Data Quality
Fuzzy Match
Groups values that are spelled alike — Jon and John, Acme Inc and ACME Inc. — and writes each row's group number, or the group's first spelling, into a new column.
When to use it
Use it to clean up hand-typed names before grouping, joining or counting, when exact matching would treat every typo as a different customer.
In a pipeline
File InputFuzzy MatchSummarize (Multi-Aggregate)
Settings
| Setting | Type | Required | Default or hint |
|---|---|---|---|
| Columns to compare | select | Yes | Pick at least one column |
| Algorithm | select | — | levenshtein |
| Similarity (0–1) | number | — | 0.85 |
| Output | select | — | group |
| New Column Name | text | — | Default: Match Group / Best Match |
| Insert After | select | — | End of table |
| Max comparisons | number | — | 5000000 |
Example
One spelling per customer
- Columns to compare=Customer
- Algorithm=Jaro-Winkler (short names)
- Similarity (0–1)=0.85
- Output=First spelling in the group
| Customer |
|---|
| Jon Smith |
| Mary Jones |
| John Smith |
| Marie Jones |
| Customer | Best Match |
|---|---|
| Jon Smith | Jon Smith |
| Mary Jones | Mary Jones |
| John Smith | Jon Smith |
| Marie Jones | Mary Jones |
Worth knowing
- Levenshtein suits typos in longer text; Jaro-Winkler suits short names; Soundex groups names that sound alike (Robert, Rupert) and ignores Similarity.
- Case, accents, punctuation and extra spaces are ignored before comparing. Blank values get no group.
- Groups chain: if A is like B and B is like C, all three share a group even when A and C are not alike. Raise Similarity if groups grow too broad.
- Every pair of distinct values is compared, which grows fast — 10,000 distinct values is 50 million comparisons. Max comparisons (5,000,000 by default) stops the run before that happens; filter first or use Soundex.