How Razorpay Built India's First Fully Agentic Payments Infrastructure

Executive Summary

Razorpay never set out to be a consumer brand. It set out to be the layer businesses stop thinking about, the infrastructure underneath checkout, payroll, lending, and reconciliation that simply works, invisibly, at whatever scale a business grows into. That invisibility is now compounding into one of the more consequential infrastructure bets in Indian fintech. Razorpay processes an annualized total payment value of roughly 180 billion dollars across more than 12 million merchants, reported FY25 revenue of 3,783 crore rupees, up 65 percent year on year, and completed its reverse flip, relocating its parent entity from the United States back to India, in 2025 as direct preparation for a domestic public listing. In June 2026, the company confidentially filed its Draft Red Herring Prospectus with the Securities and Exchange Board of India, targeting a raise of 5,000 to 6,000 crore rupees and a valuation in the 5 to 6 billion dollar range. Weeks later, at its annual Sprint event, Razorpay unveiled more than 100 product launches under a single framing: the age of agentic payments, in which AI systems do not merely process a transaction a human initiated, but initiate, negotiate, and complete transactions on a business's behalf. This case study examines how Razorpay built that agentic infrastructure layer by layer, what it has produced commercially on the road to a public listing, and what the model reveals for leaders thinking about AI transformation in businesses that compete not on consumer visibility but on operational depth.

‍

The Market Razorpay Was Actually Solving For

India's fintech story is usually told through its consumer-facing winners, the apps people open every day to pay a bill, check a score, or split a dinner. Razorpay was never trying to win that layer. Founded in 2014 by Harshil Mathur and Shashank Kumar, two IIT Roorkee alumni who had personally experienced how difficult it was for a small business to accept a digital payment online, the company built itself around a much less visible but far more structurally important problem: businesses in India, from single-founder startups to enterprise-scale unicorns, needed financial infrastructure that behaved like a global standard while working within India's specific banking, compliance, and settlement rails.

That problem does not reward the same playbook that built Meesho or CRED. A consumer platform wins by shaping a single interface millions of people interact with directly. An infrastructure platform wins by becoming so deeply embedded in a business's operational stack, payments, payroll, lending, banking, reconciliation, that removing it becomes more expensive than staying. Razorpay's expansion from a single payment gateway into a full-stack platform spanning RazorpayX business banking, point-of-sale hardware, payroll, merchant lending, and cross-border payments was not diversification for its own sake. It was the deliberate construction of exactly that kind of embeddedness, where 70 percent of Indian unicorns reportedly now run some part of their finance function through RazorpayX, and where more than 12 million merchants depend on Razorpay for the single most operationally critical function any business has: getting paid.

Serving this market well requires a fundamentally different relationship with AI than a consumer app does. A consumer-facing AI feature has to be delightful, visible, and easy to explain in a product screenshot. An infrastructure-facing AI feature has to be reliable at a scale most consumer products never approach, invisible enough that a finance team barely notices it working, and precise enough that a single hallucinated output in a payroll approval or a payment routing decision could cost a business real money within seconds. Razorpay chose to build its AI architecture for that second, much less forgiving standard, and its 2026 product strategy makes that choice explicit through a single organising idea: agentic AI, not as a feature category, but as the operating model for the entire platform.

‍

The AI Architecture Razorpay Built for a Market That Runs on Trust, Not Attention

Razorpay's AI build spans five distinct layers, each targeting a different point in a business's financial operations: the checkout and discovery layer where a customer's intent turns into a transaction, an operational automation layer that handles disputes, fraud, and returns without human intervention, a business banking layer that runs a company's cash management autonomously, a risk and authentication layer securing the payment gateway itself, and a platform layer that lets any business, developer, or even another AI system plug directly into Razorpay's infrastructure. Understanding each layer separately is useful, but the strategic insight is in how deliberately they were built to interlock into a single agentic stack rather than as five unrelated feature launches.

‍

Agentic Payments for Conversational and LLM-Native Commerce

The most forward-facing piece of Razorpay's 2026 strategy is agentic payments, a set of capabilities built for a world in which a growing share of product discovery and purchase intent happens inside a conversation rather than inside a traditional storefront. Razorpay's infrastructure now lets merchant chatbots complete purchases autonomously inside in-app conversations, enables conversational product discovery and UPI payments directly inside large language models, and, notably, allows any business to upload its product catalogue and go live with a fully native checkout experience inside ChatGPT. A parallel voice payments capability lets customers approve and complete a transaction over a phone call without opening a screen at all.

