Fuzzy Join

VLOOKUP for messy data: join two CSVs on approximately-matching keys ('Acme Ltd.' = 'ACME Limited') with Jaro-Winkler or Levenshtein scoring and an interactive match-review queue.

ETLUniqueBrowser workspace

Free to use. No account, no daily limit, and no backend processing for tool input.

Tip: use samples, upload, copy, download, and send-to actions inside the workspace where available.

Actions

Smart Data Inspector

Bounded preview analysis. No upload, no backend.

Score

Rows

Columns

Empty

Duplicates

FormatTEXT

Memory0 B

EncodingBrowser text

Quality checks

WARN Paste or drop a CSV, JSON, XML, or text sample to inspect it.

Recommended

Timeline

Waiting for input

Data workflow

Inspector-supported workspace

Fuzzy Join can work with structured samples, and the sidebar inspector can profile a bounded preview without uploading your data.

Preview analysisColumn-awareClient-side

Use the inspector for quick row, column, missing-value, duplicate, type, and quality signals.
For large files, the inspector analyzes a capped preview so memory use stays predictable.
Use related tools to continue into cleaning, validation, conversion, or reporting workflows.

The inspector is a lightweight preview helper; the main tool remains responsible for the full transformation or export.

Related tools

Related guides

All guides

Fuzzy Join Online GuideVLOOKUP for messy data: join two CSVs on approximately-matching keys ('Acme Ltd.' = 'ACME Limited') with Jaro-Winkler or Levenshtein scoring and an interactive match-review queue. Use Fuzzy Join in your browser with no signup, no daily limit, and no backend processing for tool input.

What is Fuzzy Join?

Fuzzy Join matches two CSVs on keys that nearly agree instead of exactly agreeing — "Acme Ltd." against "ACME Limited", "O'Brien" against "OBrien". You choose the similarity algorithm (Jaro-Winkler for names, Levenshtein for typos), set a threshold, and then review the borderline matches in a queue rather than trusting a number. That review step is the important one: every fuzzy join has a band where only a human can tell a real match from a coincidence.

How to use Fuzzy Join

1Load both CSVs and choose the column in each to match on.
2Pick the scoring algorithm — Jaro-Winkler favours matching prefixes, Levenshtein counts edits.
3Set a similarity threshold to divide automatic matches from ones needing review.
4Work through the review queue, accepting or rejecting each borderline pair.
5Export the joined result once you are satisfied with the matches.

Features

Jaro-Winkler and Levenshtein similarity scoring
Adjustable threshold separating automatic matches from review
Interactive review queue for borderline pairs
Per-match score shown so you can judge the cutoff
Unmatched rows from both sides reported rather than silently dropped
Runs in the browser — neither file is uploaded

Common uses

Joining company or customer lists from two different systems
VLOOKUP-style matching where the keys were typed by hand
Reconciling records after a merger or migration
Deduplicating a list against a master reference

Used by

Data engineers
Analytics engineers
Database administrators
Operations teams

Common errors

Setting the threshold too low and accepting coincidental matches — short strings score highly against each other far too easily.
Skipping the review queue, which is where the genuinely ambiguous cases live.
Matching on a column with a common prefix ("The ", "Dr ") without normalising first, which inflates Jaro-Winkler scores across unrelated records.
Assuming one-to-one matching: without a constraint, several rows on one side can all match the same row on the other.

How it works

1Every candidate pair of key values is scored by the algorithm you chose, producing a similarity between 0 and 1.
2Pairs above your threshold are matched automatically; pairs in the band below it are sent to the review queue instead of being discarded.
3Your accept and reject decisions are applied to build the final joined output, with unmatched rows from both sides reported separately.

Recommended workflow

1
Check key overlap first
See how well the two key columns correspond before joining.
Open tool
2
Normalise the keys
Trim, case-fold and standardise before matching to raise accuracy.
Open tool
3
Merge cleanly afterwards
Combine the matched result with other datasets.
Open tool

Frequently asked questions

Jaro-Winkler or Levenshtein — which should I use?: Jaro-Winkler for names and company names, because it rewards matching prefixes and handles transpositions well. Levenshtein when you expect typos or missing characters anywhere in the string, since it simply counts the edits needed.
What threshold should I set?: Start around 0.85 for Jaro-Winkler on names and adjust while watching the review queue. Any single number will be wrong for some pairs, which is exactly why the queue exists.
Why did two obviously different rows match?: Short keys. "AB Ltd" and "AC Ltd" differ by one character out of six, which scores highly on any edit-based measure. Match on a longer or more distinctive column where you can.
Are my files uploaded?: No. Both files are parsed and matched in your browser; the site has no backend to send them to.

Fuzzy Join

ReadyHow to use

VLOOKUP for messy data: join two CSVs on approximately-matching keys ('Acme Ltd.' = 'ACME Limited') with Jaro-Winkler or Levenshtein scoring and an interactive match-review queue.

