Remarketing

Remarketing shows the best cost per result in almost every account, and the reason is not that the advertising is better. It is that everyone in the audience had already visited the site, viewed a product or abandoned a basket. They were selected precisely because they showed intent.

Attribution then credits the conversion to the last ad they saw. Some of those people were genuinely brought back by it. Many would have returned regardless, because they had already decided and were waiting for payday, or comparing, or simply had not finished.

This is why remarketing is the first place to run an incrementality test. It is where the gap between attributed conversions and caused conversions is widest, and the gap is measurable in a fortnight with a holdout.

None of this means remarketing does not work. It means the reported figure overstates by an amount nobody knows until they measure it.

The audience pool is produced by prospecting. Every person in a remarketing list arrived from somewhere else first, so remarketing is downstream of an activity it cannot perform.

The failure mode follows from that, and it is common enough to be predictable. Remarketing reports well, so budget shifts toward it. Prospecting shrinks, so fewer new people enter the pool. The pool ages, frequency climbs against a smaller audience, and results decline six to eight weeks later for reasons that look like fatigue and are actually starvation.

By the time it shows up, the diagnosis is awkward: the account got worse because the efficient part of it was funded at the expense of the part that fed it.

The defence is structural. Set the prospecting budget first, as a floor rather than a remainder, and let remarketing take what is appropriate to the audience size rather than what its reported ROAS would justify.

Browser restrictions have steadily shortened how long a site visitor can be recognised. Safari Intelligent Tracking Prevention has limited third-party cookies since 2017 and caps the lifetime of script-set first-party cookies; Firefox blocks third-party cookies by default. The practical effect is that a 30-day remarketing window often is not really thirty days.

Consent is the other half in Europe. A visitor who declines is never added to the audience at all, so the pool reflects the share of traffic that accepted rather than the traffic that arrived.

The response most advertisers have landed on is first-party data: customer lists uploaded and matched, and server-side event collection through the Conversions API or enhanced conversions, which survives browser restrictions because it does not depend on the browser.

That shift moves remarketing from a tactical setting to something that needs the client CRM and their cooperation, which changes who has to be in the conversation.

Platforms restrict what audiences may be built and targeted around sensitive categories, including health, finance, and characteristics such as religion or sexual orientation. Building a remarketing list from visitors to a page about a medical condition can breach personalised advertising policy even though the mechanism is the same as any other list.

Uploading a customer list is a disclosure of personal data to a third party. Under GDPR it needs a lawful basis and it has to be covered by the privacy notice those customers actually saw. Consent given for a newsletter does not automatically extend to advertising.

The obligations sit with the client as controller, which is exactly why an agency should not be making this call quietly. Confirming the lawful basis before the first upload takes one email and removes the version of this conversation that happens after a complaint.