The strategic logic behind this layer is that the venue where commerce happens is shifting away from a merchant's own app or website and toward whatever AI interface a customer is already using, whether that is a voice assistant, a chatbot embedded in a messaging app, or a general-purpose AI system like Claude or ChatGPT. Razorpay's own public materials frame this directly: it now bridges the gap between discovery and checkout for large merchants including Vodafone and Zepto, providing infrastructure that lets a merchant's own AI assistant handle a transaction for a user precisely at the point where purchase intent is created, rather than requiring the user to navigate away to a separate checkout flow. A business that does not build for this shift risks losing the transaction entirely to whichever competitor's product the AI assistant recommends instead, which makes agentic payment infrastructure less of an optional add-on and more of a defensive necessity for any merchant operating at meaningful scale.

The Agent Studio and Operational Automation

Where agentic payments sits at the front end of a transaction, Razorpay's Agent Studio sits at the back end, automating the operational work that traditionally consumed significant human hours inside a finance or operations team. The Dispute Auto-Responder agent reviews chargeback cases, verifies proof, and submits evidence to protect a merchant's revenue without a human drafting a single response. A Cashflow Insights agent analyses return-to-origin patterns across pincodes, products, and customer segments to identify preventable return drivers before they compound into meaningful revenue loss, a particularly relevant capability for the D2C and e-commerce merchants that make up a significant share of Razorpay's customer base. A Subscription Recovery agent proactively intervenes on failed recurring payments before they translate into customer churn, and a Cashflow Forecaster agent predicts a business's cash position three to seven days ahead, pushing a daily digest covering unified balances, payroll obligations, and potential shortfalls directly to a founder or finance lead.

What makes this layer strategically significant rather than merely convenient is what it signals about where Razorpay believes the actual value of AI sits in financial infrastructure. Consumer fintech companies tend to spend their AI investment on the interface, the part a user sees and interacts with directly. Razorpay spent a comparable share of its AI investment on the part no user ever sees: the operational plumbing that determines whether a dispute gets resolved in a business's favour, whether a return gets prevented before it happens, and whether a founder finds out about a cash shortfall three days before it becomes a payroll crisis rather than three days after.

‍

Agentic Business Banking via RazorpayX

RazorpayX, the company's business banking product, extended the same agentic logic directly into a company's treasury function. Its Insights Agent continuously tracks a business's burn rate, balances, and runway, surfacing risks early enough that a founder can act on them rather than discover them retroactively during a board update. A Receivables Agent tracks unpaid invoices and automatically follows up, including over calls, to accelerate collections without manual chasing. A Payouts Agent lets a finance team simply name who they are paying, after which the agent fetches the relevant details, applies the correct tax deduction history, and moves the transaction to authorisation in seconds rather than the several minutes a manual payout typically requires. A Bookkeeping Agent posts accounting entries directly into a business's ERP system based on predefined rules, removing manual reconciliation, while a Reporting Agent generates financial reports, flags anomalies, and shares downloadable summaries without a finance team member building a spreadsheet from scratch.

Razorpay's own senior leadership has been explicit that this is not a peripheral feature set. Asheesh V, the company's senior director of product for business banking, has stated publicly that RazorpayX is already used by roughly 70 percent of Indian unicorns, framing the agentic layer as an extension of that existing trust relationship rather than a new pitch to a cold audience. The company also launched an AI Payslip assistant that gives employees instant clarity on their salary, deductions, and taxes without needing to raise an HR query, and a Payroll Approvals Agent that autonomously approves leave and reimbursement requests within policy limits, materially reducing the operational bottleneck that reimbursement approvals typically create inside growing companies.

‍

Biometric and Intelligent Risk Infrastructure at the Gateway Layer

Underneath all of the customer-facing and operations-facing AI sits a risk and authentication layer that most merchants and their customers will never directly notice, which is precisely the point. Razorpay introduced biometric card authentication, shifting card payment approval from OTP-based verification to RBI-compliant fingerprint or facial verification, directly addressing the same OTP interception and SIM-swap fraud vectors that have plagued digital payments broadly across the Indian market. A Smart AML Risk Screening system built into Razorpay's international payments infrastructure predicts anti-money-laundering risk early enough to prevent bank-level settlement delays, and a chargeback fraud protection system draws on data from more than 10 million international cards and over 2 billion transactions to reduce fraud-related disputes before they occur rather than resolving them after the fact.

