How to Evaluate an Identity Graph for Healthcare Media

Graphs are hard to evaluate because vendors describe them in large numbers: billions of devices, hundreds of millions of people. Size is not quality. Evaluate four separate properties.

1. Coverage

Coverage is how much of your intended audience the graph can find. Ask for coverage against your audience, not overall. A graph can have excellent national coverage and poor coverage of older adults, rural areas, or specific specialties.

Evidence to request: match rate of a sample of your audience, broken down by key segments, with deterministic and probabilistic shares shown separately.

2. Correctness

Correctness is whether the links are right. A graph that links a person's email to a neighbor's TV is wrong, even if the match rate looks good.

Evidence to request: validation against a truth set, such as a panel or first-party logged-in data, with precision and recall reported. Ask how often false links are found and removed.

3. Freshness

People change devices, move, and switch providers. A link that was right a year ago may be wrong now.

Evidence to request: how often links are refreshed, how long unverified links are kept, and what share of links have been observed recently. See healthcare audience data freshness.

4. Fitness for use

A graph that is fine for broad awareness may not be fine for measurement, where false links bias results. It may be fine for households but not for individuals.

Evidence to request: documentation of intended uses and known limitations.

The evidence sheet

Filling this in forces the conversation away from headline numbers.

Privacy considerations

Identity graphs involve personal data and, in health contexts, can reveal sensitive information by association. Review how the vendor handles opt-outs, deletion requests, and sensitive data categories. The NIST Privacy Framework is a useful structure for this review.

Graphs and measurement

In outcome studies, a false link can count someone as exposed who was not, or miss someone who was. Both weaken the study. When a measurement partner uses a different graph than the activation partner, exposures can be lost between them. This is one reason Rx match rates drop.

Practical takeaway

Ask every identity vendor to complete the evidence sheet for your audience specifically. If they cannot provide validation evidence for correctness, treat their match rates as upper bounds.

Editorial note. Analysis and frameworks are the author's own and do not represent Acxiom or any current or former employer, client, or named platform. Examples labeled hypothetical or illustrative are not results from real campaigns. Nothing here is legal, regulatory, or medical advice.