Picture a whiteboard covered in dots and lines instead of a spreadsheet with names in rows. Each dot is one identifier a business or vendor happened to observe: a cookie value, a mobile ad ID, a hashed email, an IP address, a loyalty card number. Each line is a claim that two dots probably belong to the same real person or household. That whiteboard is what an identity graph actually is. Marketing language tends to describe it as "knowing your customer across every touchpoint," but structurally it is a graph of guesses, some far more confident than others.
What's actually stored: nodes and edges, not identities
An identity graph does not contain a field that says "this is Jane Smith." It contains nodes, the raw identifiers, and edges, the claimed links between them, each carrying a confidence weight. A person who logs into a service, then later browses anonymously on the same laptop, then opens an email on their phone, generates three separate identifiers. The graph's entire job is deciding whether to draw edges connecting those three dots into one cluster. Vendors differ enormously in how aggressively they draw those edges, and that aggressiveness, not the size of the database, is what actually separates a conservative identity graph from an inflated one.
Deterministic edges vs probabilistic edges
Edges come in two flavors and conflating them is where most confusion starts. A deterministic edge is built from something unambiguous: a visitor logs into the same account on two devices, so the graph links both devices with near-certainty. A probabilistic edge is built from a statistical pattern instead: two devices tend to appear on the same home wifi network at similar hours, so the graph guesses they belong to the same household, without any login ever confirming it. Deterministic edges are rare and reliable. Probabilistic edges are common and make up the bulk of most commercial graphs, because logins are the exception, not the rule, in anonymous browsing.
The hop problem: why confidence should not stay constant
Most explanations of identity graphs skip this part entirely. A graph rarely links two identifiers directly. More often it chains them: cookie A links to device B through a probabilistic match, device B links to email C through a separate probabilistic match, and the graph reports that cookie A belongs to the same person as email C, several hops removed from any direct observation. Each hop carries its own error rate, and those error rates compound. A chain of three edges, each individually 90 percent likely to be correct, does not produce a 90 percent likely conclusion; it produces something meaningfully weaker, because all three have to hold at once. Graphs that report a single flat confidence score regardless of how many hops produced a match are hiding this compounding. It's the one question worth asking any identity resolution vendor: how many hops sit between the identifier you started with and the identity you were handed back?
Identity graph vs CRM vs identity resolution
These three terms sit next to each other in vendor pitches and get blurred together, but they are structurally different things.
| CRM | Identity graph | Identity resolution | |
|---|---|---|---|
| What it holds | Known, opted-in contacts a business collected directly | A network of identifiers, both known and anonymous, with claimed links between them | The process of running that graph to output a single "who is this" answer |
| Source of entries | Signups, purchases, form fills, first-party consent | First-party data plus, often, third-party broker data and cookie syncing | Not a data store; a matching operation performed on top of a graph |
| Confidence | High. Every record was volunteered | Variable, edge by edge and hop by hop | Inherits whatever confidence the underlying graph produced |
| Typical size | Bounded by actual customers and leads | Can run to hundreds of millions of anonymous identifiers | N/A, it is an action, not a store |
A CRM is a ledger of people a business actually has a relationship with. An identity graph is a much larger, much less certain structure that tries to connect anonymous activity back to that ledger, or to build a picture of someone the business has never directly heard from at all. Identity resolution is simply the query that walks the graph and returns an answer. Businesses that treat identity graph output with CRM-level confidence are making the mistake this whole page exists to prevent.
Identity graph marketing: what it's actually used for
The commercial use cases cluster into a few patterns. Cross-device advertising uses a graph to recognize that the phone and the laptop that just converted are probably the same household, so a campaign does not keep bidding on someone who already bought. Frequency capping relies on the same clustering to avoid showing one person the same ad forty times across devices. B2B enrichment tools use a version of this idea to take an anonymous website visitor's IP address and walk a graph back to a company record. In every case, the graph is doing the same underlying job: taking a fragment of anonymous signal and betting on which known cluster it belongs to.
Where Raydar sits relative to this
Raydar does not run an identity graph and does not sell access to one. Its visitor identification is narrower and more literal: a hashed version of the click's IP address, tagged with an explicit high-confidence or low-confidence label, with no chaining across cookies, devices, or third-party broker files. That is a deliberately smaller claim than what a full identity graph vendor sells, and it is worth understanding the difference described here before evaluating either kind of tool, including how de-anonymization generally works and what a match rate figure is actually measuring. For businesses trying to identify visitors by company rather than by individual, the mechanism differs again, covered in person-level vs company-level identification.