Razorpay's ChatGPT Commerce Bet, Explained

For years, ecommerce in India has followed a predictable path: a consumer searches on Google, scrolls through a marketplace, compares reviews, and finally checks out through an app or website. That pattern is starting to crack.

AI chatbots are increasingly becoming the first stop for product discovery — and in some cases, the last stop for the actual purchase. Nykaa gave the clearest evidence of this shift yet in its Q1 FY27 management commentary, revealing that it has become the top-ranked beauty brand on both ChatGPT and Gemini through a practice it calls Answer Engine Optimisation (AEO) — essentially, SEO for AI assistants instead of search engines. The company is going a step further too, working to bring its Nykaa Beauty and Nykaa Fashion shopping experiences directly inside ChatGPT as part of a wider OpenAI tie-up.

That shift creates a new problem for brands: getting discovered inside a chatbot is only half the battle. Without a way to actually complete a purchase inside that same conversation, all that visibility goes to waste. This is exactly the gap Razorpay is racing to fill.

Razorpay's Early Bet On Conversational Commerce

Long before AI-led shopping became a boardroom talking point, Razorpay had already begun laying the groundwork. The Bengaluru-based payments major started working with OpenAI and the National Payments Corporation of India (NPCI) on infrastructure that would let merchants build storefronts natively inside ChatGPT and complete transactions without ever leaving the chat window.

According to Razorpay co-founder and CEO Harshil Mathur, the company is currently working with roughly 50 brands and several ecommerce marketplaces piloting agentic commerce, with skincare label Innovist among the first to go live for select users. The payment flow itself leans on two existing UPI capabilities — UPI Reserve Pay, which lets a user pre-block funds for future merchant debits, and UPI Circle, which delegates UPI authentication so a transaction can be completed inside ChatGPT without redirecting to a separate app.

Razorpay isn't alone in chasing this opportunity. Rival Cashfree has rolled out "Cashfree Here," a payments extension built for AI applications, enabling in-chat transactions across platforms including ChatGPT and Claude. The two companies, however, are approaching the problem differently — a distinction that matters for how the market eventually shakes out.

How Razorpay's Approach Differs From The Competition

Building a merchant's own MCP (Model Context Protocol) server and checkout flow from scratch is typically an eight-to-ten-week engineering project. Razorpay's pitch is that it collapses this into roughly half an hour for a Shopify merchant — auto-syncing the product catalogue and handling the entire submission process to OpenAI on the brand's behalf.

Cashfree's agentic payments tooling, by contrast, still requires merchants to build and plug their own AI agents into its MCP layer — a more hands-on approach that shifts more engineering work back onto the brand.

Mathur believes this difference could become a meaningful edge over time — not because AI platforms will exclusively favour one payment provider (they won't; ChatGPT will support multiple gateways much like a website can host any payment option), but because merchants will likely commit to just one or two payment partners for agentic commerce specifically. Getting those relationships locked in early, he argues, is where the real competitive advantage lies. "The merchants that we are taking live only work with us," Mathur told Inc42, describing the current cohort as having some degree of exclusivity.

The Trust Problem At The Centre Of AI Commerce

Every executive in this space keeps circling back to the same question: will Indian consumers actually trust an AI chatbot with their money?

Cashfree CEO Akash Sinha has pointed out that while AI promises real efficiency gains for payments infrastructure, the agentic commerce layer will inevitably run into consumer hesitation around handing financial data and decision-making over to an AI agent.

Razorpay's answer is architectural rather than reassurance-based. Per Mathur, the payment data itself never reaches the underlying AI model — ChatGPT never receives a customer's actual payment credentials. Every transaction still routes through NPCI's UPI rails and requires explicit user authorisation. In practice, a shopper discovers and selects a product conversationally, but the moment checkout begins, control passes to Razorpay's payment stack, with authorisation happening through the customer's own UPI app and standard two-factor authentication — not through the chatbot itself.

Razorpay says its system is built around two non-negotiable safeguards: payment data staying completely outside the LLM's reach via the UPI Reserve Pay protocol, and every transaction requiring either pre-set customer consent or two-factor authentication. "It's not like you give a customer's details and let the agent shop and the customer has no control," Mathur said. The wider regulatory framework governing this new transaction layer, though, is still very much a work in progress.

Why Brands Are Jumping In Early

Unsurprisingly, the earliest adopters are digital-first, direct-to-consumer (D2C) brands — a segment that has been squeezed hard by rising customer acquisition costs on Meta and Google, alongside steep commissions charged by ecommerce marketplaces. For these brands, a new, lower-friction discovery-to-checkout channel is an attractive escape valve.

Razorpay, for now, isn't charging merchants any additional platform fee during this early testing phase — the company says it's prioritising adoption over monetisation while it figures out product-market fit. Mathur has floated the idea of an eventual SaaS or distribution-style fee, but is careful to frame it as speculation rather than a firm plan. "Unless we see scale, I don't think we'll focus on monetisation in this domain," he said.

Part of the appeal for smaller merchants is technical, not just financial. AI-native commerce requires exposing structured product catalogues, live pricing, inventory data and checkout logic in formats an AI assistant can actually parse — something most small D2C brands simply don't have the engineering bandwidth to build on their own. Razorpay is positioning itself as the layer that absorbs that complexity on their behalf.

Beyond ChatGPT: Razorpay's Wider AI Ambitions

The ChatGPT integration is just one piece of a broader AI push at Razorpay. The company's Agent Studio platform — currently in beta — already has around 200 businesses using it to deploy AI agents for tasks like payment reconciliation, abandoned-cart recovery, subscription recovery and finance operations. Newer features include agent "memory" that retains context from previous customer interactions, and merchant-controlled settings for how assertively an agent should follow up with a customer who abandoned their cart.

Notably, Razorpay isn't charging for Agent Studio either. "We don't even know what is the right way to charge for it," Mathur admitted — a sign of just how early-stage this entire category remains, even for the company building it.

Building this stack hasn't been cheap in terms of effort, even if Razorpay hasn't disclosed hard numbers. "Investment is significant in terms of mostly engineering effort," Mathur said, pointing to the need for a dedicated team continuously shipping capabilities for agentic commerce. The company's ambitions extend well past ChatGPT alone — Razorpay says it's building what it calls a distribution layer spanning merchant websites, independent AI agent platforms, and multiple large language model providers, including plans to eventually support Anthropic's Claude and Google's Gemini for agentic payments.

Can ChatGPT Really Become India's Next Shopping Mall?

That remains genuinely uncertain. India's ecommerce habits are still deeply rooted in dedicated apps and marketplaces, where consumers browse images, compare reviews and window-shop visually — behaviours that conversational interfaces have only just begun to approximate.

Mathur draws a parallel to the early days of ecommerce itself, when Indian consumers started with small, low-risk purchases before gradually trusting digital platforms with bigger-ticket spending. He expects AI-led commerce to follow a similar trust curve rather than an overnight shift.

For now, Razorpay's strategy is unambiguous: chase adoption first, worry about monetisation later. "We'll have to scale it to a point where enough and more customers are transacting to the point that we say, hey, now we need to monetise it," Mathur said. "Right now we're in the adoption phase of the journey."

The broader signals, though, suggest the industry isn't waiting around to find out. Retailers like Nykaa are actively tracking their visibility across AI answer engines, while payment players including Razorpay and Cashfree are simultaneously building competing layers of infrastructure for this emerging channel. Whether AI assistants end up becoming a genuine shopping destination or simply a complementary discovery layer will hinge entirely on how fast Indian consumers warm up to the idea of an AI handling — even partially — their money.