A data structure that maps every known identifier for a person into one connected record. It links email addresses, device IDs, phone numbers, cookies, and account IDs so downstream systems can recognize one customer across channels.
An identity graph is a data structure. Not a dashboard, not a profile, not a database table. It maps relationships between identifiers: this email address is connected to this device ID, which is connected to this phone number, which appeared alongside this loyalty card number. The graph does not store behavioral data or preferences. It answers one question: which identifiers belong to the same person?
How it works
When a new data point arrives (a form submission, a login event, an app install), the graph attempts to connect it to an existing node. If the new email address matches one already in the graph, the associated device ID gets linked. Over time, the graph grows as more identifiers cluster around known individuals.
Most graphs use a combination of deterministic matching (exact identifier overlap) and probabilistic matching (statistical inference from patterns like IP address, browser fingerprint, and behavioral signals). The blend determines both coverage and accuracy.
Scale vs. accuracy
The biggest misconception is that a bigger graph is a better graph. Scale without accuracy produces false connections where two different people get merged into one node. That creates worse outcomes than having no graph at all: wrong personalization, bad attribution, wasted spend.
Graph hygiene matters more than graph size. Stale identifiers like a device ID from 3 years ago or a cookie that has been cleared create noise. The best implementations expire old connections and require periodic re-verification through authenticated events.