At its core, cross-device tracking connects a person's interactions across phones, desktops, tablets, and connected TVs into a single journey. Two linkage methods power this:

• Deterministic: ties devices using authenticated IDs such as logins, hashed emails, customer IDs, or consented ad IDs. It's precise, consent-dependent, and strongest where users sign in.
• Probabilistic: infers links from signals like IP, time, OS, browser, language, and device attributes with statistical models. It extends coverage but requires strict privacy controls, has confidence thresholds, and is sensitive to signal loss.

The practice is changing fast. Signals from third-party cookies and mobile IDs are limited by regulations and platform policies. Teams are shifting to:

• First‑party identity: durable account-based IDs, server-side tagging, and customer data platforms to unify profiles under consent.
• Clean rooms: privacy-safe joins with media partners using hashed identifiers and aggregation.
• Modeled measurement: MMM, geo experiments, and incrementality tests to validate what attribution suggests.
• Granular consent: honoring purpose-based consent and regional choices to meet GDPR, CCPA/CPRA, and Apple's ATT.

When done correctly, cross-device tracking improves attribution quality, frequency management, and personalization without over-relying on any single identifier.

Use this pragmatic sequence to deploy cross-device measurement with control and compliance:

1. Map identity surfaces: list every login surface, CRM ID, newsletter, app account, and offline customer key. Decide the primary person ID, and how device IDs and cookies map to it.
2. Collect with consent: implement clear notices, purpose-based opt-in, and an audit trail. Respect ATT for iOS apps and regional choices server-side.
3. Instrument events consistently: standardize event names and properties across web, app, and CTV. Capture timestamps, campaign parameters, and device context in a unified schema.
4. Link devices: use deterministic joins first (login, hashed email). Add probabilistic links with confidence scores and recency windows. Never overwrite deterministic links with probabilistic ones.
5. Choose your attribution mix: run a cross-device multi-touch model for day-to-day optimization, and validate with experiments or MMM for causal impact. Use data-driven models only where sample sizes support stability.
6. Control ad frequency: set person-level caps and guardrails per channel. Reconcile platform caps with your own identity graph to avoid overserving.
7. Govern data: define retention windows, access controls, and deletion workflows. Minimize data fields to what you truly need for measurement.
8. Prove value: create a recurring report that shows lift from de-duplication, corrected ROAS, and budget shifts driven by cross-device insights.

Success criteria: higher attributed conversions without inflation, lower cost per incremental conversion, and fewer wasted impressions from duplicate reach.

Cross-device tracking pays off when it sharpens the metrics you act on. Focus on these:

• De-duplicated reach and frequency: person-level reach by channel and device, with overlap matrices. Watch for high overlap that signals waste.
• True conversion paths: path analysis that shows device order (e.g., mobile discovery → desktop purchase). Use this to tune creative and landing experiences.
• Cross-device ROAS: revenue and margin tied to the person, not the device. Compare to platform-reported ROAS to identify over-crediting.
• Incremental lift: geo split tests or holdouts to validate what your attribution suggests. Treat lift as the north star when reallocating spend.
• Time-to-convert by device: latency distributions that reveal where assists happen. Shorten the path with better handoffs and remarketing rules.
• Suppression and cap savings: quantify impressions avoided from person-level suppression and the CPM saved.

Use these insights to adjust budgets toward devices and touchpoints that reliably assist or close, refine bidding, and set creative that fits the screen where each step happens.