Phonetic matching lets the engine treat names that sound the same as potential matches, even if they're spelled differently. Smith and Smyth. Johnson and Johnsen. Catherine and Katharine.
It's on by default for the Person profile and off by default for everything else. Understand when to turn it on and when to leave it off, because in the wrong context it hurts accuracy.
How it works
For each name column, the engine computes a phonetic code — a compact representation of how the name sounds. Two names with the same code are phonetic matches.
The phonetic coding handles:
- Silent letters (
Knightis phonetically equivalent to names starting with theNsound) - Vowel variation (
CatherineandKatharinecode the same way) - Foreign name conventions (
Johnsen,Johanssen,Johansencode the same way) - Ambiguous pronunciations (multiple codes considered, any overlap counts as a match)
Phonetic coding is language-aware but not perfect; some name pairs that native speakers hear as identical won't match phonetically, and rarely the reverse.
When to turn it on
Always on for:
- Person profile with name columns
- Healthcare provider matching (last names often have spelling variation)
- Genealogy or research matching
Optionally on for:
- Company profile if company names include surnames (
Smith Manufacturing,Jones & Sons) - Custom profile matching on text fields that represent names
When to turn it off
Off for:
- Company profile with corporate-style names (
Acme Inc,Global Solutions LLC) - Product / SKU profile (product names don't have phonetic variation worth matching on)
- Identifier profile (identifiers are exact by definition)
- Custom profiles matching on non-name fields
Phonetic matching on non-name data is usually harmful — you'll get false positives from unrelated records that happen to code similarly.
What phonetic matching does to scores
When two names phonetically match:
- If the spelling also matches (exact or high fuzzy similarity), phonetic is irrelevant — the score stays high.
- If the spelling differs but the phonetic code matches, the score floor is raised — a pair that would have scored 55 on pure string similarity might score 75, pushing it out of
unmatchedintoreviewormatch.
The specific score boost from a phonetic match is profile-dependent. For the Person profile, phonetic agreement meaningfully lifts the name subscore.
Blocking and phonetic codes
Phonetic matching also affects blocking. If phonetic is enabled and your blocking key includes phonetic-of-last-name, then records with the same phonetic code end up in the same block, getting compared against each other. Without phonetic, they might sit in different blocks and never be scored.
Enabling phonetic matching therefore increases both the number of candidate pairs compared and the score each gets — which is why it's such a strong lever.
Configuration
From the Configure step:
Phonetic matchingtoggleDefault: Profile-dependentMaster toggle. Enable or disable for the entire match.
In custom profiles:
Phonetic per fieldboolean per columnEnable phonetic for specific name fields only. Useful when you want phonetic on last name but not first name.
Phonetic boost weight0–30Default: 20How much the phonetic agreement contributes to the field subscore. Lower if you're getting too many phonetic false positives.
Common pitfalls
Too-aggressive phonetic matching
Turning phonetic on in a Company profile where names are Alpha Corp and Beta Corp will cause them to phonetically match on the Corp portion. Turn off phonetic for company names.
Phonetic on non-Latin scripts
Phonetic coding is designed for Latin-script input. For Chinese, Japanese, or Arabic names, phonetic matching falls back to transliterated-name comparison — it works but with reduced accuracy. For non-Latin primary script data, consider matching on native script first with phonetic off, then a separate pass on transliterated data with phonetic on.
Cross-cultural phonetic mismatches
A Japanese name transliterated as Yamashita and the same name transliterated as Yamaxita may not produce the same phonetic code. For cross-cultural name matching, consider using handling international names strategies in addition to phonetic.
Related reading
- Handling nicknames and abbreviations — the other name-normalization lever
- Handling international names — when phonetic isn't enough
- Match profiles — profile-level phonetic defaults
