UPI and the trust pact in its second act
As tap-and-pay, UPI AutoPay mandates and agentic commerce take hold, more payments in India will go through without a customer entering a PIN or approving each transaction. At a closed-door roundtable hosted by Mindgate Solutions with PayU, indust...
What that next phase would entail in terms of growth and innovation is the subject of Mindgate Solutions’ series on UPI, unfolding in closed-door roundtables, in association with PayU, for the leaders who build, lend and collect on the rail. The first conversation examined the trust question, especially as UPI scales further, exploring themes of merchant viability, platform accountability, and structural sustainability. . This second roundtable brought together leaders from lending, fintech, cross-border payments, and payments businesses to discuss the economics of trust and beyond.
In its first phase, UPI was often described as a cashback-fuelled contest for installs and habit, while merchants and banks stayed off the customer’s screen, with ownership staying with the app with customer oversight at every step. The second phase shifts towards agentic commerce, where a pre-defined mandate will increasingly replace the open-scan-confirm sequence and previously scheduled mandates will alter how money is debited. In this shifting landscape, AI agents will search, negotiate with the merchant’s agent and pay, with no human approving the individual transaction.
However, as ownership shifts, so does risk, because the system must now answer what the PIN once assured: who authorised the payment, within what limits, who is liable when an agent errs, and how a chargeback works when an agent enters the equation.
The gaps that need to be solved
UPI now handles more than 23 billion transactions a month and accounts for roughly 85% of India’s digital payment transactions. The ecosystem routinely runs above 10,000 transactions per second, and the largest banks have reached that level on their own.Mindgate powers UPI technology for around 14 large banks and financial institutions, which together process more than 12 billion transactions a month, a little over half of national volume. From that vantage, through its work with banks, the company sees a few underlying expectations to be fulfilled in the current landscape.
The first is technical declines. By global standards, UPI’s failure rates are low, but the tolerance is so tight that anything above 0.1% is treated as an aberration, and at this volume even a sliver of failure means millions of disrupted payments. The second is checkout friction. With more than 40 fintechs and third-party application providers (TPAPs) in the market, online payments still bounce customers between apps, and every redirection is a chance to drop off; native, in-app journeys are the industry’s answer. The third is what happens after the payment: reconciliation, settlement, disputes and unfulfilled transactions. Settlement is already approaching near-real-time, with the ecosystem working towards a 15-minute window within a year or two.
Beyond the QR code
One speaker presented an offline-online split, with roughly 70% of UPI payments made offline, at physical stores, and about 30% online, while noting that online still dominates by value. Interoperable AutoPay, agentic payments and near-field communication (NFC)-based tap-and-pay were highlighted as the features most likely to drive the next leg of growth.Another speaker, from a large conglomerate’s payments business, mapped what comes next in each channel. In physical stores, the shift is from scanning a QR code to tapping a phone at the counter. The feature National Payments Corporation of India (NPCI) launched in September 2026 lets users pay by tapping a phone at a point-of-sale terminal without opening the UPI app, and that, by some, is being seen as UPI’s answer to Apple Pay. Large enterprises want to remove app redirects altogether and keep the payment inside their own interface, through merchant plugins and in-app journeys.
Another view made the case for a product that already exists: UPI AutoPay, which lets a customer authorise recurring debits such as loan instalments. Lenders usually verify a borrower’s bank account with a penny drop, a small test transfer that confirms the account is real. Setting up a UPI mandate confirms the account in the same step, so verification and registration happen together.
Agentic payments go a step further. A household could set a ₹2,000 monthly grocery budget and let an agent find the cheapest basket across e-commerce platforms and pay for it, with the buyer’s agent dealing directly with that of the seller’s agent. UPI Circle, which lets one person delegate a spending allowance to another, and Reserve Pay, which blocks funds for later use, already provide the building blocks. For this to scale, security has to hold, chargeback rules have to be written, and someone has to own the outcome when an agent gets it wrong.
With fewer human interventions, however, the burden of oversight moves from the customer to the system. When every payment needed a PIN, the customer was the last line of defence, catching errors and deciding whether to retry. Mandates, taps and agents remove that checkpoint, so payment providers have to build it in. They need real-time monitoring that spots a failed or duplicated debit before the customer does, risk scoring that asks for extra authentication when a transaction looks unusual, and clear rules on who is liable and how a chargeback works when the payer is an agent rather than the person. Merchants carry a parallel burden. A payment they cannot see is a payment they cannot trust, so they need instant confirmation, quick refunds, and reconciliation that matches every automated debit to an order without manual checking. The less a human is involved in each payment, the more every party in the chain has to be able to prove what happened.
Why reliability is the next test
That makes reliability the next test, and the familiar debate about success rates no longer covers it.At UPI’s scale, even a 99.99% success rate leaves a very large number of people stranded, and the ambition has to reach well beyond four nines. But the bigger gap is not how often payments fail; it is how they fail. An artificial intelligence (AI) layer that monitors the system and tells each user, in their own language, exactly what happened would matter more than another decimal point. A system that fails gracefully can even afford a slightly more relaxed SLA.
Ambiguity hurts merchants too. A payment stuck in “processing” leaves a shopkeeper unable to trust the customer, especially after years of fake apps that flash convincing success screens. Soundboxes have largely solved this for small merchants, but customers still often pay twice when the first payment eventually goes through. Pattern recognition could help here as well, for instance by spotting the failure types where an advance settlement to the merchant carries little risk.
