HDFC Bank built its Neev AI platform with an initial ₹2 crore

32

TL;DR:

HDFC Bank seeded its in-house Neev AI platform with ₹2 crore, favouring specialised banking models over expensive general-purpose systems. The strategy matters because it combines lower initial costs with tighter control over data, model behaviour and regulatory compliance.

Article:

HDFC Bank says it built Neev, its in-house enterprise artificial intelligence platform, with an initial investment of ₹2 crore, offering a striking counterpoint to the large budgets often associated with corporate AI programmes.

The platform is maintained by a team of about 40 people and supports reusable AI capabilities across banking operations. Ramesh Lakshminarayanan, the bank’s group head of information technology and chief information officer, said: “We spent a few more crores after we built the platform initially.”

Neev’s significance lies less in its modest starting cost than in its design. Instead of depending entirely on large, general-purpose language models, the bank is developing smaller models for tasks requiring banking knowledge, speed and high accuracy. These include document extraction, trade-document classification, customer-query support, card processing and credit decisioning.

“Banking applications require high accuracy and reliability,” Lakshminarayanan said, arguing that internal capabilities give the lender greater visibility into model behaviour and compliance before deployment.

HDFC Bank’s FY2025-26 annual report describes Neev as a secure foundation for model access, governance and workflow integration. It also says 98% of the bank’s financial transactions are digital, raising the operational stakes for reliable AI infrastructure.

The approach could become a template for other regulated enterprises: buy external computing and model access where useful, but retain control over domain logic, testing and sensitive data. However, ₹2 crore represents the initial platform investment, not its full lifetime cost.

Neev’s commercial value will ultimately depend on measurable outcomes, including faster processing, fewer errors, stronger fraud controls and lower operating costs. A cheap prototype is noteworthy; dependable AI at banking scale is the harder result.