AI at population scale: Rahul Chari’s charter for the next 300 million

Rahul Chari’s session at Global Fintech Fest (GFF) 2026 began with a practical question: what does artificial intelligence (AI) look like when it is built for a billion Indians with different languages, devices, data plans and levels of digital comfort? In a talk titled ‘AI at Population Scale: Engineering Intent-led Experiences for a Billion Indians’, PhonePe’s co-founder and Chief Product & Technology Officer argued that at this scale, the AI that matters is the kind that recedes into everyday tasks, whether it is paying a bill, checking where your money went, buying insurance or setting up a payment page, while changing how a financial platform understands and serves its users.

PhonePe’s approach to AI, Chari argued, is intensely practical and outcome-led, structured around four pillars: AI infrastructure, engineering productivity, operational productivity and end-client value. The goal is not to simply add chatbots to existing flows, but to move the platform from one that works around broad demographic cohorts to one that delivers a highly personalised “Bank Branch of One” for every user. That means moving from segment-level products to intent-level experiences, where the system understands what a user is trying to do and removes as much friction as possible between that intent and its completion.

From interfaces to intent

On the consumer side, that vision is taking shape through a set of AI-driven features designed to make the app feel less like a grid of icons and more like a conversational interface to everyday financial life. PhonePe is launching a natural-language search feature that allows users to complete tasks across payments, purchases and support through voice or text commands. Instead of searching through menus, a user can simply ask, in their own words, to send money to a contact, pay a specific bill, reorder a frequent purchase or raise a support query. The system will interpret the request, confirm details where needed and execute the task, bringing what would otherwise be a series of steps into a single interaction.


Equally important is what happens after a transaction. A new edge-cloud hybrid model allows users to query their spending patterns securely, with generative AI running partly on the device to strengthen privacy and reduce server costs. In practice, this means a user can ask questions such as ‘How much did I spend on groceries last month?’ or ‘Which merchants did I pay most often in August?’ without raw transaction data having to leave the phone unnecessarily. This way sensitive data stays closer to the user, while the processing power of larger models in the cloud can still be used for more complex analysis.

Insurance is another area where cognitive overload has long been a barrier. To reduce the friction involved in choosing a policy, PhonePe has built an agentic insurance flow where users can enter their details and allow an AI agent to find a suitable plan. The agent compares options across providers, explains key differences in plain language and surfaces recommendations based on the user’s profile and budget.

Making everyday finance conversational

If the consumer story is about intent, the merchant story is about bridging the digital divide. For millions of small businesses, the gap between having a smartphone and running a sophisticated digital operation remains wide. PhonePe’s AI tools are designed to narrow that gap by turning technical tasks into conversational ones. PG SmartPages, a zero-code solution, now allows merchants to deploy payment-ready webpages in under 10 minutes without design or coding expertise. The AI generates the copy, selects the necessary fields and optimises the page based on a simple description of the product. A kirana owner, home baker or tuition teacher can describe what they sell in their own words and end up with a mobile-optimised payment page that looks and works like something a small technology team might have built.

Putting digital tools within reach of small businesses

Behind that page lies another layer of automation that rarely gets showcased but matters enormously for adoption. Payment-gateway integration timelines for merchants have been cut from weeks to minutes through an AI Integration Agent. What once required days of reading documentation, debugging APIs and coordinating between developers can now be accomplished through a single AI interaction. The agent follows PhonePe PG’s official API specifications, generates the required payloads, handles errors according to best practices and produces a working integration that a merchant can take live almost immediately. For growth-stage businesses and small and medium enterprises (SMEs) that cannot afford large engineering teams, this turns payment infrastructure from a bottleneck into something that can be set up with far less technical effort.

