SBI uses AI to underwrite nearly Rs 1 lakh cr in MSME loans in FY26
The country's largest lender, State Bank of India (SBI), used artificial intelligence to underwrite nearly Rs 1 lakh crore in MSME loans of up to Rs 5 crore each during 2025-26, a senior bank official said on Wednesday, August 12.
"In FY26, in 12 months time, we are able to underwrite loans up to Rs 5 crore, both covering the new-to-bank customers, as well as existing customers. Almost Rs 1 lakh crore we were able to underwrite," SBI Managing Director Rama Mohan Rao Amara said at the annual FIBAC event here.
He added that the bank is also deploying artificial intelligence, including Large Language Models (LLMs), to automate processing of cheques up to Rs 10,000.
The use of AI has helped free up the bandwidth of relationship managers, who earlier spent considerable time collecting data and carrying out preliminary analysis, the official said.
Cheques of up to Rs 10,000 account for around 25 per cent of the bank's cheque volumes, and the AI model can read the cheque and verify mandatory fields and compliance requirements, the official said.
The bank has now automated processing of such cheques through a STP (straight-through processing) system, with practically no human intervention.
However, SBI has retained a control mechanism under which a dedicated control risk unit reviews a sample of cheques processed by the AI system to identify errors and determine whether the models require further training.
He said AI is being deployed across the customer lifecycle, including credit underwriting, portfolio management, fraud risk management and customer service.
SBI is also using AI-based early warning signals to identify vulnerable exposures before delinquencies emerge, by analysing sector-specific, market and other publicly available information.
On the benefits of AI, the official said quantifying the impact in terms of a defined reduction in the cost-to-income ratio would take time, but the bank was already seeing benefits in terms of customer satisfaction and release of employee bandwidth.
