Session Stitching

The core logic for defining session boundaries. Stitching algorithms override default 30-minute inactivity timeouts by analyzing inter-event arrival times and contextual signals.

• Custom Thresholds: High-velocity gaming apps may use a 5-minute timeout, while a B2B research portal might extend to 4 hours.
• Intent Signals: A session is not terminated if a user adds an item to a cart, even if a 30-minute pause occurs, preserving the purchase intent.
• Midnight Rollover: Logic must handle UTC vs. local timezone discrepancies to prevent artificially splitting a single user session across two calendar days.

Stitching employs a hybrid strategy to connect sessions. The algorithm prioritizes high-certainty signals before falling back to statistical inference.

• Deterministic Anchors: A hashed email key or login event instantly merges a previously anonymous session with a known profile.
• Probabilistic Bridges: When a user switches from a mobile browser to an app without logging in, the system evaluates IP address, device fingerprinting entropy, and Wi-Fi SSID to assign a confidence score.
• Linkage Decay: A probabilistic link's weight decreases over time, preventing a session from a coffee shop's shared IP address from permanently polluting a user's identity graph.

When two disconnected sessions are stitched, their internal state must be merged without corrupting the user experience.

• Cart Merging: If a user adds a shirt on mobile and shoes on desktop, the stitched session must display both items, deduplicating SKUs added via both devices.
• Last-Write Wins: For non-critical state like the last search query, the most recent timestamp wins.
• Immutable Event Log: The raw clickstream is never altered. Stitching creates a new canonical session ID that points to the original discrete sessions, preserving the audit trail for debugging.

The latency of the stitching process dictates the use case it can serve.

• Real-Time Stitching: Performed by a streaming data pipeline (e.g., Apache Kafka, Kinesis) within milliseconds. Essential for next-best-action models and fraud detection that require an immediate unified view.
• Nearline/Batch Stitching: A periodic job (e.g., hourly Spark job) that processes complex graph neural network linkages too computationally expensive for real-time. Used for customer lifetime value forecasting and attribution reporting.
• Lambda Architecture: A production system often combines both, using a speed layer for immediate low-latency links and a batch layer to correct and solidify the identity graph.

Stitching logic must be gated by a user's dynamic consent state. A technical linkage is only valid if legally permissible.

• Consent Gate: The stitching engine queries a Consent Management Platform (CMP) to check if a user has opted out of tracking before merging a session.
• Right to Deletion: If a user invokes a data subject access request (DSAR), the stitching logic must be able to unlink all associated sessions and reprocess the profile as purely anonymous.
• Purpose Limitation: A session stitched for security anomaly detection may be prohibited from use by the marketing personalization engine, requiring separate, purpose-bound identity graphs.

Stitching logic differentiates between a simple device switch and a fundamental channel shift to apply the correct business context.

• Cross-Device: Linking a user's iPhone to their MacBook via iCloud Private Relay masking requires advanced fuzzy matching on browser locales and screen geometries.
• Cross-Channel: Connecting a web session to an in-store point-of-sale transaction using a loyalty card number or a hashed email key captured at the register.
• Offline-to-Online: Onboarding a call center transcript to a web session by matching the dialed phone number (ANI) to the account's primary phone field, creating a true omnichannel journey.

A centralized data structure that links all known identifiers—email addresses, device IDs, and usernames—to a single unified customer profile. The identity graph serves as the persistent backbone that session stitching algorithms query to determine if two anonymous sessions belong to the same individual.

• Stores deterministic links (hashed login credentials)
• Maintains probabilistic edges with confidence scores
• Resolves identity across browsers, apps, and CTV devices

A method of identity resolution that relies on exact, verified matches of personally identifiable information (PII) to link sessions with absolute certainty. When a user logs in on both mobile and desktop, the hashed email key provides an unbreakable stitch point.

• Zero ambiguity — match confidence is 100%
• Typically uses hashed email or phone number
• Forms the anchor points in a session stitching timeline

A statistical approach that infers device ownership using non-personal signals like IP address, browser type, screen resolution, and behavioral patterns. Session stitching engines use probabilistic matching to connect anonymous pre-login sessions to known profiles.

• Assigns a confidence score (0.0–1.0) rather than a binary match
• Critical for stitching sessions before authentication events
• Models degrade gracefully as signals age

A technique that collects a device's unique configuration attributes—installed fonts, WebGL rendering hash, canvas fingerprint, and audio stack—to generate a persistent, stateless identifier. Session stitchers use fingerprints to recognize returning devices even when cookies are cleared.

• Operates without any client-side storage
• Increasingly restricted by browsers (e.g., Safari's ITP)
• High entropy but privacy-controversial

A temporal model that progressively reduces the linkage confidence of an identifier as it ages without fresh validation. A device fingerprint from six months ago carries less weight in a stitching decision than one from six minutes ago.

• Prevents stale identifiers from polluting profiles
• Uses exponential decay functions or half-life models
• Essential for maintaining golden record accuracy over time

The measurement methodology that tracks a consumer's exposure to a marketing touchpoint on one device and the subsequent conversion event on another. Session stitching is the technical prerequisite — without it, the conversion appears orphaned.

• Connects CTV ad views to mobile purchases
• Requires identity graph resolution across device types
• Foundation for incrementality testing and media mix modeling