Skip to content
Infinite AudienceInfinite Audience Docs

How matching works

Customer data is rarely clean. People use nicknames, mistype their address, change their last name, move, and fill out forms from work. Instead of requiring a perfect text match, the engine looks for agreement across every detail you send, using three approaches in order.

Your recordNames, addresses, emails, phones, DOB, or a person ID
DeterministicExact, verified connections in the graphProbabilisticWeighs every detail when data is messySpatialReal-world geography and commuting distance
Resolved personPerson ID, match level, confidence, and enrichment

Each step runs only for records that earlier steps have not yet matched to a specific person.

Deterministic matching

Section titled “Deterministic matching”

Deterministic matching finds an exact, verified connection in the graph. It’s fast and very precise, and it handles the everyday differences that defeat simple text comparison:

  • Nicknames such as Bill for William or Barb for Barbara, and two-word first names such as Mary Beth.
  • Last-name changes, confirmed by first name, date of birth, and postal address.
  • An Infinite Audience Graph Person ID you already have, which is recognized immediately with no rematching.
I · IndividualNickname at a known address
You sent
  • Bill Gallagher
  • 104 Main St
  • Springfield, IL 62704
Matched in the graph
  • William Gallagher
  • 104 Main St
  • Springfield, IL 62704

How: Bill is a recognized nickname for William, and the address agrees exactly.

I · IndividualName and email, no address
You sent
Matched in the graph

How: The email is connected to a Michael Chen in the graph, so no postal address is needed.

I · IndividualNo street address at all
You sent
  • Sarah Jenkins
  • ZIP 90210
  • Born 1988-04-12
Matched in the graph
  • Sarah Jenkins
  • 311 N Canon Dr, Beverly Hills, CA 90210
  • Born 1988-04-12

How: Name, ZIP code, and date of birth together identify one person, even without a street.

I · IndividualLast name changed after marriage
You sent
  • Amanda Taylor
  • 42 Elm St, Columbus, OH 43215
  • Born 1992-06-15
Matched in the graph
  • Amanda Miller
  • 42 Elm St, Columbus, OH 43215
  • Born 1992-06-15

How: First name, date of birth, and address all agree. The different last name doesn't block the match.

Probabilistic matching

Section titled “Probabilistic matching”

When there is no exact connection, probabilistic matching weighs every detail together. Strong agreement on some details can outweigh small differences in others, such as a typo, a road alias, a different unit format, or a neighboring ZIP code.

Some differences are never outweighed. A candidate is ruled out if the house numbers differ, the unit numbers conflict, the first names cannot be the same person, or the birth dates are more than 18 years apart.

I · IndividualSame road, different name
You sent
  • Elena Vance
  • 12558 NC-42
  • Clayton, NC 27520
Matched in the graph
  • Elena Vance
  • 12558 Highway 42
  • Clayton, NC 27520

How: State route and highway names are treated as the same road when the house number, ZIP code, and name agree.

I · IndividualTypo plus a nickname
You sent
  • Mikey Jehnson
  • 2210 Oakmont Dr
  • Austin, TX 78704
Matched in the graph
  • Michael Johnson
  • 2210 Oakmont Dr
  • Austin, TX 78704

How: House number, street, and ZIP code agree exactly. Mikey is a nickname for Michael, and Jehnson is a single-letter phonetic variation. Together the details clearly point to one person.

I · IndividualHyphenated last name and a different unit label
You sent
  • Sarah Jenkins-Smith
  • 500 Harbor Blvd, Apt 4B
  • San Diego, CA 92101
Matched in the graph
  • Sarah Jenkins
  • 500 Harbor Blvd, Suite 4B
  • San Diego, CA 92101

How: Unit labels such as Apt, Suite, Unit, and # are normalized, so both records point to unit 4B. The hyphenated last name contains the graph surname.

I · IndividualWrong or neighboring ZIP code
You sent
  • Robert Alvarez
  • 1418 Ponce De Leon Ave
  • Atlanta, GA 30307
Matched in the graph
  • Robert Alvarez
  • 1418 Ponce De Leon Ave
  • Atlanta, GA 30306

How: Same name, house number, and street in a neighboring ZIP code in the same area. This is common when a ZIP code is mistyped or an address sits near a ZIP boundary.

