Razorpay Agent Studio: The AI OS for Indian SMBs

Razorpay’s Biggest Bet: From Payments Provider to AI-Powered Operating System for India’s Small Businesses

Soumya Verma
22 Min Read
Quick Take:

  • The Announcement: Razorpay launched Agent Studio and the Agentic Experience Platform at FTX 2026 (March 12, Bengaluru) — an AI agent marketplace and builder built entirely on Anthropic’s Claude Agent SDK. Positioned as the world’s first agent studio built natively on payment infrastructure
  • What It Is: A B2B marketplace where businesses deploy purpose-built AI agents to handle post-payment operations: dispute management, cart recovery, subscription retry, cashflow forecasting, RTO reduction — without writing a single line of code. No-code agent builder also in beta
  • Production-Ready Agents at Launch: (1) Abandoned Cart Conversion (via WhatsApp/voice + payment link) | (2) Dispute Responder (auto-collects evidence, submits to gateway) | (3) Subscription Recovery (smart retry logic + customer nudges) | (4) Cashflow Forecaster (3-7 day prediction, payroll risk alerts) | (5-8) Four more agents at launch
  • Agentic Commerce Pilots: Zomato, Swiggy, PVR INOX, Vodafone Idea, Bluestone, Honasa (The Derma Co) — enabling in-app conversational commerce where users discover, decide, and pay inside one chat interface. First piloted at GFF 2025 on Vodafone Idea
  • IPO Timeline: Reverse flip to India completed March 2025. Converted to public entity April 2025. CFO targets IPO within FY27. Axis Capital, Kotak, JP Morgan, Citibank selected as advisers. IPO size expected to exceed $700 Mn. IPO a ‘few quarters away’ per Shashank Kumar (March 2026)
  • FY25 Financials: Revenue Rs 3,783 Cr (+65% YoY from Rs 2,296 Cr). Core payments business: EBITDA-profitable. Net loss attributable to Rs 1,209 Cr ESOP expense + Rs 1,245 Cr restructuring/tax costs from reverse flip. Valuation: $9.2 Bn (Series G, June 2025)

Twelve years ago, Razorpay was rejected by over 100 investors before getting its first cheque from Y Combinator. The pitch was simple: make it easier for Indian startups to accept online payments. It worked. Today, Razorpay processes $180 Bn in annual payments volume, serves 8 million businesses, has raised $742 Mn across 11 rounds at a $9.2 Bn valuation, and is preparing for one of the most anticipated fintech IPOs in India’s market history. The payments problem is largely solved.

So what’s the next act? At FTX 2026 on March 12, 2026 in Bengaluru, Razorpay co-founder and CEO Harshil Mathur gave the clearest answer yet: Razorpay is becoming the AI operating system for India’s 8 million small businesses. The vehicle is Agent Studio — a B2B marketplace of autonomous AI agents built on Anthropic’s Claude Agent SDK — and a companion Agentic Experience Platform that reimagines how merchants onboard, integrate, and manage their entire financial stack. Paired with live agentic commerce pilots on Zomato, Swiggy, PVR INOX, and Vodafone Idea, Razorpay is making the most consequential product bet in its history — and it is timed precisely to reshape the company’s IPO narrative.

StartupFeed Insight

The strategic logic is elegant — and the timing is deliberate.

Razorpay’s core payments business is already EBITDA-profitable. The problem is that India’s payment processing market is also becoming a commodity: UPI rails are free, MDR (Merchant Discount Rate) is zero on UPI, and Stripe, PayU, BharatPe, and PhonePe are all competing for the same merchant base. A pure-play payment gateway with 65% revenue growth is a good business. A pure-play payment gateway with 65% revenue growth is not a $9.2 Bn business — not at the multiples the Indian public market will assign to a fintech.

Agent Studio changes the multiple. If Razorpay can demonstrate that it is not a payment gateway but an AI-powered business operating system that happens to process payments, the revenue quality improves dramatically: from transaction-fee revenue (low-margin, commoditised) to software/SaaS revenue (high-margin, recurring, sticky). The market will price a business operating system at 8-12x revenue. It will price a payment gateway at 3-5x. On Rs 4,000 Cr of revenue, that difference is worth Rs 20,000-35,000 Cr in market capitalisation.

The risk: India’s 8 million SMB merchants are not the same as the 450,000 active enterprise customers who generate the bulk of Razorpay’s revenue. Getting a Rs 50 Cr D2C brand to use Agent Studio for dispute management is very different from getting a 10-person textile trader in Surat to use an AI cashflow forecaster. Shashank Kumar himself acknowledged at FTX 2026 that SMB adoption of agentic payments is ‘still to play out’. The IPO narrative will depend on how many of those 8 million merchants actually activate agents, not just how many have access.

