Is retargeting incremental? Ask what would have happened anyway

Think about who sees the ads. They have been on your site. Some looked at a product. Some added it to the basket. A share of them were going to buy anyway.

Now add the way platforms count. A platform credits a conversion to an ad when someone saw or clicked it within an attribution window. It does not know what that person would have done without the ad. A view-through conversion, where someone saw the ad but did not click, makes this worse for a channel that targets people close to buying.

The result is a low CPA and a high ROAS. Sellforte, a vendor that sells marketing mix modelling, notes that retargeting reaches people who already visited the site, and that platform ROAS is often higher than incremental ROAS for lower-funnel activity¹. It sells measurement, so read it as a view. The mechanism is easy to follow all the same.

A practitioner's newsletter makes the same point from the advertiser's side. It describes a retargeting-only campaign with a very good-looking cost per acquisition, where adding budget brought no more conversions². The post has no hard numbers, and we do not borrow any.

Less than the internet suggests. Here is what we can trace.

None of these is a controlled experiment on your account. They are reasons to ask the question, not answers.

Many retargeting statistics circulate with no traceable source. We do not repeat them. If someone quotes you a lift figure for retargeting, ask where it came from and which brand it was measured on.

The cleanest test for retargeting is an audience holdout test. The idea is simple: keep a random slice of your retargeting audience out of the ads. Compare their purchases with the group that saw them.

• Audience holdout. Split the retargeting list at random. One part is eligible for ads. The other is not. If the exposed group buys more, the difference is the incrementality of the ads. Because the split is random, the two groups should look alike.
• Exclusion. Turn retargeting off for a defined group or period and compare total sales. It is cheaper to set up. It is also easier to fool yourself, because anything else that changed in the period gets blamed on the ads.

Pick one and decide in advance how you will read it. Write down what result would make you cut the budget, and what result would make you keep it.

Example (illustrative): A retargeting list has 100,000 people. You hold out a random 10,000 for a month. The 90,000 who can see the ads buy at 0.20%, which is 180 purchases, and the platform credits most of them to retargeting. The held-out 10,000 buy at 0.18%. Only about 18 of the 180 purchases were caused by the ads. The rest was existing intent.

Watch out for a trap: the platform may offer to run the test for you. Its holdout tool can be a practical start, and the practitioner post above suggests exactly that². But AdExchanger warns against letting a platform decide whether sales are incremental or customers are new³. A platform grading its own homework is a conflict of interest. Check the result against your own sales data where you can. Our piece on platform reported ROAS vs incremental ROAS goes deeper.

If you want a number to compare, incremental ROAS (iROAS) is incremental revenue divided by ad spend. It is the figure the test should give you.

A common story: the team doubles retargeting spend, the dashboard still shows a healthy CPA, and total sales barely move.

That pattern has a plain reading. The audience is limited. It is the people who already visited your site. You can raise spend, but you mostly buy more impressions on the same people. If those people were going to buy anyway, more impressions add little.

So use total sales as the check. If more spend does not move total sales, the extra spend probably isn't incremental. Look at total revenue and MER over the same weeks. In our convention, MER is total revenue divided by total marketing cost (media, agency, creative and tools), not media spend alone. Check both, not just the channel dashboard.

Sometimes it does. The sponsored article in The Drum argues that retargeting can be highly incremental where the path to purchase has many obstacles and people drop out halfway⁴. Our view: retargeting earns its place when a holdout shows a clear gap between exposed and held-out buyers, and when the cost of the ads is below the margin that gap represents. The test decides, not the dashboard.

A few conditions help, in our view:

• A long or awkward path to purchase, where a reminder can matter.
• A catalogue large enough that people forget what they looked at.
• Audience sizes big enough to hold out a slice and still read the result.

None of these guarantees a result. They tell you where a test is most likely to be worth running.

Our view: size the budget on the incremental margin the test shows, not the CPA the platform reports.