Perhaps the most quietly significant piece of this layer is Razorpay's intelligent routing infrastructure, which uses real-time learning to send every payment through whichever processing route gives it the best chance of approval, reducing unnecessary declines that would otherwise cost a merchant a completed sale. In January 2026, the Reserve Bank of India granted Razorpay a Payment Aggregator Cross-Border licence, formally authorising both inward and outward international payment aggregation, a regulatory milestone that directly extends the reach of this AI-driven routing and risk infrastructure into cross-border commerce, an increasingly important growth vector as Indian D2C brands and SaaS companies sell into international markets.

‍

Ray, the Agentic Platform Layer

The final piece of Razorpay's architecture is arguably its most structurally ambitious: an agentic platform layer, internally branded Ray, that treats Razorpay's own infrastructure as something other AI systems and developers can act on directly rather than merely integrate with through a traditional API. Ray Smart Assist functions as a product finder built specifically to understand a business's specific payment needs rather than presenting a generic menu of options. Agentic Onboarding uses AI to complete know-your-customer verification and account activation in minutes rather than the days a manual onboarding process traditionally required. Ray Customer Support resolves merchant issues around the clock without tickets or human handoffs, and an Agentic Dashboard lets a business owner ask direct questions about their account, resolve disputes, and make configuration changes conversationally rather than navigating a traditional settings interface.

Most significantly, Razorpay became India's first payment gateway to launch a Model Context Protocol server, letting AI systems, including Claude directly, manage payments, resolve disputes, and track performance from inside an AI conversation rather than requiring a human to log into a separate dashboard. Combined with dedicated payment integration nodes for developer platforms including n8n, Replit, and Vercel, this platform layer represents a bet that the next generation of businesses will be built by AI-assisted developers and, increasingly, by AI agents themselves, and that Razorpay's infrastructure needs to be legible to those agents in the same way it has historically been legible to human developers.

‍

The Business Outcomes

Razorpay's financial trajectory reflects a business that has spent a decade building toward exactly the kind of embedded, high-retention infrastructure position its 2026 agentic strategy is now designed to extend. Consolidated operating revenue reached 3,783 crore rupees in FY25, up 65 percent from 2,296 crore rupees in FY24, driven by growth across the payment gateway, RazorpayX, point-of-sale, and international businesses. Gross profit grew 41 percent year on year to 1,277 crore rupees, up from 906 crore rupees the previous year. Annualized total payment value crossed 180 billion dollars, and the company's online payments segment, its largest and most mature business line, turned EBITDA-positive during the year, with founder and chief executive Harshil Mathur stating publicly that newer business lines were rapidly scaling and unlocking additional growth vectors beyond the core payment gateway.

The company did report a consolidated net loss of 1,209 crore rupees in FY25, but that figure requires context rather than a surface reading. The loss was driven largely by one-time ESOP-related expenses and tax liabilities directly tied to Razorpay's reverse-flip process, the corporate restructuring required to relocate its parent entity from the United States back to India, rather than by any deterioration in the underlying operating business. Management has indicated that consolidated profitability is targeted within two to three quarters of the core India business turning profitable, a considerably shorter runway than the net loss figure alone would suggest to an observer unfamiliar with the accounting mechanics of a reverse-flip transaction.

Razorpay now serves more than 12 million merchants, maintains a reported 94 percent merchant retention rate, and holds roughly 55 percent share of India's online payment gateway market, alongside processing an estimated 35 million UPI transactions daily. The company has raised a total of approximately 742 million dollars in funding across 12 rounds from investors including GIC, Peak XV Partners, Tiger Global, Ribbit Capital, TCV, Matrix Partners, Lightspeed Venture Partners, and, notably, Y Combinator, whose acceptance of Razorpay into its accelerator programme in the company's early years was a pivotal moment in its access to both capital and banking infrastructure relationships it had initially struggled to secure on its own.

‍

The Reverse Flip and the Road to IPO

Razorpay's decision to complete a reverse flip in 2025, relocating its parent holding entity from the United States back to India, was not a routine corporate housekeeping exercise. It was a direct, structurally necessary precondition for the company's planned Indian public listing, and it places Razorpay alongside a broader wave of Indian startups that had originally incorporated abroad during an earlier funding era and have since worked through the complex, often costly process of redomiciling in order to list on Indian exchanges. The process directly explains the bulk of Razorpay's reported FY25 net loss, since ESOP-related expenses and tax liabilities tied to the restructuring flow through the same income statement as the company's underlying operating performance, even though the two have essentially nothing to do with each other commercially.