On the enterprise side, PhonePe’s PulsePro data platform is getting its own conversational layer. “Ask Pulse” will allow users to pose complex, natural-language questions about India’s digital payments data rather than relying on predefined metric sorts. A business analyst can ask, ‘Which districts have the highest average transaction value?’ or ‘How did UPI volumes for grocery merchants change in Tier-II cities over the last quarter?’ and receive answers backed by anonymised transaction data spanning hundreds of millions of users and merchants. Earlier reporting on PulsePro has highlighted how the platform turns anonymised transaction data from over 71.5 crore users and 5 crore merchants across 99% of India’s postal codes into granular, hyperlocal market intelligence. Ask Pulse extends that capability by making it accessible to users who may not think in dashboards or data tables, but in questions.

The AI behind the scenes

None of this would be possible at PhonePe’s scale without a parallel transformation in how the company builds and runs its systems. Chari spent a significant part of his session on giving a walkthrough of internal operations and engineering scale, where AI is being used not as a buzzword but as a productivity multiplier. Central to this effort is KadhAI, a workflow orchestration platform that acts as a “cook” for code. KadhAI can plan and implement code changes in isolated sandboxes, run tests and checks, and then route the changes for human approval. Coupled with internal tools such as Lens, which gives teams deeper visibility into Android apps, and TraceMind, which automates much of the work involved in crash debugging, tasks that previously took days, from UI generation and threat analysis to root-cause diagnostics, can now be completed in minutes. According to the company, TraceMind is an automated debugging engine that correlates crash data, deployment histories and user journeys to produce root-cause summaries, while Lens provides deeper visibility into app behaviour across devices and versions.

Beyond individual tools, PhonePe has built an internal AI platform called AgentHub, where more than 200 AI agents are currently deployed. These agents handle over 1,600 queries daily for 759 weekly active users and automate up to 50% of tasks in some departments. Their roles range from support triage and fraud-investigation assistance to data reconciliation and compliance checks. The point, Chari emphasised, is not to replace humans but to free them from repetitive, low-judgement work so that they can focus on exceptions, edge cases and decisions that still require human oversight.

The impact of this operational AI is visible in some stark numbers. Despite processing about 10 billion transactions per month, PhonePe operates with around 500 customer support agents. By automating over 92% of tickets, the company maintains a ratio of roughly one support agent per 20 million transactions. According to the company, in fraud resolution, AI now handles about 80% of end-to-end investigations instantly, ​reducing the cost per investigation from approximately ₹50 to ₹5-₹15 while also clearing 800 previously unaddressed cases.

Designing for the next 200-300 million

Chari tied all of this back to a larger demographic challenge. India’s next wave of digital finance growth will not come from the already-connected urban middle class alone, but from the 200-300 million non-digitally native users who remain on the margins. Many of them are more comfortable with voice than text, with vernacular languages than English, and with guided flows than open-ended interfaces. PhonePe’s AI charter is explicitly designed to bring these users into the ecosystem by using voice, vernacular and on-device natural-language AI to reduce barriers such as digital intimidation and language constraints. The vision is to move users beyond basic peer-to-peer payments and help them discover more complex financial services, savings, credit, insurance and investments through domain-specific, agent-led experiences that feel less like navigating a bank and more like asking a trusted person for help.

Chari, however, emphasised that as AI takes on more of the heavy lifting, human oversight becomes more, not less, critical. The goal is not fully autonomous financial decision-making but augmented decision-making, where AI handles the routine and surfaces the exceptional for human review. In a sector where mistakes can mean lost money and eroded trust, the design principle is to keep humans firmly in the loop for high-stakes actions while allowing agents to manage the bulk of repetitive tasks.

The company is building the underlying systems of models, agents, orchestration layers and edge-cloud architectures that will determine whether AI at population scale remains a slogan or becomes a working system. If it works, the payoff is not simply a smarter app but a financial ecosystem capable of serving a billion intents a day, across dozens of languages and every kind of device, without collapsing under its own complexity.

The session ended where it began: with the question of scale. In most markets, AI at scale means millions of users. In India, it means hundreds of millions, soon to be a billion. Chari’s charter is an attempt to answer what it takes to make AI meaningful at that magnitude, by focusing on intent rather than interfaces, automating the routine to protect the exceptional, and treating the next 200-300 million users not as a future segment but as the core design cohort for everything that PhonePe builds next.