State Bank of India (SBI) used artificial intelligence (AI) and digital data to underwrite nearly ₹1 trillion of loans of up to ₹5 crore each to medium, small and micro enterprises (MSMEs) in fiscal year 2026 (FY26), reducing the workload of relationship managers. The bank has also seen lower delinquency in the portfolio underwritten through its business rule engine (BRE), managing director Rama Mohan Rao Amara said.
The scale of AI-assisted underwriting marks a shift in how the country’s largest lender is processing smaller business loans, with technology taking on much of the information-gathering and preliminary analysis that relationship managers previously handled.
“We were able to, using AI, underwrite a large number of loans,” Amara said at the #FIBAC 2026 event in Mumbai on Wednesday. The bank underwrote loans of up to ₹5 crore each through FY26, covering both new-to-bank and existing customers, he said.
SBI’s underwriting process draws on the digital footprint available for MSMEs, including GST Network and filing data, bank-account information, credit-bureau scores and other structured and unstructured information. The bank combines these datasets through its business rule engine (BRE) to assess borrowers.
That has shifted work away from relationship managers, who previously spent significant time collecting information and conducting preliminary analysis. “It has actually freed the bandwidth of the people who were otherwise the relationship managers, particularly, who were otherwise spending a lot of time in terms of gathering the data, doing some analysis, et cetera,” Amara said.
SBI has also seen lower delinquency in the portfolio underwritten through the BRE, he said.
SBI is extending AI into other parts of lending and portfolio management. In unsecured lending, it uses AI to assess thin-file customers such as small businesses and proprietorships, where limited information may be available beyond account and UPI data. This is helping the bank expand financial inclusion while meeting priority-sector lending requirements.
For existing borrowers, SBI uses AI to identify vulnerable exposures before conventional warning signs such as days past due (DPD) emerge. The models can ingest market- and sector-specific information and other publicly available data to generate early-warning signals, allowing the bank to intervene proactively.
The technology is also being applied to routine banking operations. SBI is using AI, including Large Language Models (LLMs), to automate processing of cheques up to ₹10,000. Such cheques account for about 25% of the bank’s cheque volumes. The AI model checks mandatory fields and compliance requirements, allowing these transactions to be processed through straight-through processing with minimal human intervention.
Human oversight remains in place. SBI’s control risk unit regularly samples cheques processed by AI to identify errors and determine whether the models need further training.
AI is also being used in fraud and technology risk management. SBI’s Security Operations Centre uses AI to process large volumes of IT infrastructure logs, while its Resiliency Operations Centre uses the technology to identify and predict potential breakdowns.
The bank is now moving some generative AI experiments into full-fledged applications. Its customer-care operations, for instance, have evolved from using AI to assist human agents to deploying AI bots that can handle customer calls end-to-end, with only complex cases transferred to employees. This has helped reduce handling time and improve customer satisfaction, Amara said.
In corporate banking, SBI is piloting an agentic AI-based digital assistant called YonoG, embedded in Yono Business. The tool can collect and read financial and other unstructured documents, populate loan lifecycle management systems and conduct risk analysis.
The shift is also changing the role of relationship managers, Amara said. “Now the role of the relationship manager is more in terms of validating that, looking at even the stress testing, whatever is coming out of that model,” he said.
SBI has yet to quantify the overall impact of AI through a defined reduction in its cost-to-income ratio. Amara said the benefits were already visible through improved customer satisfaction, better risk management and the release of employee bandwidth for higher-value work.
For the quarter ended June (Q1FY27), SBI’s gross advances rose nearly 19% on year to ₹50 trillion. Domestic corporate advances grew over 18% on year to ₹14 trillion, while its SME book rose over 22% on year to ₹6.46 trillion.
Subhana Shaikh is a business journalist at Mint, where she covers the Reserve Bank of India, monetary policy, and India’s bond markets. She has seven years of experience in reporting on financial markets, with a focus on banking and the broader financial system.<br><br>She began her career after completing her postgraduate diploma at the Indian Institute of Journalism and New Media, Bengaluru. She then spent five years at Informist Media, a news wire agency, where she closely tracked bond markets and the BFSI sector, developing a strong foundation in market reporting. She later moved to NDTV Profit, where she expanded her coverage across a wide range of business and economic stories.<br><br>At Mint, Subhana focuses on explaining central bank decisions, bond market movements, and banking trends for her readers. Her reporting combines on-ground inputs with careful analysis to help audiences understand complex financial developments.<br><br>Based in Mumbai, she is interested in exploring stories across the business landscape. Outside of work, she enjoys reading and spending time with her three cats.
