Is the Match Score Fully Explainable?

Yes — but "fully explainable" has a precise meaning in this context that is worth understanding, especially when discussing the platform with regulators or clients.

What "fully explainable" means

Every match score is the output of a logistic regression model. The exact inputs going into every score are known and traceable: the seven influencing factors (exact names, equivalent names, inexact names, unmatched names, year of birth, countries, and name rarity) are calculated from the underlying data and fed into the model. ComplyAdvantage knows precisely which numbers went in and how they were transformed into the final probability.

What the influencing factors view shows

Rather than exposing the full mathematical detail to every user, Seeker surfaces a human-readable version: which factors drove the result, in which direction, and with what relative strength. This is sufficient for compliance teams to understand and defend their screening decisions to governance teams and regulators.

What the client needs to understand for each profile result is:

• There are specific factors underpinning the final score

• Each factor affects the score positively or negatively, with varying strength

• The final score expresses the system's confidence that the result is relevant and should be escalated

Important caveat: The influencing factor values cannot be summed to arrive at the match score percentage. The score is a point on a logistic curve, not an arithmetic total. A factor that moves the score by 60 points in one match might move it by only 5 points in another — depending on where the overall score sits on the curve. This is expected behaviour, not a limitation.

The full methodology

The exact model weights and the full breakdown of individual raw scores are ComplyAdvantage intellectual property. They are not exposed to clients, but they are fully auditable internally. This is consistent with how explainability works in other model-driven scoring systems — analogous to a credit score, where contributing factors are transparent but the precise mathematical transformation is proprietary.