At Cypher 2024, Debarag Banerjee, Chief AI & Data Officer at L&T Finance presented a groundbreaking approach to reimagining credit assessment in India. The session explored how artificial intelligence and alternative data sources are transforming lending practices, particularly for underserved populations in the gig economy and rural areas.
Core Concepts of AI-Driven Credit Assessment
L&T Finance has developed a sophisticated AI-powered credit assessment system called Cyclops, which fundamentally challenges traditional lending approaches. The core innovation lies in expanding beyond conventional credit signals to create a more comprehensive view of potential borrowers’ creditworthiness.
Traditional credit assessment typically relies on:
- Credit history
- Formal employment records
- Standard documentation
In contrast, L&T Finance’s approach incorporates:
- UPI transaction data
- Bank statement analysis
- Geographical location insights
- Alternative data sources like payment aggregator information
Challenges and Innovative Solutions
The primary challenge in Indian lending, especially for gig economy workers, is the lack of traditional credit signals. More than 36% of L&T Finance’s customers come without a credit history, making conventional assessment methods ineffective.
Key solutions include:
- Developing machine learning models that use lookalike performance as a target variable
- Creating ensemble probability models for different data sources
- Utilizing a granular classification system that divides India into 200 metro grids
- Analyzing affordability and fraud risk at a hyper-local level
Unstructured Data: A Revolutionary Approach
L&T Finance has pioneered the use of unstructured data in credit assessment, particularly in microfinance. Their innovative methods include:
- KYC Photo Analysis
- Using multimodal large language models to analyze applicant photos
- Developing a “lifestyle index” based on visual cues
- Extracting socioeconomic signals from images of applicants’ homes
- Video Verification Insights
- Analyzing video call backgrounds
- Extracting contextual information about applicants’ living conditions
- Using AI to generate objective creditworthiness signals
- Satellite Image Assessment
- Examining neighborhood characteristics
- Comparing applicant’s home location with surrounding infrastructure
- Generating additional creditworthiness indicators
Generative AI Integration
The organization views generative AI as a tool, not a complete solution. Their approach involves:
- Using prompted large language models to generate initial scores
- Combining generative AI outputs with traditional machine learning models
- Creating more nuanced and accurate credit assessment frameworks
Industry Impact and Future Vision
L&T Finance is methodically expanding its approach:
- Currently focusing on two-wheeler loans
- Planning expansion to urban and rural lending
- Exploring applications for tractors, personal loans, and potentially mortgages
The company aims to democratize credit access by:
- Building trust in underserved communities
- Providing financial opportunities for gig economy workers
- Leveraging advanced AI technologies to create more inclusive lending practices
Conclusion
The presentation highlighted a transformative approach to credit assessment, demonstrating how AI and alternative data sources can create more equitable financial systems. As Debarag Banerjee emphasized, in an environment where trust is scarce, technology can be magical—offering hope and opportunity to millions previously excluded from traditional financial services.
