Identity Resolution

The process of linking identifiers based on an exact, verified match key, such as a hashed email address, phone number, or account login. This method provides absolute certainty of identity linkage.

• Match Keys: Common keys include user_id, email_sha256, or loyalty_card_number.
• Zero Ambiguity: If the key matches, the profiles are merged without a confidence score.
• Limitation: Fails to connect users across devices or channels where they haven't authenticated.

Uses statistical models and machine learning to infer identity linkages based on patterns in non-unique attributes like IP addresses, browser fingerprints, device types, and behavioral cadences. This generates a confidence score rather than a binary match.

• Fuzzy Logic: Weighs dozens of weak signals (e.g., operating system, screen resolution, typing speed) to calculate a similarity score.
• Threshold Tuning: Administrators set a minimum confidence threshold (e.g., 95%) to trigger an automatic merge, routing lower scores to a manual review queue.

The underlying data structure that stores the relationships between all known identifiers for an individual. It maps transient identifiers (cookies, mobile ad IDs) to a persistent master profile.

• Node-Edge Model: Each identifier is a node; a deterministic match or a high-confidence probabilistic link forms an edge.
• Temporal Decay: Weights on edges decay over time to deprecate stale connections, such as a shared IP address that changes.
• Graph Traversal: Enables the system to connect a new anonymous cookie to a known CRM record by traversing intermediate nodes.

Extends identity resolution beyond a single individual to group profiles into a household (B2C) or a buying center (B2B). This prevents duplicate marketing pressure and reveals organizational purchasing intent.

• Shared Signals: Uses shared physical address, Wi-Fi network fingerprints, or corporate email domains.
• B2B Graph: Maps individual contacts to a master company account, aggregating engagement scores to qualify the entire account.

The non-negotiable architectural requirement to separate the identity graph from raw personally identifiable information (PII). The system operates on irreversible pseudonymous tokens.

• Tokenization: Emails and names are replaced with vaulted tokens before entering the resolution engine.
• Data Partitioning: The identity graph stores only the relationships between tokens, while the raw PII remains encrypted in a separate, access-controlled vault.

The capability to resolve an anonymous visitor to a known profile within milliseconds of the first page view, enabling in-session personalization. This requires high-performance key-value stores and edge-deployed logic.

• Latency Budget: Resolution must complete in under 50ms to avoid blocking the page render.
• Cache-First Architecture: Recently active identity graphs are held in-memory to avoid a slow disk lookup on every hit.

A data structure that maps all known identifiers for an individual—email addresses, device IDs, usernames, and offline keys—to a single master profile. Identity graphs serve as the persistent output of the resolution process, linking anonymous and known states across touchpoints.

• Deterministic links: Exact matches on hashed PII like email or phone
• Probabilistic links: Statistical matches on IP, browser fingerprint, and behavioral patterns
• Householding: Groups individual profiles into shared physical or economic units

A probabilistic device identification technique that combines dozens of subtle browser and operating system attributes to create a unique identifier without relying on cookies. Attributes include canvas rendering, WebGL vendor, installed fonts, and audio stack characteristics.

• Generates a hash from 30–100+ browser signals
• Entropy of 15–20 bits is typical for stable identification
• Increasingly restricted by browser privacy protections like Intelligent Tracking Prevention

A marketer-managed system that creates a persistent, unified customer database accessible to other systems. A CDP ingests data from online and offline sources, performs identity resolution natively, and exposes segments to execution channels.

• Aggregates first-party behavioral, transactional, and demographic data
• Maintains a golden record with full historical lineage
• Exposes unified profiles via API for real-time personalization engines

The process of grouping a sequence of individual user events and pageviews into a single coherent visit. Session boundaries are typically defined by a 30-minute inactivity timeout, though custom thresholds apply for streaming or gaming contexts.

• Critical for stitching pre-login and post-login behavior
• Enables accurate attribution of conversions to specific sessions
• UTC midnight rollover handling prevents split-session artifacts

An identity resolution method that links records using exact, verifiable identifiers such as hashed email addresses, phone numbers, or loyalty card numbers. This approach produces high-confidence matches with near-zero false positive rates.

• Requires explicit user authentication or PII submission
• Forms the backbone of cross-device login-based stitching
• Limited by coverage: only matches authenticated touchpoints

A statistical approach that links identifiers using machine learning models trained on non-PII signals like IP address, user agent, geolocation, and behavioral patterns. Confidence scores quantify match likelihood.

• Extends identity coverage to anonymous traffic
• Models require continuous retraining as device ecosystems evolve
• Typically combined with deterministic links in hybrid resolution engines