PhonePe CTO Srijon Biswas outlined the efficiency gains from its use of AI, with more than 200 agents used by around 1,600 employees daily.

Reducing the time and resources required for routine workflows can help fintechs manage costs while freeing employees for other functions.

Biswas said PhonePe has saved about 22 person-days per app release and shortened specific audit processes from four days to three minutes.

AI is helping PhonePe reduce operating costs and speed up internal workflows, with specific audit processes shortened from four days to three minutes and about 22 person-days saved per app release, according to its CTO Srijon Biswas.

Speaking to Inc42 CEO and cofounder Vaibhav Vardhan at the inaugural ‘The CTO Summit 2026’, Biswas detailed the gains across operations and software development, alongside the fintech’s more cautious approach to consumer-facing AI.

During the session, “How PhonePe Is Combining DPI And AI For 720 Mn+ Users”, Biswas said the company now has more than 200 AI agents, used by around 1,600 employees daily.

In operations, PhonePe has deployed AI in specific workflows to reduce costs and manual work. Biswas said voice agent costs have fallen to about one-sixth of their previous level.

In certain KYC and onboarding workflows, the company has also removed the need for human agents. “We have been able to reduce 20 agents to zero,” he said, clarifying that the employees were moved to other functions.

Some audit processes have seen substantial time savings, with Biswas saying the time required has fallen from four days to three minutes.

Speeding Up Software Development

The savings extend to software development, where PhonePe has reduced the work required for each app release by about 22 person-days. Biswas said the frequency of releases makes the cumulative saving significant.

In one development workflow, a change that passed from a product manager to a designer and then a developer previously took two to three days. Automated code generation has reduced this to around 10 minutes, he said.

Unit testing on the app has similarly fallen from three days to about half a day.

Biswas said the company broadly evaluates its AI efforts across three areas: operational efficiency, developer efficiency, and consumer-facing applications.

While operational efficiency focuses on reducing costs, developer efficiency is about becoming faster “without losing my quality bar”, he said.

A More Cautious Consumer AI Approach

PhonePe is taking a more cautious approach to consumer-facing AI, where applications require greater engineering effort and careful consideration of the user experience.

One example Biswas cited allows users to ask questions about their transaction history in plain English, instead of navigating multiple filters.

For a regulated fintech, AI agents also need to operate within existing access controls. Biswas said PhonePe took time to build a layer that ensures models and automated systems follow the same role-based access rules.

“We decided to go slow a little bit and made sure that we have a layer which even when a model or any automation wants to access these systems, they have to go through the same roles,” he said.

The emphasis on access controls also featured in a separate discussion at the CTO Summit, where PhonePe and OceanBase executives highlighted fresh data and clear access rules as foundations for reliable AI applications.

Looking ahead, Biswas expects AI to change the composition of engineering teams, with greater emphasis on senior judgement and decision-making.

“You need more people who can make stronger decisions and get them correct more often,” he said.