What is Fraud Score?
A fraud score squeezes many signals—device, behavior, payment, corridor risk, name‑screening results—into a single number or label that drives decisions. Approve, step up, hold, or decline. It’s triage math, not truth. Inputs change; weights drift; regulators ask why you chose that threshold yesterday and a different one today.
Design the score to be explainable. Log top features with contributions; keep versions; run champion/challenger. Segment thresholds by product and geography. Low scores should glide with minimal friction; mid scores deserve targeted challenges; high scores get holds or hard declines. Bias hides inside data—watch cohorts and error rates, not just global AUC.
