Retargeting consistently reports the strongest return in an account. The number is real and the interpretation is usually wrong, because of who the audience contains.
The audience is selected for intent
A retargeting pool is built from people who visited the site, viewed a product or began a checkout. Every one of those actions signals existing interest.
Advertising to that group produces conversions at a high rate, which is expected given who they are.
The question the reported return does not answer is how many would have converted with no advertising at all.
Credit is claimed for returning visitors
Someone who abandoned a basket often intends to return, and many do so directly, through a bookmark or a search for the brand.
If a retargeting ad was served and clicked in the interim, that conversion is attributed to the campaign.
The ad may have been the reason, or it may have been the last thing touched on a path that was already going to complete.
Nothing in standard reporting distinguishes the two cases, and the campaign is credited identically whether it changed the outcome or merely observed it.
The measured return rises as the audience narrows
Restricting retargeting to the most recent and most engaged visitors improves every reported metric, which looks like optimisation.
What is actually happening is that the audience is increasingly composed of people who were going to buy regardless.
Taken to its limit the campaign advertises only to people mid-purchase and reports an extraordinary return while contributing almost nothing.
Holdout testing gives the real number
Withholding retargeting from a randomly selected share of the eligible audience and comparing conversion rates isolates the incremental effect.
Most platforms support this directly, and where they do not, a suppressed segment achieves the same thing with more effort.
The incremental figure is typically far below the reported one, and knowing the gap changes how much budget the channel should receive.
Where it earns its place
Retargeting works best where the purchase genuinely stalls, on considered purchases with long decision periods and multiple sessions.
It works least well immediately after a visit, when the person is still deciding and needed no reminder.
Delaying entry into the pool by a few days, and excluding recent converters, removes most of the spend that was buying conversions already in progress.