Base Rates Beat Gut Feel: Bayesian Thinking for Business Decisions

A fraud detector flags a transaction. The vendor says it is 99% accurate. What is the chance this transaction is actually fraud? Most people answer "99%". The real answer, when 0.1% of transactions are fraudulent, is roughly 9%. Nine out of ten flags are false alarms — not because the tool is broken, but because the base rate is low. Ignoring base rates is the most expensive habit in applied statistics, and Bayesian thinking is the cure.

The One Calculation Worth Memorizing

Out of 10,000 transactions, 10 are fraud (0.1% base rate). A 99%-accurate test catches ~10 of them (true positives) and wrongly flags 1% of the 9,990 honest ones — about 100 false positives. So 110 flags contain 10 real frauds: 10 / 110 ? 9%.

The general form is Bayes' rule, but you rarely need the formula — you need the habit: start from the base rate, then update. The test result moves you from 0.1% to 9%, a ninety-fold update and genuinely useful triage. The error is starting from 50/50 ("flagged or not, who knows?") instead of from the base rate.

Where Base Rates Get Ignored

• Hiring and admissions. An impressive interview moves the needle — but from the base rate of applicants like this one, not from zero. Structured rubrics exist to force the update to be explicit rather than vibes-based.
• Medical-style screening analogies. Churn-risk scores, lead scores, anomaly flags: any rare-event detector lives in the fraud-detector arithmetic above. Always ask for precision (of all flags, how many are real?) alongside accuracy.
• Project forecasts. Your team's plan says six weeks. The base rate for similar projects is fourteen. The plan is evidence, but it updates from fourteen, not from six — this is the planning fallacy, and reference-class forecasting (ask "how long did the last five like this take?") is the Bayesian fix.
• Alarms and alerts. An alert that fires constantly trains everyone to ignore it — rationally, because its precision collapsed. Every alert threshold is a bet about base rates; tune accordingly.

A Practical Bayesian Habit

For any uncertain judgment, write down three numbers before deciding: the base rate (how often does this happen in general?), the hit rate (if it were true, how likely is this evidence?), and the false-alarm rate (if it were false, how likely is this evidence anyway?). Even rough guesses beat gut feel, because gut feel systematically skips the base rate and overweight the vivid evidence in front of you.

Two honesty notes. First, uncertainty about the inputs is itself worth reporting — a forecast with a calibrated interval beats a confident point estimate built on guessed priors. Second, base rates cut both ways: they restrain hype about rare events, but they also protect unglamorous bets with strong base rates from being talked down by one scary anecdote. The discipline is symmetric — start from what usually happens, update with what you see, and show your working.