Hello everyone,
Welcome to the 220th edition of The Pattern, a weekly look at what's changing across finance, technology, and the economy.
I'm writing this from Dubai, where I'm at MEBIS, the Middle East Banking Innovation Summit. We launched AI agents here that take on some of the parts of underwriting that nobody enjoys doing manually, and can save teams a lot of time. They can check whether an application is complete, trace who actually owns a business, pull clean data out of messy documents and sort what's actually been submitted, with only the exceptions going to a credit or risk analyst.
Between sessions, I ended up in a dozen conversations with lenders from across the region, and almost all of them came back to the same thing: there’s a lot of interest in AI in credit, but there’s still a lot of figuring out where it actually makes sense. Some use cases are pretty straightforward. Others get more complicated once you start looking at the data involved, the existing lending process, and what would actually need to change.
Our co-founder Srijan made a similar point a few editions back when he wrote about AI shifting from assistance to execution in credit, and being here has only made me more convinced he's right.
I had a similar thought while looking at what’s happening back in India. There is a difference between something looking ready and seeing how it really works when it is being used. It came up at the event, and then I noticed it again in two very different stories from India, first in the RBI’s credit numbers and then in a Supreme Court judgment.
Very different stories, but they left me with the same question: how much of what looks strong has actually been tested?
Let’s get into it.
The credit boom nobody's calling skill
That same question comes up in India's credit numbers. Banks pulled in $127 billion through the RBI's special FCNR(B) deposit scheme this year, far more than anyone expected. That kind of liquidity doesn't stay idle for long. Much of it has moved into lending, which helps explain the sharp jump in credit growth.
And it has. Bank credit grew 19.1% year-on-year by the end of August, the fastest pace in a decade and nearly double last year's 10.3%, with outstanding credit now above Rs 224 lakh crore. NBFCs saw a version of the same thing: credit there grew 14.9% in July against 10.6% a year earlier, with retail loans up 21.4% and vehicle finance up 19%. Two separate RBI releases, one shared story.
The interesting part is that even the people closest to this growth seem to think the current pace won't last. Motilal Oswal's Nitin Aggarwal has already guided his desk down to 14.2% credit growth for FY27, which is really him saying he expects the current run to fade. Axis Bank's Amitabh Chaudhry went further, calling the current 18-19% pace abnormal and pointing straight at the FCNR(B) money as the reason. When the people closest to the numbers are already flagging that the current pace won't last, that's worth taking seriously.
Which is probably why RBI just announced an open market sale of Rs 1 lakh crore in government bonds, its first net bond sale in two years, spread across three tranches on September 17, 21 and 28. In plain English: the same central bank that flooded the system with liquidity two months ago is now pulling a chunk of it back, because that surplus pushed short-term rates below the repo rate and made its own policy harder to hold. Bond markets didn't wait around, the 10-year yield hit a three-month high the day the sale was announced.
If the OMO sales do their job, I'd expect credit growth on both sides to settle toward the mid-teens over the next couple of quarters, which lines up with what Aggarwal and Chaudhry's own economists are already forecasting. NBFCs usually feel a liquidity squeeze earlier than banks do, so their funding costs, not their growth charts, are the number I'd watch first.
The difference between having a rule and following it
The second story is very different, but it comes back to the same question. This time, the test came in a Supreme Court judgment.
A borrower had taken a commercial vehicle loan from a large vehicle-finance NBFC and defaulted. One night at 1am, four unidentified men broke his steering lock and drove the truck away while it was parked after a delivery. No notice, no signed paperwork. The lender sold the truck for Rs 4.50 lakh soon after. The Allahabad High Court had thrown out his case, saying he'd waited too long to complain.
The Supreme Court didn't agree. It said RBI's loan recovery guidelines had "existed only on paper" because they weren't being properly enforced, and told RBI to make sure banks and NBFCs actually follow them. The bench also referred to an older ruling that said banks cannot use "goondas" to take vehicles by force.
The interesting bit, for me, was that the lender did have the right to repossess the vehicle without going to court. But that right came with conditions. There had to be notice, paperwork and a proper process. Once those were skipped, the lender couldn't rely on that right anymore. That's a pretty important difference between having a rule and actually following it.
The Court left the sale itself standing, since the truck was long gone, but it ordered the lender to close both loan accounts, refund the sale amount with 6% interest, and pay Rs 10 lakh in compensation for the borrower's loss of livelihood, a violation, it said, of his constitutional rights.
Two different stories, one honest question
The timing is what caught my attention. Both happened within days of each other, just as lenders were showing some of their strongest growth numbers. The recovery guidelines aren't new, and the RBI was always going to pull some of that excess liquidity back. But when the numbers look good, it's easy to stop asking what's driving them. Sometimes you only get those answers once something forces you to look closer.
I'd put what I heard in Dubai in the same bucket. There's genuine interest in what AI can do for credit, and some of the use cases are already working well in production. But plenty of lenders are still figuring out where it fits into their existing workflows, and that's a perfectly reasonable place to be. The credit numbers looked strong on the way up. The recovery rules looked fine on paper. In both cases, it took something actually putting them to use, or to the test, to see where the real gaps were.
Something can look ready, but you only really know once people start using it. That applies to credit growth, rules that have been around for years, and AI too. You figure things out as you go. You see what works, what needs changing, and what you hadn't thought about in the first place.
Cheers,
Mayank
Reading list
- RBI's OMO bond sale and why it's pulling ₹1 trillion out of the system
- Bank credit growth accelerates to a decadal high of 19.1% in August
- NBFC credit growth hits 14.9%, retail loans up 21.4%
- Supreme Court's judgment directing RBI to enforce loan recovery guidelines
- By Srijan Nagar: when AI shifts from assistance to execution in credit
- From a past edition: AI is coming for your loan recovery playbook
- UPI for credit card bill payments: Will MDR apply to credit card bill payments as well?
Thank you for reading. If you liked this edition, forward it to your friends, peers, and colleagues.
All opinions expressed are my own and do not necessarily reflect the views of FinBox or its promoters.
