Bajaj Finance analysed 2 crore phone calls with AI last quarter and converted them into ₹1,600 crore in fresh loan disbursals — roughly 10% of their entire Q3 volume.
That number deserves a second read. A single AI deployment — voice-to-text on customer service calls — added a disbursement channel that didn't exist a year ago.
The Bajaj Finance story is worth unpacking because the logic is simple but the execution wasn't.
Most NBFCs treat call centre transcripts as support data. Bajaj Finance ran AI across 52 lakh customer interactions, extracted latent signals, and generated 1 lakh new personalized loan offers for customers they already had data on but no offer for.
That's a credit origination engine built from data most lenders throw away.
CEO Rajeev Jain signalled they're going further: 10 crore AI-powered calls in the coming year and 800+ autonomous AI agents across sales, collections, risk, and dealer management operations. These are capital allocation decisions, not roadmap slides.
While Bajaj Finance focused on acquisition, L&T Finance went after the underwriting bottleneck directly.
Project Helios — their AI underwriting co-pilot — processes bureau and banking data to support credit decisions. The numbers after rollout: SME turnaround for SEP customers dropped from 21 hours to 14 hours. For SENP customers, from 37 hours to 25 hours.
That's 1.5 hours saved per case, across 5,000+ underwriting decisions so far.
For a lender running thousands of SME loans a month, this isn't a marginal efficiency gain. It's structural capacity. The same credit team processes more, faster, with less manual review burden.
Here's the uncomfortable part. Both Bajaj Finance and L&T Finance are large. Bajaj Finance has the data volume to make voice AI work. L&T Finance had the resources to build a bespoke underwriting co-pilot and run it through thousands of cases before scaling.
The medium-sized NBFC — ₹500 crore to ₹3,000 crore AUM — doesn't have 2 crore call records to train on. They often lack a clean, structured data pipeline from sourcing through collections.
This creates a widening asymmetry. The top 10 NBFCs are building AI moats on proprietary data. Everyone else is watching from the sidelines, debating whether off-the-shelf tools will close the gap.
They won't. Not without structured intelligence at the account level first.
The RBI has been explicit: as of 2026, AI underwriting models entering production require documented bias testing across protected characteristics. That's a compliance obligation, not a suggestion.
What that means in practice: before any NBFC deploys AI in underwriting, someone has to audit the training data, document the variables, and validate outputs against the regulator's framework.
Lenders who haven't done basic data hygiene — clean bureau pulls, structured CAM workflows, consistent rejection reasons — are not AI-ready. The model is the easy part. The data pipeline beneath it is the actual work.
The gap between Bajaj Finance and a mid-tier NBFC isn't computing power. It's years of structured, consistent, labelled credit decisions feeding a model that gets smarter with each disbursement.
Bajaj Finance plans 10 crore AI-powered calls next year. That's not a projection. It's a budget line.
Every credit head at a mid-tier NBFC is doing the math right now.
How much did AI contribute to Bajaj Finance's loan disbursals in Q3 FY26? AI-powered call centre operations contributed ₹1,600 crore in loan disbursals — approximately 10% of Bajaj Finance's total Q3 FY26 volume. The system analysed 2 crore calls and extracted credit signals to generate 1 lakh new personalised loan offers.
What is L&T Finance Project Helios and how much did it reduce underwriting time? Project Helios is L&T Finance's AI underwriting co-pilot that processes bureau and banking data to assist credit decisions. It reduced SME underwriting turnaround from 21 hours to 14 hours for SEP customers, and from 37 hours to 25 hours for SENP customers — saving approximately 1.5 hours per case across 5,000+ decisions.
Does the RBI require documentation for AI underwriting models in 2026? Yes. The RBI now requires NBFCs to conduct and document bias testing across protected characteristics before AI underwriting models enter production. Lenders must validate model outputs and maintain audit trails for regulatory examination.
Abhishek Gupta is Co-Founder at Dekrypt Labs, building BIOS — a Business Intelligence Operating System for Indian businesses. dekryptlabs.com