How should you attribute sales to retargeting campaigns?
Retargeting has the best-looking numbers in most ad accounts and the least trustworthy. The channel targets people who already visited, already added to cart, already showed intent, then claims the sale when they convert. Some of that credit is earned. Much of it is harvested. Telling the difference is the whole job.
The harvested-intent problem
Consider a shopper who adds a product to cart on Monday, sees a retargeting ad on Tuesday, and buys on Wednesday. The retargeting platform claims the sale. Would the shopper have bought anyway? Often yes: cart abandoners convert on their own at meaningful rates, especially within a few days. The ad may have accelerated the purchase or merely accompanied it. Last-click attribution cannot distinguish the two, so it credits the ad for all of it.
This is not fraud; it is measurement design. Any channel that targets high-intent audiences will show great attributed numbers, because the audience was going to convert at high rates regardless. The question is never did retargeting touch the sale. The question is did retargeting cause the sale, or cause it sooner, or cause a bigger basket. Those are three different claims and they need three different measurements.
Holdouts are the only honest answer
The clean way to measure retargeting is a holdout: suppress the retargeting audience for a random subset of eligible users and compare conversion rates. The difference between the exposed group and the holdout group is the true incremental lift. Everything else is modeling.
Holdouts feel expensive because you are deliberately not advertising to people you could reach. Reframe the cost: you are buying the knowledge of what the channel is actually worth, which determines how much of your budget it deserves. A retargeting program that shows 5x ROAS on last-click and 1.4x on holdout-measured incrementality is not a 5x channel. It is a 1.4x channel with great reporting. Budget accordingly.
Windows should match the buying cycle
Retargeting windows are usually set by platform default: 30 days, sometimes 180. These defaults are too long for most products. A shopper who viewed a product 45 days ago and buys today was not persuaded by the ad; the ad was wallpaper. Long windows inflate attributed revenue with conversions the channel did not cause.
Set windows from your own data. Look at the time between first visit and purchase for your customers, and set the window to cover the period where retargeting could plausibly influence the decision. For most DTC products that is days, not months. Shorter windows will make the channel's numbers look worse and its budget decisions look better, which is the trade you want.
Separate the jobs retargeting does
Retargeting actually does three different jobs, and they deserve separate measurement. Cart abandonment recovery targets the highest intent and mostly accelerates purchases that were likely. Product view retargeting nudges consideration and has genuine persuasive power. Winback retargeting to old customers is closer to retention marketing. Lumping all three into one line item hides which job is earning its keep.
Split them in reporting and in holdout design. You will typically find that cart recovery shows the smallest incremental lift (the audience converts anyway) while mid-funnel view retargeting shows the largest. That finding should reshape the budget: spend less on harvesting carts, more on persuading browsers. Most brands do the opposite because the dashboard rewards it.
Retargeting is not a scam, but its default measurement is flattering to the point of fiction. Holdouts, honest windows, and split reporting turn it from the channel with the prettiest dashboard into a channel you can actually scale with confidence. The brands that do this work usually discover retargeting is worth keeping but worth less than they thought, which is valuable information at any budget.