Match Algorithms

This page explains how Data Cleanser matches records under the hood. For day-to-day setup, start with the Match Rules page — return here when you need to tune scenarios, debug unexpected groups, or explain behavior to administrators.

Composite duplicate keys

Every match rule normalizes a field value into a token. Tokens are concatenated with underscores into one composite key stored on each CRM record as DCM_Duplicate_Key__c. A human-readable version lives in DCM_Duplicate_Key_Re__c.

All match rules in a scenario are ANDed. Every populated rule must produce the same token for two records to group.
Final token normalization: remove spaces, uppercase.
Blank fields are omitted from the key. If Email is empty, that segment is skipped — not treated as “match on blank.” Two records can still group when their populated segments align, which can be broader than users expect.

Fuzzy is phonetic, not general typo tolerance. “Fuzzy” uses a 4-character Soundex-style code. Two values match only when their codes are identical after normalization — not when they are merely “close” in spelling. Use Word Mapping for synonyms like Corp → Corporation.

Matching types — implementation reference

Matching type	What the engine does
Exact	Raw field value, then space-stripped uppercase.
Fuzzy	Soundex-style 4-character code (see below).
Word Mapping	Lookup in Word Mapping list; fallback to raw value.
Word Mapping Fuzzy	Word Mapping normalization, then Soundex code.
Word Removal	Split on spaces; drop Remove Words tokens; concatenate remainder.
Word Removal Fuzzy	Word removal, then Soundex code.
Alpha and Numeric characters only	Extract `[A-Za-z0-9]+` segments and concatenate.
Numeric	Strip letters and common punctuation/symbols.
Relaxed Phone	Extract digit runs; drop a leading `1` (US country code).
First Word only	First space-delimited token only.
Street Address	Uppercase; replace CRESCENT→CRES, ROAD→RD, STREET→ST only.
X-Letters	First N characters, where N = count of X in the match type name.

How “Fuzzy” (Soundex-style) works step by step

Uppercase the input value.
Map letters to digit groups: BFPV→1, CGJKQSXZ→2, DT→3, L→4, MN→5, R→6 (other letters → 0).
Keep the first letter; deduplicate consecutive digit codes; strip zeros.
Pad or truncate to exactly 4 characters.
Two records match on a Fuzzy rule only when both codes are identical.

Examples: “Smith” and “Smyth” often share a code; “Acme Corp” and “Acme Corporation” will not unless Word Mapping normalizes them first.

Word Mapping and Remove Words pipeline

Dictionaries live under Settings → Word Mapping and Removal and are stored in the org-wide custom setting Word_Mapping_Public__c.

Word Mapping tab — Is_Remove__c = false. Maps Name → Matching_Word__c (e.g. Corp → Corporation).
Remove Words tab — Is_Remove__c = true. Tokens in Name are stripped before comparison.
Loaded once per scan transaction and cached in memory for that run.

Cross-object match keys

Lead side: field API names strip the Lead# prefix.
Contact side: uses Cross_Field_Name__c, strips Contact#; supports Account#Field to read from the contact’s related Account.
Cross-object keys are stored in DCM_Cross_Duplicate_Key__c — not the same-object DCM_Duplicate_Key__c field.

See also Cross-Object: Converting Duplicate Leads for the user-facing conversion workflow.