On June 12, 2026, Razorpay confidentially filed its Draft Red Herring Prospectus with the Securities and Exchange Board of India, using the confidential pre-filing route that allows a company to share sensitive business and financial data with the regulator without making it public until closer to the actual listing. Reports around the filing indicate the company is targeting a raise between 5,000 and 6,000 crore rupees, structured as a mix of fresh issue and an offer for sale by existing shareholders, with Axis Capital, Kotak Mahindra Capital, JPMorgan, and Citi appointed as lead managers. The targeted valuation range of 5 to 6 billion dollars represents a markdown from the company's 2021 peak private valuation of 7.5 billion dollars, a compression that reflects both the broader recalibration of fintech valuations globally since the 2021 funding peak and a public market environment that, as one market analyst put it in coverage of the filing, now demands demonstrated profitability alongside growth rather than growth alone.

That the IPO, the reverse flip, and the Sprint 2026 agentic product launch all landed within roughly the same window is itself a strategically coherent sequence rather than a coincidence. A company preparing to face public market scrutiny for the first time has every incentive to walk into that process with its most differentiated, most defensible growth story fully articulated, and an AI-native, agentic positioning gives Razorpay a materially different narrative to bring to public investors than a straightforward payment gateway growth story would on its own.

‍

Where the Model Faces Structural Pressure

The scale and sophistication of Razorpay's agentic build are genuine, and the pressure building around the business as it approaches a public listing is equally genuine.

The consolidated net loss, even with the reverse-flip context fully accounted for, means Razorpay is walking into public market scrutiny without a clean profitability story at the consolidated level, and public investors, as multiple analysts covering the DRHP filing have noted, are considerably less forgiving of growth-without-profit narratives in 2026 than they were during the funding environment that produced Razorpay's 2021 peak valuation. The gap between the company's targeted 5 to 6 billion dollar IPO valuation and its 2021 peak of 7.5 billion dollars is itself a visible marker of that recalibrated investor appetite.

Competitive intensity in Indian payment infrastructure has not slowed. PayU, Cashfree, and a resurgent Paytm continue to compete directly for the same merchant relationships Razorpay has built its business around, and the emergence of agentic commerce as a category means Razorpay's early-mover advantage in agentic payments and Agent Studio is unlikely to remain uncontested for long, particularly given how directly replicable individual features like dispute automation or cashflow forecasting are once a competitor decides to prioritise them.

The sheer breadth of Razorpay's 2026 product launch, more than 100 individual updates spanning six distinct business lines, carries its own execution risk. A company can build agentic infrastructure quickly, but sustaining reliability across that many autonomous, revenue-adjacent systems simultaneously, particularly ones making decisions inside a business's core financial operations, is a considerably harder operational discipline to maintain than building the systems in the first place. A single high-profile failure inside an autonomous payout agent or a bookkeeping agent, in a business category built entirely on trust, would carry reputational cost disproportionate to the technical scope of the failure itself.

Finally, Razorpay's infrastructure positioning, while commercially powerful, is also inherently less visible to the end consumers whose behaviour ultimately drives merchant revenue and, by extension, Razorpay's own growth. Unlike Meesho or CRED, which can point directly to consumer engagement metrics as evidence of product-market fit, Razorpay's growth is entirely dependent on its merchants' own growth and retention, meaning any broader slowdown in Indian D2C, SaaS, or startup activity would compress Razorpay's growth indirectly, through its customers, rather than through any direct weakness in Razorpay's own product.

‍

Strategic Implications for Leaders

1. Infrastructure businesses should build AI for reliability at scale before building it for visibility. Razorpay's most significant AI investment sits in places a user or merchant rarely sees directly, dispute resolution, cashflow forecasting, payout routing, precisely because that is where autonomous decision-making carries the highest operational stakes and the least tolerance for error. Leaders building AI into infrastructure-layer products should resist the pull toward AI features that demo well and instead prioritise the unglamorous operational layer where reliability compounds into genuine switching costs for a customer.

2. The venue of commerce is migrating faster than most merchant-facing businesses have adjusted for, and infrastructure providers that build for that migration early capture disproportionate strategic leverage. Razorpay's bet on agentic payments inside chatbots, LLMs, and voice interfaces is a direct response to the recognition that purchase intent is increasingly forming inside AI conversations rather than inside a merchant's own storefront. Leaders should audit where their own customers' purchase or decision intent is actually forming today, not where it formed two years ago, and build infrastructure that meets that intent at its actual point of origin.

