Cross-device measurement: why your customer sees the ad on their phone and buys from their computer

A customer is on the metro on their way to work, sees your ad on Instagram from their phone, is interested, but doesn't buy right there (nobody enters their card details standing in a packed train carriage). They get to the office, remember it, open their computer and search for your brand on Google. They buy. Which channel do you credit that sale to? If your answer is "the Google search, because it was the last click before buying", you're making the most common (and most expensive) measurement mistake in online advertising: ignoring that the same customer uses several devices before deciding.

This is called cross-device behaviour, and the problem isn't that it exists (it always has, people have been comparing prices on their phone and buying on their computer for years), it's that most measurement systems, by default, don't see it properly, and that leads to wrong decisions: pausing Instagram campaigns that were actually generating sales, just credited to a different channel.

Why this happens, technically

Each platform (Google, Meta, TikTok) measures mainly within its own ecosystem and within the same device, unless the user is logged into their account on both devices and the platform can link that identity. Meta, for example, can recognise a user who saw the ad on their phone with the Instagram app open and then bought from their computer if in both cases they were logged into the same linked Facebook or Instagram account. But if they use a personal phone and an office computer with different profiles, or clear cookies, or use Safari on an iPhone (which by default blocks much of cross-site tracking), that link breaks and the conversion appears to come out of nowhere, credited to the last channel that could actually see it: usually, a direct search for your brand on Google.

The practical effect: underrating what works

This has a very concrete and very costly consequence: top-of-funnel campaigns (the ones that generate initial interest, typically on social media or display) rack up clicks and impressions but few "attributed" conversions, while brand search on Google gets the credit for sales that actually started somewhere else. A business that judges its campaigns only by what each platform reports in isolation ends up, with apparent but mistaken logic, cutting budget precisely from the campaign that was bringing those customers in at the start of the journey.

In a real case with a fashion ecommerce client, when we switched on an attribution model that accounted for the full journey (not just the last click), we discovered that the Instagram campaigns, which looked like they had a sky-high cost per sale viewed in isolation, were actually involved in 40% of the purchases that Google Ads ended up claiming as "brand search". The budget about to be cut from Instagram was, in fact, the very thing sustaining much of what got credited to Google.

What to do about it (without needing a data department)

The first piece is simple and free: switch on data-driven attribution in Google Ads and Google Analytics 4 if you haven't already, instead of leaving the default "last click" model. This model splits conversion credit across all the channels involved along the way, not just the last one, using aggregated data from millions of conversions to estimate the real weight of each step.

The second is accepting that platforms alone will never give you the full picture, and compensating for that with simple surveys: a "how did you hear about us?" field on your contact form or checkout process, with options like "social media", "Google search", "someone recommended us". It isn't perfect and doesn't replace data, but it works as a common-sense check against what each platform says on its own.

The third, more advanced but accessible to almost any business with some volume, is measuring by "incremental lift": pausing a specific campaign for a week or two and watching whether total business sales (not just the ones attributed to that channel) drop noticeably. If pausing Instagram makes total sales fall, even though Instagram "only" appeared attributed to 5% of conversions, that's a sign it was influencing far more than its individual report showed.

The phone's role as "discoverer", not always "buyer"

It's worth understanding the general pattern, because it repeats a lot in commerce with a mid-to-high average order value: the phone is the device of discovery (scrolling, curiosity, first contact) and the computer remains, for many categories, the device of the final decision, especially when there are long forms to fill in, several tabs to compare at once, or payment details to enter more carefully. This isn't universal (fashion and impulse-buy products do close a lot on mobile), but knowing the pattern for your specific sector saves you from drawing the wrong conclusions from your own reports.

Tools that help without being expensive

Google Analytics 4 already builds, by default, "conversion path" reports that show, for each sale, the sequence of channels involved before it closed, not just the last one. Checking that report once a month (you don't need it open daily) is usually enough to spot whether you're underrating a channel. If you use a CRM, noting down the source the customer themselves states on the first call or email also adds a layer of truth no pixel can give you.

An example with numbers: the real journey behind a furniture purchase

An office furniture manufacturer selling directly online reviewed, over one quarter, the full journey of a hundred random purchases using Analytics' conversion path reports. The most repeated pattern (present in 41 of the hundred purchases) was: discovery on Instagram from a phone, a visit to the site without buying, a brand search on Google a few days later from the office computer, and the final purchase on that second device. If that company had judged Instagram's performance solely by the conversions the platform itself credited to it in its dashboard, it would have concluded that channel barely generated any business, when in reality it was involved in over 40% of total sales.

The second most common pattern (23 of a hundred purchases) was the reverse: discovery through a direct Google search from the office computer, with no detectable prior social channel. This confirmed something the team suspected but had no way to verify: a real share of demand arrived through word of mouth between colleagues, without passing through any measurable advertising channel, which is also valuable information when deciding where it isn't worth forcing more investment.

The role of first-party data in all this

Part of the long-term solution involves relying less on identifying the user through third-party cookies (increasingly restricted) and more on first-party data the business collects directly: a customer account with login, a points or loyalty programme, or simply a checkout that asks for an email before showing the final price. The more users voluntarily identify themselves through your own system, the less you depend on whatever platform managing to link their devices on its own, and the more reliable your own internal measurement becomes compared with whatever any external platform reports to you.

A simple exercise to do this week

You don't need a perfect measurement system in place to start benefiting from this idea. A one-hour exercise any business can do with the tools it already has: log into Google Analytics 4, find the "multi-channel conversion paths" report (under the advertising or conversions section) and review the ten most common paths to a sale. You'll almost certainly find at least one channel that shows up frequently as an intermediate step but is rarely the last click, and that has probably been underrated in your budget decisions for a while.

That finding, on its own, doesn't have to immediately change how you split your budget, but it should change how you interpret the reports each platform gives you separately. The next time someone proposes cutting a channel because of its seemingly high cost per result, that conversion path is the first place to check whether that conclusion holds up against a fuller look at the customer's real journey.

How the new generation of privacy-focused browsers affects this

Beyond Safari, more and more browsers (Firefox with its enhanced tracking protection, and recent versions of Chrome progressively rolling in third-party cookie restrictions) are adopting stricter privacy policies by default. This trend isn't temporary and isn't going to reverse: it's the whole industry's general direction, pushed both by regulatory pressure and by users' own demand. Any business relying on measurement based exclusively on third-party cookies is going to see, as the years go by, an ever-widening measurement gap unless it adapts its strategy toward first-party data and more robust attribution models.

Common mistakes when trying to fix cross-device measurement

The first mistake is turning on data-driven attribution in Google Ads but still looking, out of habit, at "assisted conversions" reports or Meta's platform breakdown as if they were absolute truth, without realising both still operate within their own ecosystem and can't see what happens on the competing platform. The second mistake is implementing measurement changes (enabling Google signals, asking for the email earlier) and expecting visible results the same week: these adjustments take time to build up enough history for the attribution model to have reliable data to work with, usually between four and eight weeks.

The third mistake, perhaps the costliest, is using the excuse of "it's hard to measure precisely" to stop measuring altogether and manage the budget by gut feeling. Cross-device measurement is more complicated than looking at a single number, but that doesn't mean it can't be approximated with a reasonable degree of accuracy using the tools described in this article. The alternative (measuring nothing and splitting budget by eye) almost always ends up more expensive than accepting some degree of uncertainty in the available data.

Frequently asked questions