HDFC ERGO runs a similar Agent AI co-pilot in its Mumbai and Pune branches since January 2025. Its claims show a 14% lift in conversion rates against Niva Bupa’s 19.4%.
Niva Bupa’s head of business development Nikhil Metha circulated an April 2 WhatsApp note asking agents to report any instances where the co-pilot’s answers differed from policy documents.
Metha warned: “If the co-pilot contradicts the master policy wordings, quote page 3, paragraph 4.2 will be considered final.” The line echoes Niva Bupa’s 2024 policy wording dispute where ₹34 crore claims were repaid after courts accepted the printed clause over call-centre scripts.
Chief Digital Officer Sandeep Gulati told this reporter over a late-night call that the final gauge of success is whether the tool shortens the funnel so Niva Bupa meets its FY27 growth target: ₹5,800 crore gross written premium against FY24’s ₹4,700 crore.
Gulati declined to tie the AI pilot to that target, but he said “we expect it to shave at least ₹200 crore off administrative costs by FY27.”
Active field agents will triple-check the AI before submission, as the Insurer remains cautious after last year’s ₹18 lakh fine for wrongful data usage by an outsourced analytics firm.
Raja summed the gamble simply: “If it saves one agent 20 minutes a day for 250 days a year, that’s one extra policy per agent. 5,200 agents equals 5,200 policies—that’s ₹260 crore extra premium at ₹5 lakh average ticket size.”
That’s still a target, not a guarantee.
