Max Levchin claims any single data point that seems to dramatically improve underwriting accuracy is a red herring. He argues these 'magic bullets' are brittle and fail when market conditions shift. A robust risk model instead relies on aggregating small lifts from many subtle factors.

By eliminating outdated constraints like the six-month activity rule and incorporating time-series data and alternative inputs like rent payments, modern credit scoring models can assess millions of creditworthy individuals, such as military personnel or young people, who were previously unscorable.

Grab leverages its rich transaction data—like a merchant's daily cash flow or a driver's income—to create proprietary credit scores. This allows it to safely underwrite loans for unbanked individuals and small businesses, a segment traditional banks avoid due to a lack of data.

In the 80s, credit was binary: a high score got a card, a low score got nothing. Capital One pioneered an "information-based strategy," using data to test and price risk for consumers just below the traditional cutoff, effectively creating the modern data-driven lending model.

Lending platform Split completely rejects FICO scores, finding them unhelpful. Instead, it built its own foundation model for underwriting based purely on cash-flow analysis and trained on its own data. This approach yields stunningly better performance, particularly for demographics like millennials whose financial lives aren't captured by traditional credit metrics.

Fintechs often promise partner banks that their proprietary user data will lead to better credit underwriting. However, FICO scores are "unreasonably effective," and most alternative data sources that genuinely improve predictions are illegal to use, like zip codes, due to redlining concerns.

By leveraging a complete financial picture of its users—income, spending, bill payments, and government data—Kaspi's super app can approve 99.9% of loan applications automatically in under six seconds with impressively low default rates.

AI is not replacing credit analysts but augmenting them like a "driver assist" feature. It rapidly parses data rooms, finds hidden connections, and reduces cognitive load, allowing analysts to perform more iterations of their investment thesis in the same amount of time.

With many "Buy Now, Pay Later" (BNPL) services not reporting to credit bureaus, lenders face "stacking" risk where consumers take on invisible debt. To get a holistic view, lenders are increasingly incorporating cash flow data, like checking account trends, into their underwriting processes.