Razorpay’s Biggest Bet: From Payments To Becoming The AI Brain For India’s Small Businesses

Razorpay’s CEO Harshil Mathur has AI agents walking him through his day’s calendar, reading out the most important stories of the day to him and planning his travel itinerary. 

As the chief of one of India’s largest fintech companies, Mathur is one of the key figures who has advocated for the wide adoption of AI agents for simplifying daily tasks of the white-collar workforce. 

“I am a techie at heart. I couldn’t resist using this wonderful technology and I am obsessed with what agents can do for you,” Razorpay CEO told Inc42 on the sidelines of the company’s annual flagship event FTX, which concluded last week. 

Mathur and his team brought the biggest global leaders of the AI world, like OpenAI, Anthropic, Nvidia, Emergent, and ElevenLabs, under one roof at FTX. Razorpay announced multiple partnerships at the event, with “AI” at the centre of it all. 

This included unveiling its Agentic AI studio in partnership with Anthropic’s Claude, working with ecommerce partners like Swiggy, Zomato to let agents orders and complete payments for your food in addition to tie-ups with PVR Inox, Dermacore, BigBasket, and LinkedIn, among others.

Mathur told us that the way agentic AI has evolved in the past couple of years, Razorpay had to think of this direction sooner or later. The company, he said, had a huge advantage in this regard as millions of small merchants are already using Razorpay for more than just payments.

As a company with a payments infrastructure DNA, this also meant stepping out of the fintech comfort zone and creating something that extends into marketing and commerce. This is perhaps why the Razorpay AI-Native Agents Studio is the biggest and riskiest gambit in the company’s storied life thus far. 

“We are transforming into a business operating firm where we help our merchant partners operate in a simplified, more cost-efficient manner. Think of it as a team of AI specialists that businesses can hire in one click — each agent purpose-built to handle a specific commerce challenge, from recovering failed payments to settling disputes before they escalate,” the CEO said about the AI-Native Agents Studio, a marketplace and builder platform, built on top of Claude’s agentic platform. 

Mathur added that thanks to the marketplace approach, businesses don’t need to stitch together separate tools for AI. They simply pick the agents they need from a growing marketplace, right where they already manage their payments. 

This makes Razorpay something like Shopify, which also has an in-house marketplace for plugins and connecting to various services.  

Razorpay’s Agentic Avatar

Razorpay’s core insight is that while payments themselves have become seamless, the operational complexity around them remains deeply fragmented. Nearly 80% of its customers are small businesses, digital first brands and startups which have to deal with operational complexities that go beyond payments.

“When a customer drops off during checkout, someone needs to investigate why. If a subscription payment fails, someone needs to retry or recover the transaction. Finance teams spend hours reconciling settlement files with bank statements. Disputes and chargebacks require documentation and manual responses. Large companies can build teams and workflows to manage these processes. Smaller businesses, however, often lack the bandwidth or resources,” Mathur told us

He believes that these points of friction are why small businesses stay small and cannot scale up. Simply put, they don’t have the bandwidth to do all of the things that a large company can unless they spend a lot of cash. This is simply not possible for smaller startups and businesses without the access to investor funds. 

The irony, as Mathur sees it, is that small businesses need to spend cash to build operational capacity but this comes at the cost of working capital. “You are spending all of your cash flow buying inventory to sell. Where do you get the cash flow to build all of this,” Mathur exclaimed, adding that Razorpay’s Agent Studio is planning to plug this serious gap. 

The Agent Studio launches with four production-ready agents. 

The Abandoned Cart Conversion Agent steps in when a customer drops off mid-checkout, initiating a voice-led conversation, understanding why the buyer left, and offering an incentive such as a loyalty discount before sending a fresh payment link. 

The Dispute Responder Agent handles chargeback investigations, gathering evidence and filing responses autonomously. The Subscription Recovery Agent works to re-engage lapsed subscribers. And the Cashflow Forecaster Agent analyses payment patterns to anticipate liquidity gaps before they become crises.

Mathur explains that using any of this requires no technical skill whatsoever. There is no code, no configuration panel, no technical onboarding required. “A very small business from any small town can do it. You don’t even need to type it in English. You can speak and tell it that this is what I want, and it will get done.”

The cost architecture is designed to match the realities of small business economics. Unlike a salaried operations team that draws a fixed cost regardless of business activity, the agents are on-demand — they activate only when triggered by a transaction or an event, and stand down when idle.

