· 7 min read · Dawn Lith Advisory
The comfort of volume
Many firms run screening at loose thresholds because a missed match is the worst outcome. The cost is a queue of false positives, analysts who clear alerts on reflex and a real risk that a true match is dismissed with the rest.
Test before you tune
Tuning without testing is guesswork. Build a test set of known names with controlled variations: transliterations, reordered names, missing elements, common misspellings. Run it at several threshold settings and measure what is detected and what is missed.
The result is a curve, not a single answer. It shows where additional alerts stop adding detection, which is the evidence you need to justify a setting.
Document the decision
Record the test data, the results, the chosen settings and who approved them. Repeat the exercise when lists, data or the system change. A supervisor will accept a reasoned threshold far more readily than an unexplained default.