PhonePe is deploying AI across three fronts: enhancing engineering efficiency, accelerating AI-first features for users, and driving automation across the organisation. Rahul Chari, Co-Founder and CTO, PhonePe, speaks with FE FUTECH about the company’s phased, human-in-the-loop approach to ensuring reliability at scale, and why building the right AI foundations is critical before widespread deployment.
Edited Excerpts
In which areas at PhonePe has Artificial Intelligence moved from PoC to production?
At PhonePe, AI has moved beyond experimentation into three core production pillars: the software development life cycle (SDLC), internal operations, and consumer-facing products.
We are seeing a significant efficiency uplift—in the high 20% range—within our engineering teams. This is not limited to code assistance; AI is now embedded across code migrations, unit test coverage, and the QA lifecycle.
Beyond engineering, our internal ‘Agent Hub’, a self-serve agent marketplace, has become a central platform for knowledge sharing and configurable workflows to automate day-to-day tasks across the organisation.
On the consumer side, we have launched AI-powered natural language search, enabling users to perform complex tasks through simple voice or text commands such as “Recharge my FASTag” or “Send ₹50 to ABC,” while also gaining insights into their spending patterns within the app.
PhonePe’s biggest upside is monetisation beyond UPI (loans, insurance, wealth). How has AI helped here? Are you still in PoC or live?
Beyond UPI, we are leveraging AI to build intelligent, assisted experiences that support our monetisation efforts across lending, insurance, and wealth management.
Our approach focuses on simplifying complex financial journeys for users. Across these businesses, we are building AI-driven, do-it-yourself (DIY) experiences that allow users to initiate and manage their financial journeys seamlessly—from credit and insurance to investments.
These AI assistants are already demonstrating value in high-complexity areas. For instance, in the equity markets vertical, a digital voice assistant proactively helps users navigate complex KYC requirements.
Similarly, contextual AI chatbots are deployed across mutual funds, insurance, and credit cards, enabling users to ask detailed questions about coverage, benefits, and products in real time.
You believe companies should build AI foundations before scaling. What baseline are you setting at PhonePe?
Our strategy is guided by a simple principle: foundations must come before scale. Internally, we refer to this as building the right organisational scaffolding.
Before rolling out any AI tool at scale, we ensure that the underlying engineering—data access, authorisation, auditability, and observability—is robust and aligned with the highest standards of safety and compliance.
This baseline enables us to re-architect our SDLC and empowers engineering teams to build more resilient systems with significantly higher efficiency. By securing this foundation first, we are able to accelerate the rollout of AI-powered features with greater confidence.
We prioritise outsized impact over incremental gains that may not justify their cost. At PhonePe, we are not focused on the fastest path—we focus on the right path—ensuring that our architecture can support high-velocity transactions without bottlenecks, while maintaining the highest standards of data security and privacy for our customers and merchants.
In fraud detection and control, how are you using AI to reduce losses and enhance trust? Any numbers you can share?
The most immediate measurable outcome of our AI deployment is in the automated resolution of fraud complaints. Investigations into fraudulent activity, as reported by law enforcement agencies (LEAs), are inherently complex. Today, however, 80% of these investigations are completed end-to-end using AI.
By automating the majority of fraud investigations, we have significantly reduced resolution times. This has fundamentally transformed our risk management operations, allowing specialised teams to focus on identifying emerging and sophisticated fraud patterns (modus operandi), rather than handling manual case processing.
We are also seeing a measurable improvement in customer experience (CX), as transaction insights and data are leveraged to resolve user queries faster and more accurately.
