Cross-Device Tracking & Attribution: A Complete Guide

Quick Answer

Cross-device tracking links one shopper's phone, laptop and tablet into a single journey, and since iOS 14.5 it has had less to work with: apps must ask permission before tracking people across other companies' apps and websites. Logged-in matching still works; the rest is guesswork on shrinking signals. So measure what each channel caused instead, where the device doesn't matter.

The Multi-Device Mess: Why Your Attribution Is a Lie

A customer finds your brand on TikTok on the morning commute, researches on a work laptop and buys on a tablet at home. Unless something ties those visits together, attribution sees three people instead of one customer, and whatever brought them to the tablet gets the credit for the whole sale.

Platform-reported ROAS doesn't fix it, because each platform only sees its own slice. Google counts the search click, Meta counts the ad view, and the same sale can show up in both reports.

How Cross-Device Tracking Works (And Why It Fails)

See also: Cross-Channel Attribution Models Explained: Stop Guessing

Deterministic Matching

Deterministic matching links devices with a known identifier, such as an email address or a login. It's accurate, but only for people who sign in on each device. GA4 works the same way: its reporting identity joins devices through a User-ID you collect, usually from a login, and otherwise counts each browser or app separately.

Probabilistic Matching

Probabilistic matching guesses which devices belong together from signals like IP address, device type and browsing patterns. It covers more people but gets more of them wrong, and Apple's App Tracking Transparency, which lets people refuse tracking, cuts the signals it relies on.

The Privacy Problem

Apple's ATT, browser tracking protection and privacy laws like GDPR keep cutting the signals cross-device attribution depends on. Safari and Firefox block or isolate third-party cookies by default, and fingerprinting runs into browser protections and consent rules. Attribution that depends on following people across the internet rests on shrinking ground.

Attribution Models: From Bad to Worse

Last-click attribution gives 100% of the credit to the final touchpoint and ignores awareness and consideration. First-click attribution does the opposite. Linear models spread credit evenly, which sounds fair but is arbitrary. Time-decay models favor recent touchpoints. None of them answers the question that matters: what caused the sale? And across devices, each one only sees the touchpoints that got stitched into the same journey.

Correlation does not equal causality. Just because a click happened before a purchase does not mean it caused the purchase. This is the foundational flaw in every traditional attribution model.

Data-driven attribution from Google uses machine learning to assign credit, but only across what Google can see: in Google Ads it shares credit among your ad interactions on Google, and Google's data-driven model is trained and reported by Google. It's a smarter guess, but still a guess. The Attribution Models Compared guide goes deeper.

Server-Side Tracking: A Partial Fix

See also: How to Fix iOS 14 Tracking for Shopify Stores (2026 Update)

Server-side tracking moves data collection from the browser to your server, so ad blockers and browser limits drop fewer events, though consent rules still apply. It improves the data, but a correlation-based model still reads it the wrong way. Server-side tracking gives you better data; causal inference gives you better answers. There's more in the Server-Side Tracking Guide.

What Works Without Cross-Device Tracking

Measuring channels instead of people sidesteps the problem: a channel either added sales or it didn't, whatever device the buyer used. That's the idea behind causal attribution, and you can test it yourself.

Three checks that don't need cross-device tracking:

1. Put the double counting in proportion. Add up the purchases Meta and Google each claim for last week and compare the total with your store's orders. Anything above your real orders is credit counted twice, not extra sales.
2. Watch the number every device ends up in. Your store records every order, whatever the device, so MER (total revenue over total ad spend) needs no stitching.
3. Test instead of tracking. Pause one channel in one region for two weeks and compare that region's revenue with the rest, each against its own two weeks before. The sales that disappear are what the channel caused, on every device.