How the Search Algorithm Works

Seeker is ComplyAdvantage's proprietary search and matching algorithm. It replaces the complex multi-parameter tuning of legacy screening systems with a single, interpretable output: a match score between 0 and 100.

What Seeker replaces

Legacy screening configurations required rules like: "only match when at least 2 out of 3 words match exactly and the third matches within edit distance 1 and is phonetically equivalent, and only do that for PEPs and sanctions but not adverse media." This is laborious to set up, maintain, and explain to regulators. Seeker eliminates all of this complexity.

The two-pass approach

First pass — Casting a wide net: Seeker uses equivalent name matching, fuzzy matching, phonetic matching (double metaphone algorithm), transliteration across scripts, and name order variations to retrieve every profile that could potentially be relevant. The goal is maximum recall — no genuine match should be missed.

Second pass — Filtering with machine learning: A machine learning model reviews all first-pass candidates and removes those that a human compliance analyst would almost certainly disregard as noise. The confidence with which each remaining profile is retained is quantified as the match score.

What makes Seeker distinctive: The ML model is trained on 10 years of real analyst remediation decisions from ComplyAdvantage clients — gathered with permission, reviewed, and annotated by in-house compliance experts. Seeker is, in effect, an automated support analyst.

How to use the match score

Set a per-risk-type match threshold. Any profile with a score above your threshold for that risk type will be surfaced. Raising the threshold reduces noise; lowering it increases recall. Thresholds are configurable per risk collection.