Spatial matching

Section titled “Spatial matching”

If neither approach can place a record and you sent a first and last name, the spatial engine looks at where the address actually is. It evaluates physical proximity and density to locate the same person within a realistic distance. This is how it finds someone who moved between buildings in an apartment community or who submitted an office address.

Explore the spatial engine, its modes, and examples →

How date of birth helps

Section titled “How date of birth helps”

Date of birth is optional, but when you have it, it strengthens matching in every approach:

Tells relatives apart

A father and son with the same name at the same address are separated by their birth dates.

Unlocks more matches

Name + ZIP + date of birth needs no street. First name + date of birth + address works after a last-name change.

Blocks wrong generations

An exact birth date raises confidence. Birth dates more than 18 years apart are never accepted as the same person.

If you don’t collect date of birth, nothing is held against the record. The engine matches on your other signals.

Match levels

Section titled “Match levels”

Every match reports a match_level. Choose which levels you accept for each request or file; all five are included by default.

Level What it confirms Typical uses
I · Individual The specific person, confirmed by name plus address, email, phone, or date of birth. Personalized outreach, direct mail, CRM cleanup and deduplication.
H · Household The home and family, confirmed by last name plus address. Household offers, direct mail, home services.
D · Digital A known email or phone, without a confirmed name. Digital advertising audiences, newsletter and web enrichment.
S · Spatial The person, placed by the spatial engine near the address you sent. Workplace leads, recent local movers, apartment communities.
A · Address The physical address only. Property and neighborhood insight, territory planning.

If a record qualifies at more than one level, the most specific result wins: an Infinite Audience Graph Person ID you sent, then I, D, H, S, and A. A spatial match can replace an address-only match because it identifies a person, not just a place.

H · HouseholdSame family, different first name
You sent
  • Linda Morrison
  • 742 Evergreen Terrace
  • Portland, OR 97214
Matched in the graph
  • Morrison household
  • 742 Evergreen Terrace
  • Members: Thomas and Karen Morrison

How: Last name and address agree, but no Linda lives there in the graph, so the result is the household rather than an individual person.

D · DigitalEmail only
You sent
Matched in the graph
  • A known person behind this email

How: The email is connected to one person in the graph. No name was sent to confirm, so the match is Digital.

A · AddressAddress only
You sent
  • 1847 Willow Creek Rd
  • Boise, ID 83702
Matched in the graph
  • 1847 Willow Creek Rd
  • Boise, ID 83702
  • Property and neighborhood attributes

How: With no name, email, or phone, the result describes the place, not a particular resident.

Match types

Section titled “Match types”

While match_level gives the category (I, H, D, S, or A), every returned match also includes a specific match_type. This records the exact combination of signals that resolved the record:

Level match_type What made the match
Prior level or empty graph_iag_person_id_match Recognized Person ID sent directly; preserves prior match provenance without establishing a new Individual match. Person IDs →
I · Individual graph_name_address_match First name + last name + street address + zip
I · Individual graph_name_email_match First name + last name + email
I · Individual graph_name_phone_match First name + last name + phone
I · Individual graph_name_zip_dob_match First name + last name + zip + date of birth (no street needed)
I · Individual graph_first_dob_address_match First name + date of birth + address (resolves across last-name changes)
I · Individual graph_fuzzy_street_match Name + street with highway/route alias or minor spelling difference
I · Individual probabilistic_name_address_match Weighted agreement across name, address, unit, or neighboring zip
H · Household graph_last_name_address_match Last name + street address + zip (different first name or family unit)
D · Digital graph_email_match Email alone
D · Digital graph_phone_match Phone number alone
S · Spatial spatial_match Name + physical coordinate proximity within residential or commuter radius
A · Address graph_address_match Street address + zip alone (property and neighborhood profile)

Confidence scores

Section titled “Confidence scores”

Every match includes a match_confidence score between 0 and 1. Matches confirmed by several agreeing details score near 1.00. Matches with less corroboration score lower.

By default, a match must score at least 0.80 to be returned. Below that, the record is left unmatched rather than guessed. Unmatched rows stay in your results so nothing disappears.

Go deeper

Section titled “Go deeper”