In this episode of Bits & Banks, Dr. Parisa Milne, Co-Founder and Chief Decision Scientist at Luminnovate, joined Aditya Khandekar of Corridor Platforms to unpack what lenders can learn from retail, from behavioral signals and data sparsity to personalized offers and pricing

Key ideas:

• Sparsity is often a continuity problem: A thin credit file is one version of data sparsity. Another is the borrower who applies four or five times, each time with a three or six-month bank snapshot, leaving gaps and overlaps. Parisa's point is that sparsity is not always missing data. Sometimes it is missing continuity and context, and open banking in Australia makes cash flow data a lot easier to access for non-bank lenders.
• Behavioral signals need proof, not just volume: Thousands of inflows and outflows become readable concepts such as ability to pay, stability, conduct and trajectory. Each signal is validated against outcomes like early delinquency, long-term delinquency, propensity to renew and recovery after a late payment, and carries its own coverage, confidence and reliability. Together they form a signal registry that works across the lending lifecycle.
• Business context makes SME signals readable: A quiet month means one thing for a cafe and another for an online retailer or a tradie. Parisa sees AI agents trained on a specific business model working alongside ML models, because ML unlocked big data while agents work well with small data. Governance guardrails matter, and the agents should not generate anything new. Aditya adds that this context can sit as rule overlays on core risk assessment and extend financial inclusion to newer businesses, within fair lending requirements.
• Retail personalization chases incremental value, not acceptance: Retail customers buy often, outcomes show up quickly, and each decision costs little. The goal is the action that changes behavior, not an offer to a customer who would have bought anyway.
• In lending, the offer changes the risk: Delinquency takes months to observe and charge-offs take longer, while a wrong call can mean financial loss, customer hardship or regulatory consequences. Price is one lever. The structure of the offer, including repayment, also changes default risk. The best offer can be a smaller amount, a longer term, repayment aligned to cash flow, a competitive rate for a valuable customer, or no offer at all.
• Pre-approval is where mid-market lenders can compete: Customer state, portfolio state and decision context give a lender a clear view of what it knows about each borrower and how confident it is, then roll up into the portfolio for monitoring and early warning. Aditya's point: pre-approval is where most offers are generated, and community and mid-market institutions can compete with tier-one lenders when they keep borrowers' credit healthy and avoid overextending them at origination.