What is entity resolution in AML screening?
Entity resolution is the technique of deciding whether the party named in a customer record and the party named in a watchlist entry are, in fact, the same real-world person or organisation. It is not string comparison; it is a judgement about identity that weighs names, dates of birth, identifiers, jurisdictions and relationships together. This guide sets out what entity resolution is, how it differs from fuzzy matching, why it reduces false positives without loosening thresholds, and how it powers deep-tier ownership analysis.
Entity resolution is the process of determining whether two records, typically a customer profile and a watchlist entry, refer to the same real-world person or organisation. In AML screening it is the step that turns a raw name comparison into a defensible decision: this alert is the same individual, that one is a different person who happens to share a name. Where a simple name match produces a long list of look-alikes, entity resolution assesses each candidate against everything known about both parties and reaches a conclusion about identity, which is the question a screening control actually needs answered.
What is entity resolution?
Entity resolution treats a match as an identity problem rather than a text problem. When a customer record is screened, the engine finds candidate watchlist entries and then has to decide, for each one, whether it describes the same party. Resolution makes that decision by combining several signals at once: the name and its variants, dates of birth, national and passport identifiers, registration numbers, country and jurisdiction, and known relationships. A candidate that agrees on name but disagrees on date of birth and nationality is a different person; a candidate that agrees across several strong identifiers is very likely the same one. The output is not a similarity percentage but a reasoned view of who the party is, which is what a reviewer, and an examiner, needs to act on.
How does it differ from fuzzy matching?
Fuzzy matching measures how similar two strings are, and it is a genuinely useful input, but it is not entity resolution. A fuzzy algorithm scores "Mohammed Al Sayed" against "Muhammad El-Sayed" and reports a high similarity because the characters and phonetics line up. What it cannot tell you is whether those two labels belong to the same human being. Two unrelated people can share a near-identical name, and one person can appear under spellings that a string comparison scores as distant. Entity resolution uses the fuzzy score as one piece of evidence and then asks the identity question by bringing in the other attributes: birth dates, identifiers, jurisdiction, roles and relationships. Fuzzy matching narrows the field; resolution decides. Treating a high name score as a match, without resolving identity, is exactly how a screening program drowns reviewers in look-alike alerts.
Why does it beat loosening match thresholds?
The instinct when a screen misses a real match is to loosen the matching threshold so more candidates surface, and the instinct when reviewers are overwhelmed is to tighten it so fewer do. Both are blunt, because a single threshold trades recall against precision along one axis: the name. Loosen it and you catch more true matches but bury them under look-alikes; tighten it and the noise falls but real matches with spelling variation slip through. Entity resolution changes the trade by adding axes. Because it decides on identity using several attributes, it can keep the name threshold sensitive and still discard weak candidates that disagree on date of birth, identifier or jurisdiction. You do not have to choose between catching variants and controlling volume. This is the mechanism behind reducing false positives without weakening detection, explored further in our guide to reducing false positives in AML screening.
How does it apply the 50% Rule and ownership?
The clearest place entity resolution earns its keep is ownership analysis. Under the 50% Rule, an entity that is majority owned or controlled by a sanctioned party is itself treated as sanctioned, even when it appears on no list. Testing for that means tracing ownership across subsidiaries, holding structures and nominee arrangements, and at every layer you have to answer the same identity question: is the shareholder named in this corporate filing the same party as the one on the watchlist, or merely a namesake? That is entity resolution, applied repeatedly across a graph of companies and individuals. Without it, deep-tier ownership mapping either misses control that hides behind spelling variants and intermediaries, or floods the analysis with false links. The 50% Rule is a screening obligation that a name match alone cannot satisfy, which is why it is set out in the complete guide to AML screening.
What does the data need to make it work?
Entity resolution can only weigh the attributes it is given, so the quality of the decision depends on the data on both sides of the match. A watchlist entry that carries nothing but a common name forces resolution to lean on the one signal that is least reliable; an entry with aliases, transliterations, a date of birth, identifiers and a jurisdiction gives it the material to separate the real party from the look-alikes. The same holds for the customer record: a screen against a bare name will always resolve less confidently than a screen against a record with structured identifiers. The table below sets out the attributes resolution relies on and what each one contributes.
| Attribute | What it contributes | Effect when missing |
|---|---|---|
| Name and variants | Finds candidate matches, including aliases and transliterations | Nothing to match on; the screen fails outright |
| Date of birth | Separates namesakes born in different years | Two different people collapse into one alert |
| National and passport identifiers | Strong confirmation or exclusion of identity | Resolution falls back to weaker signals |
| Jurisdiction and address | Distinguishes parties by country and location | Cross-border look-alikes are hard to split |
| Ownership and relationships | Supports the 50% Rule and associate networks | Control through intermediaries stays hidden |
Structured, richly attributed data is not a nicety here; it is the precondition for resolution to reach a confident decision rather than defaulting to the name.
