What Is an Identity Graph? How It Works, How to Build One and How to Choose One

An identity graph is a map of every identifier that belongs to the same person: their emails, phone numbers, customer IDs, browser cookies, device IDs and ad click IDs, joined by the evidence that ties them together. Each identifier is a node, each piece of evidence is an edge, and each connected cluster is one person. It is the data structure that identity resolution builds and reads from.

When a shopper browses on their phone, adds to cart on a laptop and buys with an email a week later, your store collects a pile of identifiers that look like strangers. The identity graph is what turns them back into one customer, so your ad platforms, your email platform and your reports all agree on who bought and what brought them in.

What is an identity graph?

An identity graph, sometimes called an ID graph, stores people as networks of identifiers rather than as rows in a table. That shape matters because a shopper does not arrive with one ID. They arrive with whatever the current browser and device can offer, and they add more each time they log in, sign up, click an email or buy.

Nodes: the identifiers

A node is any value that points at a person or a device. For an online store, the common ones are:

Edges: the evidence

An edge records why two nodes belong together. Some edges are certain: an email entered at checkout on a browser links that email to that browser's cookie. Others are inferred: two browsers that share an IP address and similar settings might be the same person. A good graph stores the type of evidence and how confident it is, not just the fact that a link exists.

Clusters: the people

Follow every edge from a node and you reach every identifier that belongs to the same person. That connected set is the person. When a new edge joins two clusters, for example the same email showing up on a phone and a laptop, the graph merges them into one customer.

---

How an identity graph works: one shopper, step by step

Here is how a graph grows for a single shopper over a week.

By Saturday the graph holds one person with two browsers, one email and the Meta click ID from Monday. The purchase can now be credited to the Monday ad, and the Tuesday cart belongs to a known customer instead of an anonymous one. Without the graph, the same week looks like two visitors and an unattributed order.

---

How to build an identity graph

Built in-house or bought, a graph is only as useful as these steps make it.

1. Decide which identifiers you collect. Emails and phone numbers from checkout and signups, your customer IDs, a first-party browser ID, and every ad click ID that reaches your site. Anything you do not capture cannot become a node.
2. Normalize before you link. The same email typed two ways must become one node. Lowercase and trim emails, and store phone numbers with a country code. Meta's customer information parameters spell out the format it expects for each field, and following them also makes your own matching cleaner.
3. Link deterministically first. An exact shared identifier, such as the same email on two browsers, is the strongest edge you can have. Build the core of every cluster from these.
4. Add probabilistic links with limits. Signals such as IP address and device characteristics can link sessions that never shared an email, but they can also join two people on the same home network. Store a confidence level with each one, and decide which uses are allowed to rely on them.
5. Write merge and split rules. Decide what happens when one email appears on dozens of devices (a shared family inbox, or a store employee testing checkout), and cap cluster size so one bad link cannot swallow hundreds of customers.
6. Plan for persistence. Browsers limit how long your identifiers survive. Safari, for example, caps cookies set by JavaScript at seven days. The graph should keep the links after a cookie expires, so a returning shopper who logs in or clicks an email reconnects to their full history.
7. Handle consent and deletion. Record the consent state for each identifier, honour opt-outs, and make sure a deletion request removes the person's nodes and edges, not just one profile.
8. Deliver the resolved identity. A graph only pays off when its answer reaches the tools that act on it: your ad platforms' conversion APIs, your email platform and your attribution reports.

Building this in-house means owning identity storage, matching logic and every destination integration. Most ecommerce teams buy it as part of a tracking or identity platform instead.

---

Types of identity graphs

Identity graphs differ in how they link, whose data they hold and who they describe.

Deterministic vs probabilistic identity graphs

A deterministic identity graph only links identifiers that were seen together, such as an email entered on a known browser. A probabilistic graph also links identifiers that are likely to belong together based on signals like IP address and device settings. Most production graphs mix the two and treat the probabilistic links as lower-confidence. Our identity resolution guide compares the two methods in detail.

First-party (private) vs third-party identity graphs

A first-party or private identity graph is built from identifiers your own store collected. It belongs to you and describes your own visitors and customers. A third-party identity graph is assembled by a data provider from many sources and covers people who may never have visited your site. Third-party graphs can identify more people, but you depend on the provider's sources and consent basis, and cross-site cookies that some of them relied on are now blocked by default in Safari (WebKit, 2020).