Agent Studio: The World’s First AI Marketplace Built on Payment Infrastructure

The premise of Agent Studio is direct: payments are only the execution layer of commerce. The real work — and the real friction — lives in everything that happens around the transaction: disputes, reconciliation, failed payment recovery, subscription lapses, refund processing, chargeback defence, and cashflow management. These workflows span multiple systems and require dedicated teams. Large enterprises have those teams. India’s 8 million small businesses do not.

Agent Studio deploys autonomous AI agents that observe Razorpay’s transaction data, reason across connected systems, and execute complex financial workflows without human intervention. Built on Anthropic’s Claude Agent SDK and integrated with Shopify, WhatsApp, Shiprocket, Slack, QuickBooks, and Salesforce, the agents have both the data access and the execution capability to handle end-to-end workflows — from identifying a disputed transaction to submitting the chargeback defence document to the payment gateway, all without a human touching the keyboard.

The 8 Production-Ready Agents at FTX 2026 Launch

Agent What It Does High-Friction Problem Solved
Abandoned Cart Conversion (SuperU) Identifies abandoned checkouts, re-engages via WhatsApp or voice, offers loyalty discounts, sends payment link. Can apply personalised nudges before completing purchase Checkout abandonment: 60-70% of Indian e-commerce carts are abandoned. Manual follow-up is impossible at scale for small merchants
Abandoned Cart Conversion (Nugget by Zomato) Same workflow as above but powered by Zomato’s Nugget AI layer for food commerce-specific context and recommendations Variant targeting food commerce merchants on Zomato’s merchant ecosystem
Dispute Responder Auto-collects transaction evidence (payment logs, delivery confirmation, order details, logistics proof), compiles optimised chargeback response, submits before deadline. Evaluates dispute win probability from historical patterns Chargebacks: merchants have a narrow window to respond with evidence. Most small businesses miss it — no one is monitoring and manual compilation takes hours
Subscription Recovery Analyses failed subscription payments, applies smarter retry logic across instruments, triggers targeted customer nudges via appropriate channels (WhatsApp, email, in-app) Involuntary churn from failed payments: up to 20-30% of subscription lapses are involuntary (expired cards, insufficient funds at billing time)
Cashflow Forecaster Predicts cash position 3-7 days ahead with alerts for payroll risk, settlement shortfalls, and payout failures. Uses Razorpay transaction data for prediction Working capital management: most SMBs cannot forecast next-week cash position accurately; payroll surprises are the #1 cause of SMB financial distress
Payment Reconciliation Agent Uploads bank statements, cross-references against Razorpay settlement reports, flags outstanding items for human review. Replaces manual Excel reconciliation Reconciliation: businesses spend hours downloading reports, comparing settlements, finding mismatches. Error-prone, time-consuming, no one wants to do it
RTO (Return-to-Origin) Reduction Agent Analyses COD (Cash on Delivery) order patterns, identifies high-risk orders, triggers pre-delivery verification flows to reduce failed deliveries RTO: Indian e-commerce has 25-35% COD return rates; each failed delivery costs Rs 150-300 in logistics + handling; RTO is the biggest margin killer for D2C brands
No-Code Custom Agent Builder (Beta) Describe the task in plain English, select the systems the agent can access, set rules — agent is created instantly. No coding required Long tail of business-specific workflows that no pre-built agent covers. Opens agent creation to non-technical operations and finance teams

The Agentic Experience Platform: Reimagining Merchant Onboarding and Operations

Alongside Agent Studio, Razorpay launched the Agentic Experience Platform — an AI-native redesign of how merchants interact with Razorpay at every touchpoint, from first sign-up to daily operations management.

Capability What Changed
Agentic Onboarding Merchant onboarding time reduced from 30-45 minutes to approximately 5 minutes. AI automates identity verification through government infrastructure (Aadhaar, PAN, GST databases). Zero manual document upload for standard onboarding
Agentic Dashboard Merchants manage payment operations through natural language. Example: upload a bank statement and type ‘reconcile against my Razorpay settlements’ — done instantly. No more navigating dashboards or exporting CSVs for routine operations
Agentic Integration Developers integrate Razorpay in under 10 minutes across AI coding environments including Claude Code and no-code platforms including Replit. Harshil Mathur demonstrated live at FTX 2026 that complete payment integration takes under 2 minutes via AI-assisted setup
AI Fraud Detection Every single transaction on Razorpay is now monitored by an AI agent that identifies whether the pattern matches known fraud profiles and blocks it before it processes. Mathur: ‘35% of payment failures in India occur due to delayed or incorrect OTPs’ — AI authentication eliminates many of these
Multi-Model Architecture (Roadmap) Currently built on Anthropic’s Claude Agent SDK exclusively. Razorpay plans to expand to multiple AI models, allowing merchants to choose models optimised for specific use cases (customer conversation, financial forecasting, fraud detection, code generation)

Agentic Commerce: When the Payment Disappears Into the Conversation

The second major bet at FTX 2026 was agentic commerce — Razorpay’s vision for what shopping looks like when AI agents replace menus, filters, and checkout forms. Harshil Mathur explained it directly: ‘You can open the Zomato app, chat with it and say, “Hey, I want samosa from this, I want chai from this”, and the Zomato app can buy it for you. Completely autonomous just by chatting on it.’

