SBI uses AI to underwrite nearly ₹1 lakh crore in MSME loans in FY26
2 min readDuring FY 2025‑26, the State Bank of India (SBI), the nation’s biggest bank, leveraged artificial intelligence to approve MSME loans amounting to close to ₹1 trillion, each loan capped at ₹5 crore, according to a senior executive speaking on Wednesday.
‘In FY26, within a twelve‑month window, we have processed loans up to ₹5 crore for both new‑to‑bank and existing clients,’ said SBI Managing Director Rama Mohan Rao Amara at the FIBAC annual gathering. ‘Our AI‑driven underwriting enabled us to sanction roughly ₹1 lakh crore.’
He further noted that the bank is employing AI, including large language models (LLMs), to streamline the handling of cheques valued up to ₹10,000.
By using AI, the workload of relationship managers has been reduced, as they no longer need to devote extensive time to data gathering and initial analysis, the official explained.
Cheques up to ₹10,000 represent roughly 25 % of the bank’s total cheque volume; the AI system can scan these instruments, check required fields, and ensure compliance, he added.
Such cheques are now processed via a straight‑through‑processing (STP) pipeline that operates with minimal human involvement.
Nevertheless, SBI maintains a safeguard: a specialist risk‑control team periodically audits a sample of AI‑processed cheques to spot mistakes and decide if the models need additional training.
He emphasized that AI is being applied throughout the customer journey — covering credit underwriting, portfolio oversight, fraud detection, and service delivery.
The bank additionally employs AI‑driven early‑warning alerts to flag at‑risk exposures before they turn delinquent, by analysing sector‑level data, market trends, and other publicly sourced information.
Regarding AI’s advantages, the official noted that measuring a precise impact on the cost‑to‑income ratio will require time, yet the bank is already observing improvements in customer experience and freeing up staff capacity.
Published on August 12, 2026