How the Search Algorithm Works

The match score is a number between 0 and 100 assigned to each profile returned in a screening case. It represents the predicted probability that a compliance analyst would escalate that profile as a true positive alert.

The seven influencing factors

• Exact names (No effect to Very Strong) — exactly matched name tokens and word order. Scales with name rarity.

• Equivalent names (No effect to Very Strong) — initials, hypocorisms, abbreviations. Scales with name rarity.

• Inexact names (No effect to Very Strong) — edit distance 1, phonetic match, concatenation. Scales with name rarity.

• Unmatched names (Very Strong Negative to Very Strong Positive) — unmatched word components. Always negative.

• Year of birth (Very Strong Negative to Very Strong Positive)) — exact year match. Not available for companies. Year only, not full date of birth.

• Countries (Very Strong Negative to Very Strong Positive)) — all country associations: birth, residence, office, media jurisdiction.

• Name rarity (scalar) — multiplies the three name factors. Rare names amplify the score; common names dampen it.

Key insight on common names: An exact match on "Muhammad Ali" produces a score of 34 — name commonness limits its evidential weight. Adding a country match raises it to 67. Adding year of birth raises it to 95. Providing more data always helps for common names. For rare names like "Concetta Fierravanti-Wells", an exact match alone scores 98.

Score determinism

The score is deterministic: the same search on the same configuration always produces the same score. Scores change if you modify input data, change threshold configuration, or if ComplyAdvantage releases a model update (clients are notified of model changes).