AI Credit Decisioning with Alternative Data

The challenge

• Thin-file and new-to-credit customers are declined or offered low limits, even when many of them can repay
• Manual assessment is slow and inconsistent, while customers expect fast digital approval
• Customers who decline an offer or are declined often drop out, and the bank never learns why or offers a better-fitting alternative
• Early-warning and collections work still relies on fixed rules that miss each customer’s context
• Risk models must be explainable, auditable and compliant with regulatory expectations and PDPA

Stream’s approach

• Combine many data sources: the bank’s own data (deposits, transactions, product history), credit-bureau data and consented alternative data such as bill-payment behaviour or e-commerce data for merchants
• Machine-learning risk models with explainable AI that show which factors drove each customer’s score, so credit officers can understand and explain the outcome
• Decision engine that combines model scores with credit policy to set approval, limit and terms, with human review where judgement is needed
• AI agents across the credit lifecycle: role-based agents such as Underwriting & Portfolio and Early Warning & Collections, each made up of several task-based agents
• AI-enabled loan origination from partner VeriPark, covering auto, personal and home loans, credit cards, commercial loans, scoring and servicing, and a digital lending portal
• Model risk management: performance monitoring, drift and bias detection, and documentation for audit

Example AI agents in lending

• Instant pre-qualification: a mortgage or auto-loan chatbot gathers the details it needs, fills gaps, and sends structured data to the decision engine for a real-time, non-binding loan recommendation
• Point-of-sale pre-approval: a time-limited, pre-approved financing offer before the customer goes shopping, through conversational AI
• Smarter credit-limit increases: when a customer turns down a limit increase, the agent explores why, explains the benefits and offers better-fitting alternatives, such as a smaller increase or a later review, within policy
• Alternatives after a declined offer: understands the concerns of customers who decline a loan offer and presents other pre-approved options
• Routing silent accounts: detects non-responsive delinquent customers and hands them to staff with a context-rich summary
• SME lending: alerts on covenant breaches and early warning signs, offers grace periods when short-term financial stress appears, and prompts renewals of loans nearing maturity

Benefits

• Wider access to credit for customers the old process could not assess, with risk kept under control
• Faster, more consistent decisions
• Higher offer acceptance and customer satisfaction while staying within credit policy
• Earlier risk detection and better recovery while preserving the customer relationship
• Explainable, auditable decisions that support model governance and PDPA
• Credit officers can focus on complex cases

Who it is for

• Commercial banks and financial institutions growing retail, credit-card and SME lending
• Digital lenders that want responsible automated decisioning