3. A regulatory and corporate restructuring milestone, like a reverse flip, should be sequenced deliberately alongside a company's most differentiated growth narrative, not treated as a separate, purely administrative workstream. Razorpay's decision to complete its reverse flip, file its DRHP, and unveil its most ambitious AI product suite within the same several-month window meant the company walked into public market preparation with its strongest possible story fully assembled rather than fragmented across a longer, less coordinated timeline. Leaders preparing a company for a major capital markets event should treat product strategy and corporate structuring as a single coordinated narrative rather than two independently managed tracks.

4. Agentic AI creates genuine competitive differentiation only when it is built as an interlocking system rather than a scattered feature list. Razorpay's more than 100 individual launches at Sprint 2026 could easily have read as feature sprawl. What made the announcement strategically coherent instead was that every layer, checkout, operations, banking, risk, platform, fed into a single agentic thesis: a business's financial infrastructure should think, decide, and act alongside it rather than simply process what it is told to. Leaders should evaluate their own AI roadmaps against that same coherence test, whether each new AI feature strengthens a single unifying thesis about how the product works, before greenlighting it.

5. Infrastructure providers carry a structurally different kind of AI risk than consumer platforms, and that risk should shape both governance and rollout pace. A hallucinated recommendation on a consumer app is embarrassing. A hallucinated decision inside an autonomous payout agent or a bookkeeping agent moves real money incorrectly, inside a customer's core financial operations, in a category built entirely on trust. Leaders deploying agentic AI into financial or operational infrastructure should build materially more conservative rollout, monitoring, and rollback discipline than they would for a consumer-facing AI feature, since the cost of a rare failure scales directly with the financial stakes of the decision the agent is making.

‍

Final Thoughts 

Razorpay's story is, in many ways, the least glamorous of the AI-native platform stories to come out of Indian fintech in the past several years, and that is exactly why it is worth studying closely. The company never had a consumer moment the way Meesho's personalised feed or CRED's credit coach did. It built, instead, a decade-long position as the infrastructure businesses stop thinking about, and it is now using that position as the foundation for one of the more ambitious agentic AI bets in the market, one built not around a single flagship feature but around a coordinated architecture spanning checkout, operations, banking, risk, and platform access simultaneously. The reverse flip, the confidential DRHP, and the Sprint 2026 launch arriving within months of each other were not separate events. They were the deliberate assembly of a single growth story, timed precisely for the moment Razorpay needs public market investors to believe it. Whether that story holds under the scrutiny a public listing invites, particularly given the consolidated net loss and the compressed valuation relative to 2021, will be one of the more closely watched tests of how public markets price agentic AI infrastructure businesses in India, as distinct from the consumer AI stories that have so far dominated that conversation.

Frequently Asked Questions

What is agentic payments and how does Razorpay use it?

Agentic payments refers to AI systems that can initiate, negotiate, and complete a transaction on a business's behalf, rather than simply processing a transaction a human already started. Razorpay's agentic payments infrastructure lets merchant chatbots complete purchases autonomously, enables conversational shopping and UPI payments directly inside large language models, and allows businesses to sell directly inside platforms like ChatGPT using Razorpay's checkout infrastructure.

‍

Is Razorpay profitable?

Razorpay's core online payments business turned EBITDA-positive in FY25, though the company reported a consolidated net loss of 1,209 crore rupees for the year, driven primarily by one-time ESOP expenses and tax liabilities tied to its 2025 reverse-flip restructuring rather than underlying operating performance. Management has targeted consolidated profitability within two to three quarters of the India business achieving it.

‍

When is the Razorpay IPO?

Razorpay confidentially filed its Draft Red Herring Prospectus with SEBI on June 12, 2026, targeting a raise of 5,000 to 6,000 crore rupees and a public listing by the end of 2026, subject to regulatory approval.

‍

What is Razorpay's Agent Studio?

Agent Studio is Razorpay's suite of autonomous AI agents that handle operational tasks for merchants, including responding to payment disputes, forecasting cash flow, identifying preventable return patterns, and recovering failed subscription payments, without requiring manual intervention from a business's finance or operations team.

‍

How big is Razorpay's business?

Razorpay processes an annualized total payment value of approximately 180 billion dollars, serves more than 12 million merchants, holds roughly 55 percent share of India's online payment gateway market, and reported FY25 revenue of 3,783 crore rupees, up 65 percent year on year.

‍

Why did Razorpay complete a reverse flip?

Razorpay relocated its parent holding entity from the United States back to India in 2025 as a direct precondition for pursuing a public listing on Indian stock exchanges, a restructuring process that several other Indian startups originally incorporated abroad have also undertaken ahead of domestic IPOs.

‍

Read our Other Case Studies

How Meesho Built India's Most AI-Native Commerce Platform for Bharat

‍