How I backtest a fraud rule before it ships
Four SQL checks for alert volume, known-fraud rate, overlap, and threshold selection before a rule reaches production.
analytics.fixelsmith.com
Fixel Smith on practical SQL, fraud detection, and public-sector data analytics.
Latest posts
What goes on a fraud team's dashboard, and what doesn't
Most fraud dashboards look better in a leadership deck than they work on an analyst's second monitor.
Five things that actually reduce noisy fraud alerts (and three that don't)
The standard fix doesn't work. Here's what does, with the SQL.
Detecting fraud rings: the social-graph problem in disguise
How to find connected fraud accounts using recursive CTEs and self-joins. Solves the graph-database problem without a graph database.
Eight window-function tricks beyond LAG and ROW_NUMBER
QUALIFY, frame specs, FILTER, gap-and-island, and the rest of the window-function patterns that turn five-line fraud rules into one-line filters.
Six SQL patterns I use to catch transaction fraud
The actual queries I run when I'm hunting fraud in transaction data. Velocity, impossible distances, suspicious amounts, merchant clusters, off-hours buys, and window functions.