Distance matching compares two text values and returns a similarity score.

It is often used for fuzzy deduplication, where values are not identical but may still refer to the same person, company, product, or address.

Common distance algorithms

Two common algorithms are:

Levenshtein focuses on the number of edits needed to transform one string into another. Jaro-Winkler is often useful for shorter names because it gives more weight to shared prefixes.

What is a matching threshold?

A threshold defines how similar two values must be to count as a match.

For example, a threshold of 0.80 means the similarity score must reach 80%.

Lower thresholds find more possible duplicates but increase false positives. Higher thresholds are stricter but can miss messy duplicates.

📌 Practical rule

Start with a stricter threshold. Lower it only when you need broader candidate groups for manual review.

Datablist workflow

Datablist supports distance matching in the Duplicates Remover, with Levenshtein and Jaro-Winkler options.

Distance matching is available on paid plans. Select the algorithm and set its similarity threshold for the chosen text property.

Use distance matching for text columns such as names, company names, addresses, or product titles. Use exact or smart matching for identifiers such as emails, URLs, and domains.

For name matching, compare distance matching with phonetic matching. The main deduplication guide includes a tested Levenshtein workflow and examples of name-only false positives.

Distance and similarity are different scales

Raw Levenshtein distance counts edits: a smaller number means a closer match. A normalized similarity reverses that direction: a larger score means a closer match. Check which scale a tool uses before copying a threshold.

For example, Acme and Acne differ by one substitution. Datablist divides that edit count by the average string length, giving 1 - 1 / 4 = 0.75 similarity. A 70% property threshold accepts the pair; an 80% threshold rejects it. This is a spelling comparison, not evidence that the companies are identical.

The property threshold and the result match score are also different. Datablist assigns an accepted, non-identical Distance comparison a 90% field score; an identical comparison scores 100%. When several properties are selected, their field scores contribute to the overall record score. Review the actual values and another identifier before merging.

For incomplete records, review the empty value rules before lowering a similarity threshold. A blank secondary field changes the combined score.