Chief Product Officer / India
Three decisions that cost us speed
- Gold-loan assets under management
- ₹4,860 Cr
- At 31 March 2026
- Active loan accounts
- 6.1 lakh
- Across 412 branches in four states
- Product team
- 46
- Including four language editors
Decisions & consequences
01Slowing app top-ups after scammers turned them against borrowers
Online top-up let a borrower add to her loan against the same pledged gold when its value rose. By mid-2024 it disbursed ₹96 crore a month, directly into the borrower's own bank account, as the RBI's digital lending rules require. Between July and November 2024, branches recorded 212 complaints from borrowers who had been called by someone claiming to work for us. The caller said an excess top-up had been credited by mistake, obtained the one-time password and then asked the borrower to return the money through a UPI request. Customers lost ₹1.36 crore. Our own systems had not been breached, but the product had made the scam easy to explain.
The growth head argued for a fraud-scoring model and customer education, preserving the instant journey. The chief technology officer preferred device checks alone. I recommended a four-hour hold on any first app top-up above ₹50,000 and on any top-up within seven days of a device or mobile-number change, with a call-back from the home branch in the borrower's language. I also asked for a confirmation screen stating the amount in words and a warning that we never request money back through UPI. The managing director approved the change in December 2024 after a presentation to the risk management committee.
Monthly app top-up value fell to ₹79 crore in the first quarter and recovered to ₹87 crore by July 2025. Complaints of this type fell to 17 in the following five months, with ₹9 lakh reported lost. Branches absorbed about 1,400 call-backs a month. The board decided separately on goodwill payments to affected customers; I supplied the complaint evidence.
The change taught us to treat persuasion as a product risk, not a customer failing. I would do one thing differently. Branch staff first heard these stories in July, but my team read only app reviews and call-centre data. Had we read the branch complaint registers weekly, we would have acted about eight weeks earlier.
Stopping a self-service first-loan journey I had championed
In October 2021 my digital team launched a journey for first-time borrowers. A customer could complete eKYC on the app, enter the approximate weight of her jewellery, receive an estimate and book a branch visit. By the end of 2022, about 1.1 lakh people had started it, but only 9% completed a loan. Branch managers reported a sharper problem. Many arrivals treated the estimate as an offer, and the appraisal often valued the gold lower because of stones, wear or purity. First-month complaints for these customers ran at 6.8 per 1,000 loans, against 2.4 for those who walked in.
I had championed the journey, and its roadmap held eleven further features and fourteen people. Our head of marketing valued it as our main digital acquisition channel, and the private equity investor's quarterly review tracked its leads. The alternative was to stop new development and move those people to the branch tablet, where appraisal would happen in front of the customer and the key fact statement would be explained in her language. We would lose a visible digital measure and write off about ₹2.6 crore of build work that we could not reuse.
In March 2023 I recommended stopping the journey, and the managing director agreed after a discussion with the investor. The fourteen colleagues rebuilt the tablet flow: the weighing and purity results appear on screen, the amount is calculated from them and the officer plays a short recorded summary in Gujarati, Hindi or Marathi before signature.
Within a year, counter time for a first loan fell from 41 to 29 minutes, and first-month complaints for new borrowers fell to 2.9 per 1,000. New-customer loans grew 11% in 2024–25, though gold prices and branch openings contributed to that. What I would do differently is test the estimate wording with thirty first-time borrowers before launch. They would have told us in an afternoon what 1.1 lakh starts took fifteen months to show.
Rebuilding the notices before an auction in the borrower's language
When a gold loan remains unpaid after its term, we must give the borrower notice before auctioning the pledged gold, and the RBI expects that process to be fair and transparent. In 2024–25, 9,800 accounts reached auction. Our notices went by registered post and an English SMS. Collections records showed that about a quarter of borrowers whose gold was auctioned visited a branch within a month afterwards, many saying they had not understood the notice. For a household, an auction can mean losing jewellery with a value far beyond its price.
The collections head worried that a longer sequence would lengthen overdue periods and invite borrowers to wait. Legal insisted that the regulatory notice and its timing remain unchanged, and risk was concerned about exposure to a fall in gold prices. My proposal kept the legal notice intact but added three reminders in the borrower's language, a branch call with a script that explained renewal, part-payment and part-release, and a UPI link to pay interest and renew from the app. The trade-off was time: we would take longer to resolve some accounts, and branches would make about 41,000 more calls a year.
The risk management committee approved a one-year trial in April 2025, on the condition that every extension had a written end date and that monthly overdue balances were reported. I agreed with collections that a reminder could never postpone an auction already scheduled under the notice.
In 2025–26, accounts sent to auction fell to 6,300. About 2,900 borrowers renewed, part-paid or took a part-release after the second reminder. The average period from first missed payment to resolution lengthened by 19 days, and overdue loans at the year end were ₹146 crore higher than the previous trend suggested. The committee extended the approach with tighter limits. I would involve branch staff earlier next time; their first draft of the call script was better than ours.