Remarketing Campaigns: When They Work and When They Waste Budget

Remarketing campaigns report the best numbers in almost every account. They also contain more attribution illusion than any other campaign type, because they are structurally positioned to take credit for conversions that were going to happen anyway.

Both things are true at once, and separating them is the whole discipline.

The incrementality problem, stated plainly

A remarketing audience is by definition made up of people who already visited your site. Some proportion of them would return and convert with no advertising at all. When you show them an ad and they convert, the platform attributes the conversion to the ad.

The reported return on ad spend can therefore be excellent while the incremental return is close to zero. This is not a flaw in the campaign; it is a flaw in reading the report as though it measured causation.

The test is a holdout: exclude a random share of the audience from seeing ads, and compare conversion rates between the two groups. If the excluded group converts at a similar rate, the campaign is largely reporting demand it did not create. Running that test once is worth more than a year of optimisation.

Where remarketing genuinely creates value

Long consideration cycles. In real estate and furniture, a purchase decision runs over weeks and often involves more than one person in a household. Staying present across that window is a real service, not a nudge.

Abandoned high-value baskets. The clearest case in retail, and the one where the arithmetic usually survives a holdout test.

Cross-sell to existing customers. Customer match lists against a known purchase history, with a genuinely different offer.

Excluding rather than including. The most underrated use of a remarketing list is as a negative audience — suppressing existing customers from acquisition campaigns, or removing recent converters from prospecting. This costs nothing and improves every other campaign.

Where it burns budget

All-visitors audiences. A list that includes everyone who hit any page, including bounces at two seconds, is mostly people with no interest. Segment by behaviour and recency or do not run it.

Windows that are too long. A 540-day window in a category with a two-week decision cycle is paying to reach people who bought elsewhere a year ago. Match the window to the actual cycle.

Frequency without a cap. Beyond a modest number of impressions a week, additional exposure stops persuading and starts irritating. The damage does not show in the campaign report, because annoyance is not a metric.

Brand-term overlap. Where remarketing and brand search both chase the same returning visitor, the account pays twice for one conversion and both campaigns report it.

Segmentation that earns its keep

Three axes, in order of usefulness: behaviour on site (viewed a product, started a form, reached checkout), recency (the last few days versus the last few weeks), and value (basket size or product tier).

Each combination should get a different message. The same creative served to a two-second bounce and to a checkout abandoner treats two entirely different situations identically, and it is why so many remarketing accounts plateau. Segmentation is a measurement exercise before it is a creative one.

Privacy changes have altered the mechanics

Cookie deprecation, tracking prevention in mobile browsers and consent requirements have all reduced the size and accuracy of browser-based lists. The durable alternatives are first-party: customer match from your own CRM, and server-side tagging with enhanced conversions.

Accounts still relying entirely on browser-side pixels are working with lists that are quietly shrinking and increasingly unrepresentative. Rebuilding on first-party data is not an optimisation — it is maintenance, and the accounts that deferred it are finding their audiences too small to serve.

If your remarketing numbers look too good, they may be. A holdout test settles it, and we build them into the accounts we run — the reasoning is in how we approach performance.

If remarketing is reporting your best return and you suspect it is claiming credit rather than creating demand, an incrementality test is where to start. We design and read one as part of a written review.