How does BriteBase use entity resolution?
Our screening engine treats every alert as an identity question rather than a string score. Agentic entity resolution weighs names and variants against dates of birth, identifiers, jurisdiction and relationships to decide whether a customer and a watchlist entry are the same party, which is designed to reduce false positives by up to 80% without loosening the underlying match sensitivity. The same resolution runs across ownership graphs, so the 50% Rule and deep-tier control are applied as part of screening rather than left to a reviewer's manual research. The data layer behind it is structured for resolution, with aliases, identifiers and ownership carried on the entries rather than delivered as flat feeds. It is available as the risk-intelligence layer behind an existing stack, or consumed directly through the BriteBase screening platform and the sanctions, PEP and adverse-media data layer.
FAQ
What is entity resolution in AML screening?
Entity resolution is the process of deciding whether two records, usually a customer profile and a watchlist entry, refer to the same real-world person or organisation. In screening it is the step that turns a raw name comparison into a defensible identity decision, so a reviewer can say this alert is the sanctioned party and that one is a different individual who happens to share a name. Resolution reaches that conclusion by weighing several attributes together rather than the name alone: aliases and transliterations, dates of birth, national and passport identifiers, jurisdiction, and known relationships. The consequence for a program is fewer look-alike alerts reaching reviewers and clearer evidence behind the ones that do. Because the decision rests on identity rather than string similarity, it is far easier to defend to an examiner asking why a particular match was cleared or escalated during a review.
How is entity resolution different from fuzzy matching?
Fuzzy matching measures how similar two text strings are; entity resolution decides whether two records describe the same party. The difference matters because a high name-similarity score does not establish identity. A fuzzy algorithm will rate two unrelated people with near-identical names as a strong match, and it will score one person's variant spellings as distant, so on its own it produces look-alike noise and missed variants in equal measure. Entity resolution uses the fuzzy score as one input, then brings in dates of birth, identifiers, jurisdiction and relationships to answer the identity question the name score cannot. Fuzzy matching narrows the candidate field; resolution reaches the conclusion. The practical consequence is that a program relying on name scores alone must either tolerate heavy false-positive volume or risk clearing genuine matches, whereas resolution separates the two cases on evidence rather than on where a single threshold happens to sit.
Does entity resolution loosen match thresholds?
No, and that is precisely its value. A single match threshold forces a trade-off along one axis, the name: loosen it to catch spelling variants and you surface more look-alikes, tighten it to cut noise and you miss real matches. Entity resolution does not touch that dial to control volume. Instead it adds axes, deciding identity by weighing dates of birth, identifiers, jurisdiction and relationships alongside the name. That lets a program keep the name threshold sensitive, so genuine variants still surface, while confidently discarding candidates that agree on name but disagree on the other attributes. The result is fewer false positives with detection intact, rather than fewer false positives bought by weakening the screen. This is why resolution, not threshold tuning, is the sustainable answer to alert overload as watchlist coverage and customer volume grow, and why loosening thresholds tends to trade one failure mode for another.
How does entity resolution support the 50% Rule?
The 50% Rule treats an entity majority owned or controlled by a sanctioned party as sanctioned itself, even when that entity is on no list. Testing for it means tracing ownership through subsidiaries, holding companies and nominee arrangements, and at every layer the same identity question recurs: is the shareholder named in this filing the sanctioned party, or a different person with a similar name? Answering that reliably across a graph of companies and individuals is entity resolution applied repeatedly. Without it, ownership analysis either misses control that hides behind variant spellings and intermediaries, or it over-links unrelated namesakes and buries the real exposure. The consequence is that the 50% Rule cannot be satisfied by a direct name screen alone; it depends on resolving parties consistently through the ownership chain, which is why deep-tier ownership mapping belongs in the screening layer rather than in a reviewer's manual research.
What data quality does entity resolution need?
Entity resolution can only weigh the attributes it is given, so its accuracy depends on the richness of the data on both sides of the match. A watchlist entry carrying nothing but a common name forces resolution onto the least reliable signal, while an entry with aliases, transliterations, a date of birth, identifiers and a jurisdiction gives it the material to separate the real party from namesakes. The same applies to the customer record: a bare name resolves less confidently than a record with structured identifiers. Ownership and relationship data matters too, because the 50% Rule and associate analysis depend on it. The consequence is that data quality is not a background concern but the precondition for resolution to reach a confident decision rather than defaulting to the name. Investing in structured, well-attributed records on both sides does more for alert quality than tuning the matching algorithm alone.
Sources
See entity resolution on your own alerts.
Book a demo and we will show you agentic entity resolution across sanctions, PEP and adverse-media screening, with deep-tier ownership and the 50% Rule applied as part of the work. Designed to reduce false positives by up to 80% without loosening detection.
Book a demo