State Bank of India's managing director Rama Mohan Rao Amara disclosed this week that the bank used artificial intelligence to underwrite nearly ₹1 trillion in loans to micro, small and medium enterprises during FY26 — loans of up to ₹5 crore, offered to new-to-bank and existing customers alike, assessed through what SBI calls its Business Rule Engine.

What the system actually does

The engine, embedded in the bank's YONO Business app, pulls GST Network filings, account information, credit bureau scores and other unstructured documents, and turns them into a lending decision without a relationship manager manually chasing paperwork. SBI says the same infrastructure now runs early-warning checks on existing borrowers too, scanning sector-specific and market data to flag vulnerable accounts before they slip into default — and that loans underwritten this way have shown lower non-performing assets than the bank's traditional MSME book, though SBI has not published a side-by-side NPA figure for the two portfolios.

The practical shift is what gets automated away: an MSME owner who previously needed to assemble bank statements, GST returns and collateral documentation for a relationship manager to review can now get a decision pulled directly from data the government and the bureaus already hold.

Why ₹1 trillion is a start, not an answer

Set against India's own numbers, the scale is more modest than the headline suggests. A parliamentary standing committee on finance has put the country's MSME credit shortfall at roughly ₹20–25 lakh crore, with close to half of MSME credit demand going unmet. Credit penetration among Indian MSMEs sits at around 14%, against roughly 37% in China and 50% in the United States — and only about one in five micro-enterprises accesses formal credit at all. SBI's ₹1 trillion, real as it is, covers well under 5% of the low end of that gap.

That context matters for what AI underwriting is and isn't solving. The system doesn't reach businesses with no GST registration, no digital financial footprint or informal bookkeeping — which describes a large share of India's micro-enterprises, the ones furthest from formal credit. What it does is speed up and cheapen the decision for businesses that already generate the digital data trail: registered, GST-filing, bureau-scored companies that were bankable in principle but expensive to underwrite manually at small ticket sizes.

The distinction that matters going forward:

  • Faster credit for the already-formal — GST-registered, bureau-scored businesses get quicker, cheaper decisions.
  • No change for the still-informal — enterprises without a digital data trail remain outside what AI underwriting can currently assess.

SBI's number is a genuine milestone in how India's largest lender assesses small-business risk. Whether it dents the ₹20 lakh crore gap depends on whether AI underwriting eventually extends to businesses that don't yet look bankable on paper — not just processes the ones that already do, faster.

Could not independently verify: SBI has not published the specific NPA percentages for its AI-underwritten MSME book versus its conventional book; the "lower NPA" claim comes from the bank's own disclosure.