In FY26, SBI used AI-led underwriting to process nearly Rs 1 trillion of loans of up to Rs 5 crore each for new-to-bank and existing MSME customers, managing director Rama Mohan Rao Amara said at FIBAC 2026 on Wednesday.

The bank combines GST data, bureau scores, account information and other unstructured data to assess borrowers. “Almost Rs 1 trillion we were able to underwrite,” Amara said, adding that the technology has freed relationship managers from data gathering and analysis.

He also highlighted the model had the capability to reduce the delinquency, he said. The bank is also using AI to identify vulnerable exposures before delinquencies emerge by ingesting sector-specific, market and other publicly available information to generate early warning signals.

In unsecured lending, AI is helping SBI assess thin-file customers, including small businesses and proprietorships, using account and UPI-related data. “It is helping us expand the financial inclusion and also in terms of meeting the priority sector obligations,” Amara said.

SBI has moved some generative AI pilots into full-scale use cases. The bank has automated processing of cheques up to Rs 10,000. Such cheques account for around 25% of its volumes.

The bank is also piloting an agentic AI assistant for corporate banking, which can analyse financial statements and documents and assist with risk assessment. This would shift relationship managers towards validating model outputs and focusing on customer relationships. “

While I agree, quantification by way of a defined cost-to-income reduction will take some time, but there are intangible benefits in terms of customer satisfaction, release of the bandwidth of the manpower whom we are able to deploy otherwise,” Amara said.