Fuzzy matching finds text values that are similar without being identical. It is useful when typos, spelling variations, abbreviations, or inconsistent data entry prevent an Exact match.

Examples include:

  • Katherine Johnson and Katherin Johnson
  • Northwind Trading and Northwind Tradng
  • 12 Market Street and 12 Market St

How fuzzy matching works

A distance algorithm compares two strings and returns a similarity score. Levenshtein counts the edits needed to transform one value into another. Jaro-Winkler gives extra weight to matching prefixes and is often useful for short names.

A threshold decides which scores become candidate matches. Lowering the threshold finds more candidates but also increases false positives.

📌 Practical rule

Start with a strict threshold. Lower it only when you are prepared to review broader candidate groups.

A fuzzy match is a candidate, not proof of identity

Consider Acme and Acne. One character substitution gives a Levenshtein distance of 1. With Datablist's average-length normalization, their similarity is 75 percent. A 70 percent threshold admits that pair; an 80 percent threshold rejects it.

That score measures spelling similarity. It does not mean there is a 75 percent probability that the records describe the same company. Two different businesses can have nearly identical names.

For deduplication, keep a stable identifier alongside the fuzzy field. Check an agreeing domain or contact ID, and review conflicting addresses before merging. A strict threshold alone cannot distinguish every namesake.

Fuzzy matching is not normalization

Smart matching cleans formatting differences before comparison. Fuzzy matching measures how close the remaining strings are.

Use Smart matching for predictable differences such as URL protocols, phone formatting, email aliases, punctuation, or word order. Use fuzzy matching when the value itself may contain a typo.

Use fuzzy matching safely for deduplication

Names alone are weak identifiers. Confirm fuzzy name matches with an email, phone number, domain, address, or external ID before merging records.

Datablist exposes Levenshtein and Jaro-Winkler through the Distance algorithm in its Duplicates Finder. These advanced algorithms are available on paid plans.

Compare the matching options in the data matching guide, or follow the complete list deduplication workflow.

When another comparison field is missing, check the empty value rules separately from the spelling threshold.