ETLUniqueBrowser workspace

Free to use. No account, no daily limit, and no backend processing for tool input.

Table A · keep every rowCSV

Try a sample:

Table B · lookup tableCSV

Try a sample:

Key column · Athe messy text column to match on

Key column · B

Similarity metric

Normalize before matching

Auto-accept thresholdreview band: 0.60 → 0.90

0.90

Paste or drop two CSVs, pick the key column on each side, and get a VLOOKUP-style join that survives typos, casing, and “Acme Corp” vs “ACME Corporation”.

Tip: use samples, upload, copy, download, and send-to actions inside the workspace where available.

Actions

Smart Data Inspector

Bounded preview analysis. No upload, no backend.

Score

Rows

Columns

Empty

Duplicates

FormatTEXT

Memory0 B

EncodingBrowser text

Quality checks

WARN Paste or drop a CSV, JSON, XML, or text sample to inspect it.

Recommended

Timeline

Waiting for input

Data workflow

Inspector-supported workspace

Fuzzy Join can work with structured samples, and the sidebar inspector can profile a bounded preview without uploading your data.

Preview analysisColumn-awareClient-side

Use the inspector for quick row, column, missing-value, duplicate, type, and quality signals.
For large files, the inspector analyzes a capped preview so memory use stays predictable.
Use related tools to continue into cleaning, validation, conversion, or reporting workflows.

The inspector is a lightweight preview helper; the main tool remains responsible for the full transformation or export.

Related tools

Related guides

All guides

Fuzzy Join Online GuideVLOOKUP for messy data: join two CSVs on approximately-matching keys ('Acme Ltd.' = 'ACME Limited') with Jaro-Winkler or Levenshtein scoring and an interactive match-review queue. Use Fuzzy Join in your browser with no signup, no daily limit, and no backend processing for tool input.

What is Fuzzy Join?

How to use Fuzzy Join

1Load both CSVs and choose the column in each to match on.
2Pick the scoring algorithm — Jaro-Winkler favours matching prefixes, Levenshtein counts edits.
3Set a similarity threshold to divide automatic matches from ones needing review.
4Work through the review queue, accepting or rejecting each borderline pair.
5Export the joined result once you are satisfied with the matches.

Features

Jaro-Winkler and Levenshtein similarity scoring
Adjustable threshold separating automatic matches from review
Interactive review queue for borderline pairs
Per-match score shown so you can judge the cutoff
Unmatched rows from both sides reported rather than silently dropped
Runs in the browser — neither file is uploaded

Common uses

Joining company or customer lists from two different systems
VLOOKUP-style matching where the keys were typed by hand
Reconciling records after a merger or migration
Deduplicating a list against a master reference

Used by

Data engineers
Analytics engineers
Database administrators
Operations teams

Common errors

Setting the threshold too low and accepting coincidental matches — short strings score highly against each other far too easily.
Skipping the review queue, which is where the genuinely ambiguous cases live.
Matching on a column with a common prefix ("The ", "Dr ") without normalising first, which inflates Jaro-Winkler scores across unrelated records.
Assuming one-to-one matching: without a constraint, several rows on one side can all match the same row on the other.

How it works

1Every candidate pair of key values is scored by the algorithm you chose, producing a similarity between 0 and 1.
2Pairs above your threshold are matched automatically; pairs in the band below it are sent to the review queue instead of being discarded.
3Your accept and reject decisions are applied to build the final joined output, with unmatched rows from both sides reported separately.

Recommended workflow

1
Check key overlap first
See how well the two key columns correspond before joining.
Open tool
2
Normalise the keys
Trim, case-fold and standardise before matching to raise accuracy.
Open tool
3
Merge cleanly afterwards
Combine the matched result with other datasets.
Open tool

Frequently asked questions

Jaro-Winkler or Levenshtein — which should I use?: Jaro-Winkler for names and company names, because it rewards matching prefixes and handles transpositions well. Levenshtein when you expect typos or missing characters anywhere in the string, since it simply counts the edits needed.
What threshold should I set?: Start around 0.85 for Jaro-Winkler on names and adjust while watching the review queue. Any single number will be wrong for some pairs, which is exactly why the queue exists.
Why did two obviously different rows match?: Short keys. "AB Ltd" and "AC Ltd" differ by one character out of six, which scores highly on any edit-based measure. Match on a longer or more distinctive column where you can.
Are my files uploaded?: No. Both files are parsed and matched in your browser; the site has no backend to send them to.

What is Fuzzy Join?

How to use Fuzzy Join

Features

Common uses

Used by

Common errors

How it works

Recommended workflow

Check key overlap first

Normalise the keys

Merge cleanly afterwards

Frequently asked questions

What is Fuzzy Join?

How to use Fuzzy Join

Features

Common uses

Used by

Common errors

How it works

Recommended workflow

Check key overlap first

Normalise the keys

Merge cleanly afterwards

Frequently asked questions