Is entity resolution the same as deduplication?

They are related but not identical. Deduplication removes repeated records within a single source; entity resolution links records that describe the same entity across different systems, often with incompatible schemas and keys.

Do you need a dedicated tool to do entity resolution?

For small volumes with clean shared keys, hand-written rules or SQL queries are enough. With large volumes, dirty data or no shared key, you need libraries or platforms dedicated to probabilistic matching, capable of scoring confidence across millions of record pairs.

What is entity resolution?

The discipline that recognizes when different records in different systems represent the same real-world entity.

Entity resolution is the discipline that recognizes when different records, coming from different systems, represent the same real-world entity: the same person, the same company, the same product. A CRM, a billing system and a customer registry can each hold a separate row for "John Smith", "J. Smith" and "Smith, John" without any of the three systems knowing they refer to the same person. Entity resolution links these records and consolidates them into a single identifier, the golden record: the most complete and reliable version obtained by combining attributes from the original sources. The process rests on two families of techniques, often combined: deterministic matching, which compares exact keys such as a tax ID or VAT number, and probabilistic or fuzzy matching, which estimates a match probability across attributes that are similar but not identical, such as a name, address or phone number written differently across systems.

See also: Data quality · How fit your data is for its intended use: complete, correct, fresh and consistent across systems. Measured, not declared.

Deterministic, probabilistic, and the role of the graph

Deterministic matching is fast and reliable whenever a clean shared key exists, but it fails as soon as data is dirty or sources share no common identifier. Probabilistic matching fills that gap: it weighs similarity across fields (string distance, name phoneticization, address normalization) and produces a confidence score rather than a binary certainty, which is why it needs calibrated thresholds and often a queue of ambiguous cases for review. On modern platforms the result is often represented as an entity graph, where each node is a source record linked to its golden record, a structure that extends naturally into a knowledge graph once relationships between entities matter, not just their unique identity.

A concrete enterprise example

A banking group with three commercial networks acquired over the years discovers it has profiled the same corporate client independently on each CRM system, under slightly different legal names and with no shared key. Without entity resolution the group cannot tell its total exposure to that client, nor apply a single risk limit. With an entity resolution pipeline combining deterministic matching on VAT number where available and probabilistic matching on the rest, the three profiles converge into one golden record, from which the correct aggregate exposure and a single client view for risk policies both follow.

Why it matters for decision makers

Without entity resolution every business system counts reality its own way: active customer counts, risk exposures and regulatory reporting diverge depending on which database calculates them. It is the technical foundation on which master data management and every unified customer view are built, including the ones behind a CDP: none of these initiatives work if the underlying system cannot recognize when two records describe the same entity. For the data quality discipline entity resolution depends on, see data quality.