Record matching is the linking of records from different sources to the same entity. The ideal case is a match on a unique identifier present in both databases – for companies this is the organisation's identification number. When the identifier is missing, a combination of characters with tolerance for small differences is used instead, that is, comparing the name, address, and other fields while allowing for typos and different forms of writing. This is where the risk of two types of error arises: unmatched records that belong together, and incorrectly matched records that do not belong together. The second error is worse, because it is hard to detect and it corrupts the data. A confidence threshold is therefore set, and records just below it are set aside for manual review instead of being merged automatically.
See also: Record deduplication, Data enrichment, Data normalisation.
