What is cross-device tracking?

Why it matters

Real journeys span devices. Someone researches on a phone at lunch and buys on a laptop that evening. Measured per device, that is two unrelated strangers: one who browsed and left, one who bought out of nowhere.

The consequences run in both directions. Research activity looks like abandonment, purchases look unattributed, frequency capping fails because the same person is counted as several, and audience sizes are inflated by duplicate identities.

Deterministic and probabilistic matching

• Deterministic. A shared login links the devices. Accurate, and available only where users sign in.
• Probabilistic. Devices are inferred to belong together from shared IP, timing and behaviour. Broader coverage, real error rates, and increasingly restricted by both regulation and platform policy.
• Household-level. Devices on one network treated as a unit — useful for some media questions, and wrong for anything personal.

Why the practice narrowed

Probabilistic cross-device matching depended on signals that have been deliberately degraded: third-party cookies, stable IP addresses, and device characteristics that browsers now actively blur.

What remains is deterministic matching on first-party login, which works well and only covers users who sign in. For most consumer businesses that is a minority of traffic, so cross-device journeys are now substantially unobservable — a permanent condition rather than a gap awaiting a technical fix.

This is a large part of why multi-touch attribution degraded: the path it reconstructs is missing whichever devices the user was not logged in on.

What to do instead

Invest in first-party identity where the product genuinely justifies an account, and do not manufacture reasons to force one. Then measure at a level that does not require identity at all: incrementality tests and media mix modeling both read total outcomes and are indifferent to how many devices a person used.