What is an identity graph?
An identity graph is a database that links scattered digital identifiers, such as cookies, device IDs, IP addresses, and hashed emails, to the single person or account they belong to, with a confidence score on each connection.
Most descriptions of an identity graph stop at the matching problem: stitch together cookies(opens in a new tab), device IDs, and IP ranges until enough overlap points to the same person. That is the mechanics. The outcome that matters is whether a rep can act on the connection once it is made. A graph can be technically thorough and still useless in practice if most of its connections sit at low confidence(opens in a new tab), because a list of maybes is not a list sales should call. A graph that earns its place in a go-to-market stack links first-party(opens in a new tab) engagement signals, pricing page views, return visits, ad clicks, to a named contact and account, and it scores how confident that link is before the name reaches a rep's queue.
How an identity graph gets built
An identity graph starts as a set of nodes: a cookie ID, a hashed email, a device fingerprint, an IP address, a LinkedIn profile. Each node on its own says almost nothing. The graph's job is to draw edges between nodes that plausibly belong to the same person or account, then attach a confidence score to each edge. A visitor who lands on a pricing page(opens in a new tab) from a known IP range, then returns a week later from a resolved email address, generates two nodes and one edge. Enough high-confidence edges, and the graph resolves an anonymous browsing session into a named contact at a named account. A form fill(opens in a new tab) or a logged-in session adds a deterministic node, the kind that anchors the rest.
Why edge confidence matters more than coverage
It is tempting to judge an identity graph by size: how many nodes, how many matched profiles, how wide the coverage. A graph with thousands of low-confidence edges moves the guessing downstream to sales, who then have to decide whether a name at a mid-size company is worth a call. A smaller graph that only surfaces edges above a real confidence threshold gives reps fewer names, and every one of them is a name they can act on without checking first. Fewer trusted matches beat a wide net full of maybes.
The same logic explains what happens when a graph gets an edge wrong. The wrong person gets attached to a signal, and a rep ends up calling someone who never looked at the product. A graph that flags uncertain edges instead of forcing every signal into a match keeps those false positives out of the queue.
Identity graph versus data warehouse
A data warehouse stores raw records: tables of page views, form fills, and ad clicks, with no opinion about how they relate. An identity graph sits on top and draws the connections between those records, with a confidence score on each one, so raw rows become a resolved person or account. Third-party cookies are one possible input, and a shrinking one. Graphs built on first-party engagement signals, resolved emails, and device data hold up as browsers block third-party cookies, and those signals tend to produce higher-confidence edges because they come from the visitor's own behavior on your properties.
Account-level versus contact-level identity graphs
An account-level identity graph resolves traffic down to a company: an IP range maps to a firmographic record, and the trail ends there. A contact-level(opens in a new tab) graph attaches a named person to that account, tying the pricing page visit to a buyer on the committee rather than to the logo alone. A useful contact-level graph proves itself when the confidence scoring behind it holds up once a rep reaches out.
Where the identity graph fits in an ABM program
An identity graph rarely stands alone. It feeds the Find stage of an ABM program, where accounts and the buyers inside them are picked for attention; the Activate stage, where named contacts go into an ad audience or a sales handoff; and the Prove stage, where the team looks back at which touches influenced a deal. When the graph underneath is unreliable, the account gets staged on bad data, the ad audience reaches the wrong person, and the proof at the end does not hold up. Everything downstream depends on this layer being right.