Flows

A 2026 Boston Consulting Group (BCG) report with Indian banking bodies found that 86% of Indian banks had generative artificial intelligence (AI) use cases under implementation. A September 2026 survey of 31 US mortgage firms found an average of about 10 of 38 listed AI use cases in production. The figures show strong activity, but they do not tell a lender which use cases will justify a 2027 budget.

Quick Answer

Generative AI in lending is in production mostly for language-heavy work: extracting data from documents, summarizing files, and assisting staff and call centers. Credit decisioning is much less common, and regulators in both India and the US are still setting the rules for it. Lenders should start with document and servicing workflows and ask for measured outcomes before funding anything beyond them.

Key Takeaways

• Production use is concentrated in language heavy work such as writing, summarisation, document extraction and staff knowledge search.
• Evidence is thinner for underwriting decisions, fraud detection and credit risk analytics than it is for productivity use cases.
• Reported operational results often combine generative AI with speech recognition and conventional machine learning, so attribution needs care.
• Lenders should fund use cases with a defined owner, measurable outcome, human review path and post deployment monitoring.

How Far Has Generative AI Spread in Lending

Adoption looks high in India and uneven in the US. The two sources measure different things, so the figures are not directly comparable.

In India, BCG, FICCI, and the Indian Banks’ Association (IBA) published Winning in the AI Era in August 2026. Per press coverage, 86% of banks have generative AI use cases under implementation in 2026. The figure was 44% in 2025 and 10% in 2024.

“Under implementation” is not the same as in production. The report names scaling as the main challenge, citing data readiness, talent shortages, regulatory concerns, and uncertain returns.

In the US, the American Association of Residential Mortgage Regulators (AARMR), the Mortgage Bankers Association (MBA), and BCG ran a joint survey. It covered 31 mortgage lenders and servicers from April to July 2026. Most were independent mortgage banks, and together they represent about 40% of US mortgage volume.

Nearly all had at least one AI use case in production. On average, respondents had about 10 of 38 listed use cases in production.

Coverage differs on how many had fully scaled a use case, so this article does not use that figure.

The survey covers AI broadly. Its most common use cases, such as writing and summarization tools, rely on generative models.

Where Generative AI Is Actually in Production

The US mortgage survey gives the clearest view of what is live. Language-heavy tasks lead:

• Employee writing and summarization tools: in production at 87% of respondents.
• Code generation and developer support: 65%.
• Data extraction from documents: 61% in production, 42% at scaled production.
• Agent assistance and knowledge search: 54%.
• Investor guideline and eligibility extraction: 52% in production, 42% scaled.
• Document classification and summarization: 52% in production, 35% scaled.

Respondents reported much lower adoption of underwriting decision support, fraud detection, and credit risk analytics. The survey did not give a single figure for these.

It also flagged servicing as an area with room to grow. Early default warnings, collections prioritization, and loss mitigation decision support were the examples given.

What Indian Lenders Are Reporting

Indian non-bank lenders disclosed some of the most specific numbers. Outlook Business reviewed the December 2025 quarter earnings calls in February 2026.

Bajaj Finance’s managing director, Rajeev Jain, said AI listened to 20 million customer calls and converted them to text. He said this produced 100,000 new customer offers the company had no information for earlier.

He also said loan disbursals through AI-powered call centers were about Rs 1,600 crore. That is roughly 10% of the Rs 16,545 crore disbursed in the quarter. The company reported 46 million face matches and document extraction at 95% to 96% accuracy across 43 document types.

Tata Capital said it had moved from AI pilots to enterprise-wide deployment across marketing, sales, credit, operations, service, and collections. It described itself as an early investor in generative AI driven underwriting. The coverage reported no outcomes for that work.

These disclosures do not separate generative AI from speech recognition and classical machine learning. They show AI in operations, not generative AI alone.

Where the Hype Runs Ahead of the Evidence

Announcements are growing faster than evidence of business results. Three findings illustrate the gap:

• Evident AI Index, Q2 2026: the 50 global banks it tracks announced a record 93 new AI use cases. Generative AI use cases almost doubled. Only 27% of new use cases disclosed an outcome, down from 41% the quarter before. About 1 in 10 of those reported a revenue uplift.
• BCG, FICCI, and IBA projections: AI agents across the credit journey could cut turnaround by 50% to 90%. The report also projects operating costs 40% to 60% lower and credit mortality 20% to 30% lower. These are projections, not measured results.
• AARMR, MBA, and BCG survey: use was concentrated in employee productivity and document-heavy work. The survey found less evidence of measurable gains in outcomes such as origination cost or customer experience. Unclear return on investment was cited by 45% of respondents.