Reliability, in short, is turning into a trust product, as customers will forgive a failure they understand, but they will not forgive one that leaves them guessing about their money.
MDR and the trust paradigm
If reliability and security are what earn trust, someone has to pay for them. That is why the proposed MDR on UPI payments above ₹2,000 drew the evening’s liveliest exchange. Many small borrowers’ EMIs sit close to the ₹2,000 threshold, one noted, so one option is to move them to smaller, more frequent repayments that fall below it. For larger borrowers, the choice is starker: pass the charge on, or fold it into interest rates. There was some comfort in the fact that every lender faces the same cost, so no one is singled out. A second lender went further. It was still early, with negotiations under way and the start date uncertain, but collections might well shift towards mandates, which are excluded from the charge. More pointedly, one of the speakers questioned whether MDR was needed at all. TPAPs have already built viable businesses by treating UPI as an acquisition channel for lending and cross-selling, and one lending-led app has climbed into the top four that way. The cost, the speaker predicted, would not stay with apps for long, as small merchants would raise their prices within months, and customers would have no choice but to pay.Another speaker, who launched their TPAP app recently and stands to receive a share of the fee, offered the counterpoint: Large merchants will absorb or bundle the cost, perhaps returning it as rewards, while smaller ones may pass it to customers as convenience fees. Workarounds are already appearing, including tools that split a ₹4,000 bill into two payments just under ₹2,000. Since only a small share of person-to-merchant payments cross the threshold, the speaker estimated the annual pool at roughly ₹1,000 to ₹1,200 crore: a sting for customers, but a blessing in disguise for the apps. The fee also gives fintechs that are not yet TPAPs a reason to become one, since cashback offers have rarely paid off as a way to win UPI users.
Notably, a Mindgate representative pointed out that MDR has a history here: UPI carried it until 2020, and card payments have always had it, at significantly higher rates. Only about 4-5% of UPI transactions cross ₹2,000, and NPCI has set differential rates by merchant category. Card MDR compensates banks for lending customers money. A large portion of UPI transactions involves no lending, since payments come from savings accounts. The case for a UPI fee is different: it helps banks and fintechs recover what they spend running the system. At 0.4%, it is still far cheaper than cards.
Merchants, meanwhile, should look at their total cost per transaction rather than the headline rate, and may need a year or two to learn where the charge applies and where it does not.
A differing perspective held that payment aggregators and gateways do much of the groundwork, from onboarding merchants to installing QR codes and devices, and bear the costs involved, yet they are not among the four parties that split the MDR. Left out, they may push merchants towards payment methods where they do earn, including global wallets. Mindgate’s representative offered a different perspective: aggregators are already negotiating a cut with acquiring banks, which receive about 12 basis points of the 0.4%, and each will get a share that reflects the value and bargaining power it brings. At the same time, merchants are less likely to retreat to cash, which already costs them 0.4 to 0.5% or more to handle, while cards still cost more.
Inside the fraud fault line
NPCI has recently set out security controls that every mobile app offering UPI must carry, whether it belongs to a bank or a TPAP, and the same applies to the SDKs that embed UPI journeys inside other apps. NPCI’s certification of new TPAPs, the room heard, now scrutinises security more closely than features or design. OTP-based registration gave way to SIM binding once fraudsters learned to harvest OTPs through social engineering. SIM binding can be broken on rooted or jailbroken devices, so runtime application self-protection (RASP) became essential to detect them. Next comes silent mobile verification (SMV), which confirms the SIM and carrier at the network level without relying on the device. SMV was introduced a couple of years ago, before the industry was ready, but participants now expect it to become standard compliance. Even so, the engineering leader cautioned that fraudsters will eventually find a way around it too.AI has made vulnerabilities far easier to find, even for attackers without deep expertise, and the only answer is defensive models smarter than the offensive ones. Yet, it also offers far more embedded solutions for both enterprises and consumers. At the enterprise level, AI can be embedded in security systems as a verifier and auditor. At the consumer level, a robust personal agent could stand in front of the user, filtering out the phishing messages and impersonation scams that fool people but would not fool the agent, with two-factor authentication guarding the agent itself.
What the next 12 months should fix
Asked what NPCI and its infrastructure partners should change within a year, industry leaders pointed to five fixes. Lenders and brokers want loan repayments and capital-market collections exempted from MDR, as person-to-person transfers are, since in those cases they are only recovering their own money. Startups want room to build on UPI, instead of development staying with a closed set of players. On fraud, the imperative is a single platform where a fraudulent transaction can be reported quickly and resolved fast, rather than each app handling it alone. Cross-border players want UPI scaled internationally, so that money sent to and from India relies less on foreign rails such as card-network transfer services and SWIFT. And to make agentic commerce real, agents need their own identity on the network and the authority to pay without a person approving each transaction, so every agent payment can be traced back to who authorised it.What the future holds
UPI’s first decade was about getting people to trust a new way to pay. Its next phase asks them to trust it with more: to let a phone tap instead of scan, a mandate debit without a prompt, and eventually an agent spend on their behalf.For merchants, the question is whether UPI becomes more than an acceptance utility. For TPAPs and infrastructure providers, it is whether the system will finally pay for the trust they are being asked to guarantee. And for the customer, the test remains as binary as ever. The payment either goes through, or it does not. The next wave will be decided by how much the system can do without that answer ever being in doubt.
The Economic Times News App for Quarterly Results, Latest News in ITR, Business, Share Market, Live Sensex News & More.