Catch all the Business News, Market News, Breaking News Events and Latest News Updates on Live Mint. Download The Mint News App to get Daily Market Updates.
Oops! Looks like you have exceeded the limit to bookmark the image. Remove some to bookmark this image.
Facts Only
* SBI used AI and digital data to underwrite nearly ₹1 trillion in loans up to ₹5 crore each for MSMEs in FY26.
* The process reduced the workload of relationship managers.
* Lower delinquency was seen in the portfolio underwritten through the business rule engine (BRE).
* Underwriting draws on GST data, bank-account information, credit bureau scores, and other structured/unstructured data.
* The business rule engine (BRE) combines these datasets to assess borrowers.
* AI is used in unsecured lending to assess thin-file customers using account and UPI data.
* AI identifies vulnerable exposures for existing borrowers before conventional warning signs appear.
* LLMs are used to automate processing of cheques up to ₹10,000 with minimal human intervention.
* AI is used in the Security Operations Centre to process IT infrastructure logs and in the Resiliency Operations Centre to predict breakdowns.
* Generative AI bots handle end-to-end customer calls in customer care operations.
* An agentic AI assistant named YonoG is piloted in corporate banking for document reading, system population, and risk analysis.
* The role of relationship managers shifted to validating model outputs like stress testing.
* SBI's gross advances rose nearly 19% year-to-year to ₹50 trillion for Q1FY27.
Executive Summary
State Bank of India utilized artificial intelligence and digital data to underwrite approximately ₹1 trillion in loans, with individual loan amounts up to ₹5 crore each, for medium, small, and micro enterprises during fiscal year 2026 (FY26). This process reduced the workload on relationship managers. The bank also observed lower delinquency rates in the portfolio managed through its business rule engine (BRE). The underwriting process incorporates data from sources like GST networks, bank accounts, credit bureau scores, and other structured and unstructured information, which are combined via the BRE for borrower assessment.
The application of AI shifted information gathering and preliminary analysis away from relationship managers. This freed up their capacity to focus on validation, such as stress testing the results generated by the models. Furthermore, SBI applies AI in unsecured lending to assess thin-file customers with limited data beyond basic account and UPI information, thereby expanding financial inclusion. For existing borrowers, AI identifies vulnerable exposures before conventional indicators like days past due emerge by analyzing market and sector-specific information for early warnings. The technology is also integrated into routine operations, automating cheque processing using Large Language Models (LLMs), and managing fraud/risk through AI-driven analysis of IT logs.
Full Take
The narrative centers on the redistribution of cognitive labor through technological augmentation, moving complex data processing to AI systems and redefining human roles in finance. The initial promise—reducing managerial workload and lowering delinquency via automated risk assessment—is supported by measurable outcomes like increased bandwidth and proactive exposure identification. However, the structure implicitly frames this shift as purely efficiency-driven; the implication is that technology provides a net positive gain without significant rebalancing of power or cost allocation.
The pattern emerges in the evolution from data collection (done by managers) to data processing (done by AI), and finally to validation (the new role for managers). This suggests an emerging dependency where human oversight becomes specialized in critique rather than execution, raising questions about skill stratification and the potential for automation to solidify existing risk models. The use of AI to expand financial inclusion, while meeting regulatory requirements, necessitates scrutiny regarding whether this expansion is truly equitable or merely optimized for priority-sector lending targets.
The deployment across different functions—from credit risk to routine operations and customer service (LLMs/bots)—demonstrates a systemic approach to digital transformation rather than isolated project implementation. The focus on mitigating risk through early warning signals, juxtaposed with the development of agentic tools like YonoG for complex tasks, indicates a strategic attempt to weave AI into the core decision-making fabric. The central unstated implication is the ownership and governance of these sophisticated models; the success depends entirely on ensuring that the human validation role remains strategically vital, preventing the system from becoming an opaque, self-optimizing black box where accountability diffuses into algorithmic complexity.
Bridge Questions: If relationship managers are shifting to model validation, what specific new competencies must they acquire to maintain strategic relevance in this augmented environment? How is the governance structure ensuring that the data used for risk assessment remains free from embedded historical biases when dealing with thin-file or unsecured customers? What mechanisms are in place to ensure the benefits of efficiency gains translate into improved customer outcomes rather than simply enhanced operational speed?
Sentinel — Human
The text reads like structured financial journalism, grounded by specific stakeholder quotes and organizational processes, suggesting a human editorial hand guided by verified data.