“In an operations team, you have to incur the fixed cost. Maybe you hire 20 people, maybe no orders come in. Or maybe you have a festive jump — it went up in December, but you are paying this cost in March and nobody is buying.”

More Control, Customisation For Small Businesses 

The agent model flips this entirely. Agents only come online when an order comes in and when it is cost effective for the business. And when the task is completed, the agent goes offline. This allows smaller businesses to optimise costs, and use single-purpose and token-efficient agents instead of multipurpose products.

Looking ahead, Razorpay plans to give merchants direct control over which AI model powers their agents — and therefore, the cost control over their tailored use.

Mathur exemplifies this with a contrast: a jewellery retailer handling transactions worth ₹50,000 apiece may want the most capable model available, while a T-shirt seller moving inventory at ₹200 per unit needs something different entirely.

“The second merchant wants to use a cheaper model. They are okay with a 60% success rate, but they want a cheaper model because they can’t afford to give up too much margin. The customer will have control over the cost — you choose the model you want, you choose the cost at which you can operate, and you get the capability accordingly.”

The Agent Studio also opens the platform to custom agent creation. Businesses can describe what they want an agent to do in plain English, connect it to relevant systems — whether Shopify, WhatsApp, Tally, Quickbooks, or Slack — and deploy it without writing a single line of code. 

Mathur describes this as a deliberate transformation of the company’s identity. “It transforms us from being just a fintech financial rail provider to being an operating system for businesses.”

This move could deepen Razorpay’s integration with merchants, embedding the platform more deeply into their daily operations. And in doing so, it creates the potential for recurring revenue streams tied not just to payment processing but to AI-driven automation.

From Payments To An OS For Businesses

“Not just the payment,” Mathur explains. “We are building the entire commerce infrastructure.”

Besides B2B Agentic AI products, Razorpay has also introduced its consumer-facing agentic commerce capabilities. Its AI agents don’t just recommend products but can complete the entire transaction on behalf of users.

Razorpay has already begun piloting this with several consumer platforms, including food delivery and entertainment apps. The startup is currently working with partners including major consumer platforms, multiplex chains, and D2C brands to integrate these agentic experiences into their apps.

In one demonstration, a user opens a food delivery app and simply speaks their order in their preferred language. The AI agent interprets the request, finds the items, places the order, and completes the payment automatically.

For Razorpay, that shift would place the company deeper inside the commerce journey itself. Payments would no longer be the final step in a transaction; it would be embedded within an AI-driven workflow that begins with intent.

And the Razorpay CEO is confident that whoever owns that workflow effectively controls the transaction layer.

The company completed its reverse flip — redomiciling back to India — and is all set to return to profitability in FY26, Mathur claimed.

“When you take out that one-time cost of the reverse flip, Razorpay was profitable — that profitability streak is maintained throughout. We hope the financials will continue to improve from here as a company that’s continuously growing.” the CEO added.

The AI pivot, in this context, could also be a valuation story a fintech giant will tell to public markets when it goes for an IPO sometime later this year or in 2027. 

Globally, AI has become central to investor conversations and enterprise adoption. Razorpay seemingly is attempting to follow a similar trajectory calibrated to the realities of Indian SME and digital economy.

“India is the largest market for AI companies after the US. The opportunity for AI is really big. What we are trying to do is go through these companies and say: how can we really deploy this reasoning to solve problems that have not been solved? India-specific core problems.” Mathur explained. 

Razorpay’s role, as Mathur frames it, is not to build the foundational AI models — that is Anthropic’s and OpenAI’s domain — but to be the infrastructure layer that applies those models to India-specific commerce problems at scale. 

For investors evaluating the IPO, this transformation narrative addresses what has historically been a challenging question about Indian fintech companies, which is  to choose revenue pools beyond margin-thin payments. 

When asked what is the long-term growth engine once the addressable payment market matures, Razorpay’s answer is: merchant base, but with a dramatically expanded scope of services and a correspondingly higher revenue per merchant.

“Short-term changes are not something that we really focus on. For us it’s more about: how do we stand in the public markets after that.” Razorpay CEO said.

What has become increasingly clear is that Razorpay is building the case that it is not a payments company seeking a listing — not unlike Pine Labs in that regard. It wants to be known as an AI-native company and an operating system for businesses with payments at its heart. 

[Edited By Nikhil Subramaniam]

 

The post Razorpay’s Biggest Bet: From Payments To Becoming The AI Brain For India’s Small Businesses appeared first on Inc42 Media.