Fair, Isaac and Company (FICO)

Contribution to AI

For most of the twentieth century a lending decision was a judgement made by a person who read your application and, often, your face. The scorecard Fair Isaac began selling in 1958, with the American Investment Company as its first taker, substituted a weighted statistical model estimated from the repayment behaviour of past borrowers — an explicit prediction about an individual, produced by arithmetic, with money and refusal attached to the output. Adoption was slow and grudging for two decades. The bureau-based score released in 1989, built first with Equifax and then across all three national bureaus, changed what the prediction was: a single figure computed from a credit file rather than an application form, comparable between lenders and industries and available for nearly every adult in the country. In 1995 Fannie Mae and Freddie Mac adopted it for conforming mortgages, and a vendor's model became national infrastructure, with a cutoff on one company's number deciding who could buy a house. Falcon, developed at HNC Software from 1992 and acquired along with that firm in 2002, extended the same logic to card authorisations scored in milliseconds by neural networks trained on data pooled from competing banks. The historical consequence is that consumer credit is where a statistical model first made consequential decisions about millions of people, and therefore where regulators first demanded that such models be validated, tested for discrimination, and able to state the reasons behind a refusal.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).