Flipkart unifies technology under 'OneTech' push as IPO window nears

At the centre of the overhaul is a shift to artificial intelligence-first architecture, replacing legacy systems with large language models and agentic frameworks while keeping the business running at full scale. Balaji Thiagarajan, Flipkart’s chief product and technology officer, likened the challenge to "changing the engines of a flying plane”.

A former Google executive with stints at Microsoft, Uber and Yahoo, Thiagarajan was appointed to the role in September. He took over from Jeyandran Venugopal, who stepped down in February 2025.

“There is a flying plane that has to keep flying. It cannot land because the business is running. And our job is to upgrade that plane in flight, changing the engines,” Thiagarajan told Business Standard.

The transformation spans customer experience, supply chain operations and IPO readiness, with Flipkart continuously tightening governance around data security, compliance and fraud detection to meet the heightened scrutiny of public markets. The company is strengthening controls around Digital Personal Data Protection standards, PCI compliance and responsible AI, while tightening audit processes and business continuity systems.

Flipkart is taking a measured approach to the cost of transforming while operating at scale, shifting all new development to an AI-first model even as it runs its core business. The company is layering AI-driven features, including a conversational shopping assistant, onto existing products while building new capabilities alongside current systems. Over time, these AI "sidecar" services will replace legacy infrastructure, with transformation and operations running in parallel.

Flipkart has increased its investment in AI sixfold this year. The focus is on measurable outcomes rather than experimentation, particularly in return reduction, fraud detection, supply chain efficiency and customer support.

Thiagarajan said the company began its AI deployment with customer support, rolling out bots and co-pilots to improve post-order experience and help agents resolve queries. Similar tools have since been extended to sellers through AI-driven support and "seller intelligence" to aid pricing, promotions and demand insights. Flipkart has also invested in demand forecasting, assortment optimisation, search and recommendations, while scaling AI infrastructure, including GPU (graphics processing unit) capacity and inferencing systems.

“We have invested in AI very deeply, and we had those investments started very early on in the AI cycle,” said Thiagarajan. “In 2026, we also built upon an AI infrastructure that we started building in 2025 to have our own GPU farms.”

The challenge of deploying AI at Flipkart's scale is compounded by India's linguistic and connectivity landscape. The company serves over 500 million registered users, many of them first-time internet buyers in Tier 2 and 3 cities. Flipkart supports 12 vernacular languages and is expanding into voice interfaces, which introduce multiple interaction combinations across languages and formats. Variability in connectivity, device capability and speech patterns requires solutions such as noise handling, slang recognition and on-device models to ensure seamless user journeys even in unstable network conditions.

Trust has emerged as a parallel infrastructure challenge. Rather than being underinvested, Thiagarajan said trust infrastructure has evolved alongside the rapid scale of e-commerce adoption, with new use cases continuously exposing fresh challenges. Flipkart is building trust across the customer journey through specific products: Open Box Delivery, geocoding for unmapped lanes and AR try-ons for high-value purchases. These sit alongside biometric authentication, smoother post-order experiences, faster delivery and installation services through its Jeeves unit, exchange programmes and protection offerings.

“Trust and security are things that have been a very high-visibility, high-priority exercise for us," he said.

The use of AI at Flipkart is tightly controlled, with a clear distinction between autonomous agents and co-pilot systems based on risk and variability. Autonomy is deployed in predictable areas such as logistics optimisation, while high-stakes functions like accounting and fraud detection continue to require human oversight.

“In some areas where the technology itself is not great enough yet, humans have to play a role," Thiagarajan said.

Flipkart's resilience strategy operates on two tracks: maintaining customer service levels and managing internal cost pressures. Disruptions such as fuel shortages linked to geopolitical events can slow supply chains, but the company responds through a multi-pronged approach across first mile, long haul and last mile operations, reconfiguring movement at scale and shifting to higher-volume transport where per-unit costs are lower. If costs remain elevated, the company may absorb them while using technology to increase basket sizes and average selling prices to offset pressure.

“When Flipkart was going through the Covid times, orders dropped to a bare minimum… we decided to absorb the cost… we did not lay off our employees… we survived it, and then we hit a huge growth spot," he said.

Thiagarajan's experience building Kasu.ai—an AI platform focused on streamlining business operations—sharpened his view on one unresolved constraint. While traditional technology allows companies to build products and scale into businesses, AI requires significant infrastructure investment whose costs do not yet align with what markets are willing to pay. At Flipkart, he said this translates into a broader question the industry has yet to answer.

"AI at scale… what it is going to do in terms of AI infrastructure costs is a question that has not been answered yet."

On talent, Thiagarajan said Flipkart competes for AI engineers not through compensation alone but through the nature of the work. Unlike multinational corporations where teams are part of globally distributed systems, Flipkart operates end-to-end from India, giving engineers greater portability across functions and domains without needing to relocate.