This is not a feature. It is a redefinition of what checkout means. In the agentic commerce model, discovery, decision, and payment collapse into a single conversational flow. The user never sees a product grid, a price filter, or a payment form. They express intent; the agent resolves it. Razorpay’s infrastructure sits at the payment execution layer, invisible but essential.

Pilot Partner Agentic Commerce Integration
Zomato User chats inside Zomato app to order food; AI recommends, user confirms, payment completes in single conversational flow. No menu navigation, no checkout page
Swiggy Similar in-app conversational food ordering + payment pilot
PVR INOX In-app movie ticket discovery and booking via conversational interface. Pilot enables ‘find me two seats for Avengers on Saturday evening near Connaught Place’ to complete booking and payment in one chat session
Vodafone Idea First pilot (GFF 2025): AI understood user’s usual recharge pattern, recommended the most relevant plan, completed payment within the same conversational flow. Zero navigation to a recharge page
Bluestone Conversational jewellery discovery and purchase intent capture inside app. High-consideration purchase category testing agentic recommendation + payment bridge
Honasa (The Derma Co) Conversational skincare recommendation + checkout within app. D2C brand testing AI-led product discovery leading directly to purchase
ChatGPT + UPI (NPCI Partnership) Razorpay partnered with NPCI and OpenAI (Oct 2025) to enable agentic payments via ChatGPT using UPI Circle and UPI Reserve Pay. External AI assistants can now complete purchases through Razorpay’s UPI infrastructure

 

Security architecture for agentic payments:

The agent never sees or stores the user’s financial credentials. Bank balance, UPI handle, card details — all remain with the user or the merchant platform. The agent can initiate a payment transaction only after receiving explicit user approval, and only within pre-authorised limits and controlled payment environments. Razorpay has built this architecture to comply with RBI’s Authentication Mechanisms for Digital Payment Transactions Directions, 2025.

The IPO Narrative: How Agent Studio Changes the Valuation Math

Razorpay’s IPO preparation is the most important context for understanding the Agent Studio launch. The company is not launching AI agents because it is fascinated by technology. It is launching AI agents because the IPO market will value an AI-powered business operating system at a fundamentally different multiple than a payment gateway.

Financial Metric FY25 Actual FY26 Trajectory
Revenue from Operations Rs 3,783 Cr (+65% YoY) Tracxn reports Rs 3,930 Cr; EBITDA+ on core payments
Net Loss (Reported) Loss attributed to Rs 1,209 Cr ESOP + Rs 1,245 Cr restructuring/reverse flip costs One-time costs eliminated; India biz profitable by FY26 end per CFO
Core Payments Business EBITDA-profitable and generating strong cash flows (per CEO Mathur) EBITDA margin expanding as infrastructure costs amortise
Consolidated Profitability Not yet — dragged by non-payments businesses (RazorpayX, international, new verticals) Expected 2-3 quarters after India business profitability — i.e., H1 FY27
Annual Payments Volume $180 Bn processed $1 Tn long-term target
Valuation $9.2 Bn (Series G, June 2025) ~$12-15 Bn expected IPO target range (industry estimates)
IPO Timeline Reverse flip complete March 2025; public entity April 2025 FY27 IPO window (Q3-Q4 FY27 = Oct 2026 – Mar 2027). Advisers: Axis Capital, Kotak, JP Morgan, Citibank
IPO Size Expected >$700 Mn 10% minimum dilution on ~$12-15 Bn valuation = $1.2-1.5 Bn potential
Agent Studio Revenue Model Token-based: smaller tasks near-free; complex tasks priced by tokens/intelligence used New high-margin SaaS revenue stream alongside payment processing take rate
Why Agent Studio changes the IPO multiple:

  • Payment gateway revenue: 0.15-2% of GMV. Volume-dependent, MDR-compressed, commoditised. Public market assigns 3-5x revenue multiple to pure-play gateways
  • SaaS/AI platform revenue: Recurring, per-merchant, high-gross-margin. Public market assigns 8-15x revenue multiple. Stripe trades at ~10x revenue; ServiceNow at 12x+
  • The narrative shift: Every rupee of Agent Studio revenue carries higher margin and higher public market multiple than a rupee of payment processing revenue. Even if Agent Studio generates only Rs 200-300 Cr by IPO, it changes the story from ‘payment gateway going public’ to ‘AI-powered fintech operating system going public’
  • Comparable public market comp: Adyen (Netherlands): payment platform that expanded into financial management tools. Trades at 8-10x revenue. Square/Block: payments + business banking + AI tools. Razorpay’s IPO positioning is explicitly Adyen/Block-comparable, not pure-gateway comparable

