Identity Resolution
Identity resolution is the process of matching records from different sources and devices to determine which ones belong to the same person, then merging them into one profile. It compares identifiers such as hashed emails, phone numbers, account IDs, and device IDs across every system a company runs. The output is a single customer view that marketing, analytics, and service tools can act on.
How Identity Resolution Works
Identity resolution evaluates every incoming record against what the system already knows, then decides whether it is a new person or an existing one.
- Collect identifiers. Each record arrives with whatever it has: a hashed email from a checkout, a device ID from a web session, a loyalty number from a store purchase.
- Compare against the graph. The system checks those identifiers against the existing set of known connections.
- Merge, create, or link. A match attaches the record to an existing profile. No match creates a new one. A partial match may link two previously separate profiles.
- Assign a persistent ID. Each resolved person gets a synthetic identifier that stays stable across runs, changing only when new evidence merges two identities.
Step 4 is what makes the output usable. Without a stable ID, every run reshuffles who is who, and yesterday’s audiences are invalid today.
The Identity Graph
An identity graph is the data structure that stores the connections, and it is distinct from the profile built on top of it. The CDP Institute describes it as “the connection layer beneath your customer profile, not the profile itself.”
It is a graph in the literal sense: identifiers are nodes, and confirmed or inferred connections are edges. One person might be a cluster holding two email addresses, a phone number, three device IDs, and a loyalty account, joined by edges carrying different confidence levels.
Graphs come in two ownership models:
- First-party identity graph: built from a company’s own records in its own infrastructure. The brand controls the matching logic and owns the output, but reach stops at people it already has data on.
- Third-party identity graph: built by an external provider aggregating identifiers across many companies, then licensed back. Reach extends past your own customers, but you own neither the logic nor the data.
First-party graphs have gained ground as third-party signal degrades under browser and privacy restrictions.
Matching Methods
Three methods decide whether two records belong to one person, and most systems run all three.
- Deterministic matching joins records on an identifier known to be the same person, such as a hashed email or a logged-in account ID. High confidence, limited recall.
- Probabilistic matching infers a match from patterns in IP address, device attributes, and behavior, usually with a confidence score attached. Wider reach, and it can be wrong.
- Transitive matching connects records through inference chains. If A matches B and B matches C, the graph infers that A and C are the same person, even with no identifier in common.
Transitive matching supplies most of the reach and most of the damage, which is why confidence thresholds matter more than the choice between the first two methods. The deterministic and probabilistic trade-off is covered in full under cross-device tracking; device fingerprinting covers the probabilistic signal stack.
Survivorship Rules
Survivorship rules decide which value wins when merged records disagree, and the rule set that makes that call produces what is commonly called the golden record. Three types cover most cases:
- Recency: take the most recently updated value. Standard for addresses and phones.
- Frequency: take the value appearing most often across sources. Useful for name spellings.
- Source priority: take the value from the most trusted system. A billing platform beats a newsletter signup form.
Rules are set per field, not per profile: the winning email can come from one system and the winning address from another.
Measuring Match Rate
Match rate is the percentage of input records that resolve to a known identity, and it is the main number used to judge a setup. Upload 100,000 emails, match 62,000, and the rate is 62%. Reported person-level ranges vary sharply by market:
| Context | Reported person-level match rate |
|---|---|
| US consumer traffic, weak setup | 25% to 35% |
| US consumer traffic, strong setup | 50% to 60% |
| US consumer traffic, ceiling | 70% to 73% |
| Canada, UK, Australia | 15% to 35% |
| EU under GDPR consent | Often single digits |
Two cautions apply to any vendor figure. Match rate multiplies by accuracy, so a claimed 70% rate delivered probabilistically at 80% accuracy is roughly 50% correctly resolved records. And demo environments typically show rates three to five times higher than production, so validate on your own file before signing anything.
Where Identity Resolution Breaks
The expensive failure is the false merge: two different people combined into one profile with enough confidence that downstream systems treat the result as fact. A wrongly merged profile can surface one person’s order history to another.
- Thresholds set too low. Loose confidence settings collapse unrelated profiles, and transitive chains spread the error outward from a single bad link.
- Shared devices. A family tablet feeds one node several people’s behavior.
- Fragmented campaign data. Profiles inherit the source and medium on each touchpoint, so three spellings of one channel produce a journey that looks like three channels. Enforcing values with UTM rules keeps that history coherent before it reaches a profile.
False merges are hard to reverse. Once a merged ID reaches ad platforms and email tools, unwinding it means rebuilding every audience downstream.
Frequently Asked Questions
What is identity resolution?
Identity resolution is the process of matching records across sources and devices to work out which belong to the same person, then unifying them into one profile. It compares identifiers such as hashed emails, account IDs, and device IDs, then assigns each resolved person a persistent ID.
What is an identity graph?
An identity graph is the data structure storing which identifiers belong to which person and how confident the system is in each link. Identifiers are nodes, connections are edges. It sits beneath the customer profile rather than being the profile, and it is either first-party (owned in-house) or third-party (licensed from a provider).
What is identity resolution in marketing?
It is what makes cross-channel measurement and personalization possible, since both require knowing that a phone visitor and a desktop purchaser are one customer. It powers suppression lists, lifecycle campaigns, accurate customer counts, and offline-to-online attribution. It usually runs inside a CDP.
What is customer identity stitching?
Customer identity stitching is another name for the merging step inside identity resolution, used most often when the records joined are sessions or devices rather than full customer records. The term is common in analytics, where GA4’s User-ID stitches logged-in sessions across devices.
What is a good identity resolution match rate?
For US consumer web traffic, 50% to 60% person-level resolution indicates a strong setup, while 25% to 35% is closer to a floor. Canada, the UK, and Australia run 15% to 35%, and EU rates are often single digits under GDPR consent requirements. Judge any number by accuracy as well as reach.
Keep the campaign data feeding your profiles consistent by building tagged links with the free UTM builder at linkutm.