Razorpay’s CEO Says Even AI Titans Haven’t Mastered Scaling the Tech

When you walk into a fintech conference these days, you’ll hear the same refrain over and over: artificial intelligence is the next big thing. It’s a line that even the most polished slides can’t hide the underlying uncertainty of. Shashank Kumar, the founder‑CEO of Razorpay, cut through the hype at a recent industry round‑table, reminding everyone that the tech is still very much in its adolescence.

“We’ve seen impressive breakthroughs in labs,” he said, his tone a mix of admiration and caution, “but nobody— not Google, not OpenAI, not even the boutique AI start‑ups— has truly cracked the problem of scaling AI for everyday, high‑volume transactions.” The comment, while brief, sparked a ripple of nods across the room, as if a collective “I‑know‑what‑you‑mean” was being whispered.

Scaling, in the AI world, isn’t just about adding more servers or pouring more data into a model. It’s about weaving intelligence into the fabric of a product that processes thousands of payments per second, all while staying compliant with ever‑changing regulations. Razorpay, which handles billions of rupees every month, lives that reality daily.

“The moment you move from a prototype to a production environment, you discover a whole new set of challenges,” Kumar continued. “Latency becomes a killer, model drift can cost you money, and the compliance checklist balloons.” He cited a recent internal experiment where a generative‑AI tool, designed to auto‑classify merchant categories, produced a 7‑percent error rate once it was pushed to live traffic—a number that would be unacceptable in a high‑stakes payments ecosystem.

That 7‑percent figure, while seemingly small, translates to millions of mis‑routed transactions in Razorpay’s world. It’s the kind of tangible cost that theoretical research papers often gloss over. Kumar’s point was clear: the devil is in the details, and those details multiply when you’re talking about a platform that serves everything from fledgling startups to Fortune‑500 enterprises.

He also highlighted the regulatory maze that AI must navigate. In India, the Reserve Bank’s recent guidelines on digital payments stress transparency and auditability—principles that many black‑box AI models struggle to meet. “If you can’t explain why a model flagged a transaction, you’re dead in the water,” he remarked, half‑joking, half‑serious.

Yet, despite these hurdles, Kumar remains optimistic. He believes the journey is more about iteration than revolution. “We’re not looking for a one‑size‑fits‑all AI that solves everything overnight,” he said. “Instead, we’re building modular components, testing them in controlled environments, and learning from every failure.”

This pragmatic approach is reflected in Razorpay’s recent partnership with a leading AI research lab. The collaboration focuses on a narrow use‑case: fraud detection for high‑risk merchants. By limiting the scope, the team can fine‑tune latency, monitor drift, and, crucially, maintain a clear audit trail for regulators.

Industry observers have taken note. An analyst at a major investment bank, who asked to remain anonymous, said, “Razorpay’s candid admission is refreshing. It underscores a shift from hype‑driven promises to grounded, responsible AI deployment.” The analyst added that investors are beginning to value companies that acknowledge limitations and show a roadmap for responsible scaling.

So what does this mean for the broader AI ecosystem? According to Kumar, the answer lies in humility and collaboration. “If the biggest players can’t claim they’ve nailed scaling, then the rest of us need to work together, share learnings, and perhaps most importantly, set realistic expectations for our customers.”

In the end, the message was simple: AI is powerful, but it’s not a silver bullet. The real magic will happen when the technology is tempered with rigor, transparency, and a healthy dose of skepticism. As Razorpay continues to experiment and iterate, the fintech world will be watching— not just for the next breakthrough, but for the everyday, incremental wins that prove AI can truly function at scale.