Overreacting to a Rare Default: How One Fintech Killed Growth by Overweighting a Single Story
A real-world example of Neglect of probability in action
Context
A mid-stage fintech startup offered small personal loans through an app, using automated credit models to approve borrowers and keep growth unit economics favorable. The business relied on steady approval rates and volume to cover fixed operating costs while maintaining a target portfolio default rate around 4%.
Situation
A single high-profile borrower fraud case — amplified by social media and a local news story — showed a dramatic, emotionally charged example of a customer who defaulted after faking identity documents. The board and some senior managers pushed for immediate, broad changes to the credit decision logic to eliminate that specific pattern of risk.
The bias in action
Leaders gave the anecdote disproportionate weight compared with the model's statistical performance: a vivid story displaced the model's historical probability estimates. Instead of running targeted experiments or computing expected loss, the team implemented a blunt rule that rejected a wide cohort of applicants with several borderline signals the anecdote had exhibited. The decision emphasized eliminating the possibility of that memorable failure over the much higher-probability everyday small defaults the model already handled well. Engineers and analysts felt pressure to ship the rule quickly, so it was rolled out platform-wide without an A/B test.
Outcome
Approval rates and loan volume dropped sharply while measured portfolio risk improved only marginally. Customer acquisition costs rose because marketing had to chase a smaller pool of eligible users. The company missed quarterly growth targets and lost negotiating leverage with investors, forcing a hiring freeze and paused feature development.