Razorpay: Company at a Glance

Parameter Details
Founded 2014 by Harshil Mathur (CEO) and Shashank Kumar (MD) | IIT Roorkee graduates | Y Combinator W15 batch
HQ Bengaluru, Karnataka (redomiciled from US to India, March 2025)
FY25 Revenue Rs 3,783 Cr (+65% YoY) | Payments: EBITDA-profitable | Consolidated: loss due to one-time ESOP + reverse flip costs
Valuation $9.2 Bn (Series G, June 2025)
Total Funding Raised $742 Mn across 11 rounds | Lead investors: Peak XV Partners, Y Combinator, Tiger Global, GIC, Sequoia Capital, Ribbit Capital
Payments Volume $180 Bn+ annually processed
Merchants Served 8 Mn businesses | 450,000+ active enterprise accounts
Employees 3,028 (Aug 2025, +12% YoY)
Core Products Payment Gateway | RazorpayX (business banking, payroll, vendor payments) | Payment Links | Subscriptions | Smart Collect | International Payments | POS
New AI Products Agent Studio (8 production-ready agents) | Agentic Experience Platform (onboarding, dashboard, integration) | Agentic Commerce (6 pilot partners)
AI Infrastructure Built on Anthropic Claude Agent SDK | Integrated with Shopify, WhatsApp, Shiprocket, Slack, QuickBooks, Salesforce, ElevenLabs | NPCI + OpenAI partnership for ChatGPT-based agentic payments
IPO Status Reverse flip complete. Public entity since April 2025. IPO advisers: Axis Capital, Kotak Mahindra Capital, JP Morgan, Citibank. Target: FY27 listing. Size: >$700 Mn
Competitive Landscape Primary: Stripe (global) | PayU (Prosus) | PhonePe | BharatPe | Cashfree | Juspay | PayTM Payment Services
Geopolitical Watch Middle East expansion plans on hold due to West Asia conflict. Monitoring India-Gulf business linkages

The Risks: What Could Go Wrong

  • SMB adoption is the hard problem. Razorpay’s 8 million merchants include a vast long tail of small traders, kirana stores, and micro-businesses that have never used a dashboard, let alone an AI agent. Shashank Kumar’s own assessment: ‘It is still to play out.’ Agent Studio’s IPO impact depends on activation rates among the 450,000 active enterprise clients, not the full 8 million
  • RBI caution on AI in financial services. The RBI has been cautious about autonomous AI acting on financial tasks. Agentic payments raise accountability questions: if an AI agent executes an incorrect payment, who bears liability — the merchant, Razorpay, or Anthropic? Regulatory clarity is still developing
  • Claude dependency. Agent Studio is built entirely on Anthropic’s Claude Agent SDK. While Razorpay plans to expand to multi-model in future, the current architecture creates a single-vendor dependency for a mission-critical financial operations platform. Anthropic’s pricing changes or availability issues could be disruptive
  • Competitive response. PhonePe, PayTM, and Cashfree will all respond with their own AI agent offerings. The Razorpay distribution advantage — 8 million merchants already on the platform — is real but not permanent. Agents are the competition’s easiest product to copy
  • IPO window uncertainty. West Asia conflict has disrupted markets and delayed multiple Indian tech IPOs (PhonePe deferred its listing on March 19, 2026 citing global market volatility). Razorpay’s FY27 IPO window assumes market stability that is not currently guaranteed

What’s Next

The next 12 months are the execution window for Razorpay’s AI strategy. Agent Studio is in beta with early access signups and will move to the token-based paid model once the product’s value is demonstrated in production at scale. The agentic commerce pilots with Zomato, Swiggy, and PVR INOX will produce real data on whether India’s consumers actually want to order via chat rather than tap. And the IPO clock is ticking — every product announcement at FTX 2026 was, in part, a D-Street pitch.

Harshil Mathur’s long-term vision is unambiguous: ‘We’re enabling companies to deploy AI agents that can understand and monitor their revenue flows, resolve payment issues, and unlock insights across billions of transactions in real time.’ That is a description of a financial intelligence platform, not a payment gateway. Whether the market agrees — and whether India’s 8 million small businesses actually adopt the agents in meaningful numbers before the IPO — will determine whether Razorpay’s biggest bet becomes its biggest win.

Share This Article

Don’t Miss Startup News That Matters

Join thousands of readers getting daily startup stories, funding alerts, and industry insights.

Newsletter Form

Free forever. No spam.