Evident’s data covers global banks, not lending alone. Even so, it shows how few announcements come with a result attached.

What Regulators in India and the US Expect

India and the United States are taking different regulatory paths. India proposes to bring AI within its broader model risk framework, while the revised US interagency guidance excludes generative and agentic AI pending further work.

Topic	India	US
Main instrument	Reserve Bank of India (RBI) draft Model Risk Management guidance, released June 24, 2026	Interagency model risk guidance, issued April 17, 2026, replacing SR 11-7
Generative AI	Covers all models, including AI, machine learning, and third-party models	Excludes generative and agentic AI, calling them new and fast-changing
Governance	Board-approved framework. Board risk committee approval for high-risk models	Banks apply broader risk management practices. Agencies plan a request for information on AI
Vendor models	Institution stays accountable. Needs vendor documentation and audit rights	No model-specific rule for generative AI yet
Customer decisions	RBI Governor Sanjay Malhotra called for explainability and human oversight	ECOA and Regulation B still require specific reasons for adverse action
Status	Draft. Comments closed July 24, 2026	In effect. AI-specific guidance pending

The US Consumer Financial Protection Bureau (CFPB) withdrew its 2022 and 2023 circulars on adverse action and complex algorithms in May 2025. The circulars were guidance. The duty itself sits in the Equal Credit Opportunity Act (ECOA) and Regulation B.

For planning, an Indian lender should expect a model inventory and board-level approval for generative AI, if the draft is finalized as proposed. A US bank must rely on its own governance until the agencies act.

Controls That Weaken After Go-Live

The AARMR, MBA, and BCG survey found basic controls in place at most firms. Written AI policies were reported by 87% of respondents. Privacy controls were at 84%, human review at 81%, and vendor controls at 74%.

Controls thinned out after deployment. Only 58% reported ongoing monitoring for model drift and accuracy. Only 45% reported regular AI reporting to their boards.

More than one-quarter said employees use AI outside approved tools. Twenty percent said this happens occasionally and 7% said frequently.

Generative AI by Lending Function

Function	What the evidence shows	Read
Document extraction and classification	In production at 61% of US mortgage respondents. Scaled at 42%	Live
Staff assistants and knowledge search	87% for writing and summarization tools. 54% for agent assistance	Live, mainly staff productivity
Customer calls and voice	Bajaj Finance: about 10% of quarterly disbursals through AI call centers	Live at one large lender. Generative share unclear
Collections and early default warnings	Named as room to grow in US servicing	Early
Underwriting decision support	Significantly lower adoption in US survey. Tata Capital cites investment, no outcomes	Early
Fraud detection and credit risk analytics	Significantly lower adoption in US survey	Early
Agent-led credit journey	50% to 90% faster turnaround is a BCG projection	Projection

Questions to Ask Before Funding a Generative AI Use Case

• Is the task language-heavy, and can a person review the output before it affects a customer?
• What measured outcomes exist for this exact use case, such as turnaround, cost per file, and error rate?
• Which framework covers it? In India, expect the RBI draft. In the US, your own governance applies.
• Can adverse action reasons be produced without relying on generative output?
• Who monitors accuracy and drift after go-live, and how often does the board see results?
• Do vendor contracts give audit rights and documentation access?
• Is there an approved-tools policy that limits staff use of unapproved AI?
• What is the plan if the model changes or the vendor exits?

Bottom Line

Generative AI is already used in lending for document work, staff assistance and customer interactions. Adoption is much thinner in underwriting and credit risk, where many of the most widely cited gains remain projections rather than measured results.

Both countries are tightening oversight, in different ways. Lenders that tie each use case to a measured outcome and a named owner will be better placed than those buying on announcements.

FAQs

Is generative AI being used to approve loans?

Evidence points to limited use. In the US mortgage survey, underwriting decision support had significantly lower adoption than document tasks. Some Indian lenders say they are investing in it, but the coverage reviewed reported no outcomes.

Does the RBI regulate generative AI in lending?

Not yet through a final rule. The RBI’s June 2026 draft Model Risk Management guidance covers all models, including AI, machine learning, and third-party models. It follows the RBI’s FREE-AI committee report of August 2025, which made 26 recommendations.

How do US rules treat generative AI in lending?

The April 2026 interagency model risk guidance excludes generative and agentic AI, and the agencies plan a request for information. ECOA and Regulation B still require specific reasons for adverse action on credit decisions.