Identity Graph

The dual-engine core of identity resolution. Deterministic matching links records using absolute, verified common identifiers (e.g., a hashed email or account ID). Probabilistic matching uses statistical models to infer identity links based on patterns in non-unique attributes like IP addresses, device types, and browsing times. A robust graph orchestrates both, applying deterministic logic first for precision, then probabilistic models to expand reach, often using a confidence threshold to control match quality.

The ability to map an identity over time, handling state changes gracefully. This includes:

• Merging: Combining two profiles when a new link is discovered (e.g., a user logs in on a previously anonymous device).
• Splitting: Separating profiles if a shared device (like a family tablet) is later attributed to distinct individuals.
• Decay: Downgrading the confidence of old probabilistic links that haven't been reinforced by recent activity.

A foundational architectural requirement, not an afterthought. The graph must enforce consent boundaries, ensuring that identifiers from users who have opted out of tracking are not linked. This involves integrating with a Consent Management Platform (CMP) to ingest user preference signals and applying them as strict rules within the matching logic. The graph must also support data subject access requests (DSARs) by quickly retrieving all data associated with a single identity.

The graph's value is realized through a low-latency API. When a user visits a site, the system must ingest a live stream of identifiers (cookie, device fingerprint, login event) and return the resolved persistent ID in milliseconds. This allows a Decisioning Engine to immediately personalize the experience for a known or newly recognized user, rather than treating them as anonymous.

Unlike relational databases, identity graphs are best stored in a native graph database. This model treats identifiers as nodes and the relationships between them as edges, with properties like match type and confidence score stored on the edge. This structure is exponentially more efficient for traversing complex, multi-hop identity chains (e.g., Device A -> Email B -> Account C) than performing recursive SQL joins.

A probabilistic device identification technique that creates a unique hash from dozens of passive browser attributes (OS, fonts, screen resolution, WebGL). It serves as a persistent identifier when cookies are blocked.

• Use Case: Anonymous visitor stitching before authentication.
• Risk: High entropy fingerprints can be unstable; minor browser updates break the ID.
• Privacy: Increasingly restricted by browsers (e.g., Safari's ITP) and regulations.

The temporal grouping logic that defines the boundaries of a visit. It transforms raw event streams into coherent sessions, linking pageviews to a specific anonymous or known profile within the graph.

• Standard: A 30-minute inactivity timeout defines a session end.
• Complexity: Midnight rollovers and campaign attribution windows require custom rules.
• Impact: Incorrect sessionization corrupts frequency counts and behavioral sequences.

The primary fuel for the identity graph. This is information collected directly with user consent through owned channels—website interactions, CRM records, and purchase history.

• Advantage: Highest accuracy and compliance durability.
• Contrast: Zero-party data (explicit preferences) vs. third-party data (aggregated, often deprecated).
• Graph Role: Provides the deterministic anchors (email, phone) for identity resolution.

The technical framework governing the legal perimeter of the identity graph. It captures, stores, and signals user consent preferences (opt-in/opt-out) to ensure compliance with GDPR and CCPA.

• Integration: Must feed suppression signals into the graph to prevent stitching of opted-out users.
• IAB TCF: Standard protocol for communicating consent strings in programmatic advertising.
• Risk: Non-compliance breaks the graph's legality and exposes the organization to fines.