Bajaj Finance's AI listened to 2 crore customer calls and turned those conversations into ₹1,600 crore in loan disbursals — roughly 10% of their entire Q3 volume — without a single new customer.
The data was already there. In recordings nobody was reading.
NBFC credit teams are obsessed with structured data. CIBIL scores. GST filings. Bank statement analysis. All of it is clean, formatted, queryable.
Phone call recordings are none of those things. So they pile up on servers and nobody touches them.
Bajaj Finance's Q3 FY26 concall changed that framing. Rajeev Jain described a system that processed 2 crore calls, converted voice to text, then extracted structured data for 5.2 lakh customers — income signals, product intent, objection patterns, urgency markers. The output: 100,000 new personalized loan offers for customers the system had no previous structured data on.
These weren't existing customers with clean credit histories. These were people who had called, talked, expressed a need — and walked away without an offer because nobody processed what they said.
The traditional credit pipeline asks a fixed question: does this applicant's data meet our criteria?
Voice analytics asks a different one: what is this person telling us that we haven't formally captured?
The two questions produce different outcomes. The first rejects everyone who doesn't fit a pre-built box. The second finds opportunities inside conversations that never made it into a database.
Bajaj Finance didn't build new products. They didn't acquire new data sources. They listened to what customers were already saying — and extracted the signal.
This is not a future capability. It ran in Q3 FY26. It generated ₹1,600 crore.
Bajaj Finance has 26 product lines, a massive customer operations team, and the engineering budget to match. Most NBFCs don't.
But the underlying logic applies at any scale: your highest-quality lead data is already inside your organization. It's in call logs, in field officer notes, in WhatsApp conversations with borrowers, in rejection letters that went unanswered.
The difference isn't data availability. It's data extraction.
Smaller NBFCs running ticket sizes of ₹2–15 lakh are drowning in unstructured borrower interactions. A field officer visits a kirana store owner, writes three lines in a notebook, and the loan gets approved or rejected on that. The context in that conversation — the owner's seasonal patterns, their supplier credit terms, their expansion plans — disappears.
Bajaj Finance built infrastructure to capture that context at 2 crore call scale. Smaller lenders need to think about how to capture it at 2,000 interaction scale first.
More data doesn't automatically mean better underwriting. What matters is whether you can extract the right signal from unstructured information.
Bajaj Finance's system didn't just transcribe calls. It structured them — pulling out intent signals, organizing customer profiles, creating actionable offer triggers. The transcription was the easy part. The structuring was the value.
This is the part the fintech industry keeps underestimating. Every mid-size NBFC in India has call recordings, CRM notes, and field data sitting unprocessed. The question isn't "do we have enough data?" It's "do we have a system that turns interactions into credit intelligence?"
For most of them, the answer is still no. That's the gap.
How did Bajaj Finance use AI for credit underwriting in India? Bajaj Finance's AI analyzed 2 crore customer calls using voice analytics, converting audio to text and extracting structured credit signals for 5.2 lakh customers. This produced 100,000 new loan offers and ₹1,600 crore in Q3 FY26 disbursals from customers the company had no prior structured data on.
Can smaller NBFCs use AI for credit underwriting like Bajaj Finance? Yes, but at a different starting point. Smaller NBFCs should focus on structuring the unstructured data they already have — field officer notes, customer call logs, WhatsApp interactions with borrowers. The principle is the same: interactions contain credit signals that never make it into formal databases. The infrastructure requirement is lower at smaller scale.
What is the ROI of AI in NBFC lending? Bajaj Finance's Q3 FY26 results offer a concrete data point: AI-driven voice analytics on 2 crore calls generated ₹1,600 crore in incremental loan disbursals. That's roughly 10% of their quarterly volume from data that was already being collected but not processed. The return depends entirely on whether the AI system can extract structured signals from unstructured sources — not just store or search them.
Abhishek Gupta is Co-Founder at Dekrypt Labs, building BIOS — a Business Intelligence Operating System for Indian businesses. dekryptlabs.com