Consumer vs B2B identity graphs

A consumer identity graph resolves individual shoppers and households, which is what an online store needs. A B2B identity graph resolves people to the companies they work for, often from business email domains and company IP ranges. B2B visitor identification tools are built on the second kind, which is why they tell you which company visited rather than which customer.

---

Identity graph vs CDP vs device graph

These three terms overlap, and vendors use them loosely.

The CDP Institute defines a CDP as software that gathers customer data from every system a company uses, links each piece to the right person and keeps a persistent profile for each customer. The linking step inside a CDP is an identity graph. A device graph is a narrower graph that only covers devices and browsers. An identity graph for ecommerce needs both device links and contact identifiers, because the email is usually what proves two devices belong to one shopper.

What ecommerce brands use an identity graph for

An identity graph is useful only where its answer changes what a tool does. For an online store, that is four places.

• Ad platform matching. Conversion APIs match each event to a user from the identifiers you send. Meta's Conversions API accepts hashed emails and phone numbers as lists, plus your own external_id, the fbc click ID and the fbp browser ID (Meta for Developers). A graph lets you attach all of them to every event, including the anonymous page views and add-to-carts that happened before the shopper identified themselves. Our Event Match Quality guide covers how Meta scores those identifiers.
• Email and SMS flows. Abandoned cart and browse flows only fire for shoppers your email platform can recognize. A graph recognizes a returning subscriber on a new device and tells the platform who they are. We cover the Klaviyo side in why identity resolution is the missing link in your Klaviyo performance.
• Attribution. Cross-device journeys only make sense when the phone that clicked the ad and the laptop that bought are the same person.
• Customer counts. New-customer metrics depend on knowing that a checkout email belongs to someone who bought before under a different device or guest account.

---

How to evaluate an identity graph provider

When you compare providers, ask the same questions of each.

---

Upstack's identity graph

Upstack Signal resolves every visitor against your own first-party data and Upstack's internal device network and identity graph, which covers more than 750 million devices and 150 million customers. It links identifiers with both deterministic and probabilistic matching, and an Upstack ID persists for the life of the customer rather than expiring with a browser cookie, so a shopper who returns long after their last visit is still recognized.

Signal sends the resolved identity with every event to Meta's Conversions API and the other ad, email and analytics destinations you connect, so each platform sees the same customer. Finding shoppers who have never given you a contact detail is a separate job, handled by Upstack Audience with third-party data.

---

Key takeaways

• An identity graph stores each person as a cluster of identifiers (nodes) joined by evidence (edges).
• Build the core from deterministic links, add probabilistic links with confidence limits, and write merge and split rules before you need them.
• Persistence matters: the graph should outlive the cookie that first created a node.
• A CDP keeps customer profiles; the identity graph is the part that decides which data belongs to which profile.
• The graph only pays off when its answer reaches your ad platforms, email platform and reports.

---

Frequently asked questions

What is an identity graph?

An identity graph is a database that links all the identifiers belonging to one person, such as emails, phone numbers, customer IDs, cookies, device IDs and ad click IDs. Each identifier is a node, each piece of evidence linking two identifiers is an edge, and each connected cluster represents one person.

What is an ID graph in marketing?

In marketing, an ID graph is the identity graph a brand or vendor uses to recognize the same customer across devices, browsers and channels. It lets ad platforms match conversions to the right user, lets email platforms trigger flows for returning shoppers, and lets attribution connect an ad click on one device to a purchase on another.

How do you build an identity graph?

Collect every identifier your store sees, normalize them, link them deterministically wherever an exact identifier is shared, add probabilistic links with confidence limits, set merge and split rules, keep links after cookies expire, record consent, and send the resolved identity to your ad, email and reporting tools.

What is the difference between a deterministic and a probabilistic identity graph?

A deterministic identity graph links identifiers only when they were seen together, such as an email entered on a known browser. A probabilistic identity graph also links identifiers that are statistically likely to belong together, based on signals such as IP address and device settings. Deterministic links are more accurate; probabilistic links reach more sessions with lower confidence.

Identity graph vs CDP: what's the difference?

A CDP is the software that collects customer data and keeps a profile for each customer. An identity graph is the set of links that decides which identifiers and events belong to which customer. Most CDPs contain an identity graph, but an identity graph can also run inside a tracking platform without a CDP.

What is a private identity graph?

A private, or first-party, identity graph is built only from identifiers your own business collected, such as checkout emails, logins and your site's cookies. It belongs to you, unlike a third-party graph that a data provider assembles from many sources.