Retargeting Campaigns Turn Your Estimand from ATE to ATT

In various platforms, one has the option to retarget audiences. For example Meta’s ASC Retargeting campaigns or Google’s Pmax.

Marketers / analysts however apply the same measurment philosophy for retargeting as they do for other non-retargeting campaigns.

This in my opinion is not the right approach.

In case of Retargeting, most marketers think they are measuring overall impact. They are not. They are measuring impact on a very specific, pre-selected group.

Let me briefly explain the causal estimands of ATE and ATT.

ATE (Average Treatment Effect) -> Effect of treatment on the entire population

ATT (Average Treatment Effect on the Treated) -> The Average Treatment Effect on the Treated (ATT) specifically measures the average effect of the treatment on those individuals who actually received it.

In a clean randomized controlled world, ATE is what you aim for. But we all know the marketing reality, it is messy. RCTs are not possible in marketing.

With the definition behind us, lets understand the Retargeting measurement nuance

📌 Retargeting campaigns

Retargeting campaigns are not random.

They are explicitly designed to target:

– People who visited your site
– People who already saw your ad.
– People who added to cart but didn’t purchase
– People who showed some form of intent

This is in essence is selection by design.

📌 The change of Estimand ATE to ATT

The moment a marketer implements retargeting campaigns, the question changes from “What is the impact on everyone?” to “What is the impact on this ‘high-intent’ or already ‘exposed to treatment’ group.

Hence the estimand changes from ATE to ATT

📌 The counterfactual change.

One tricky part is, when you go from ATE to ATT, the counterfactual also changes.

You can’t meaningfully ask: “What if a low-intent or non exposed user saw this retargeting ad?” Your counterfactual group gets restricted to the people who already received the ads (treatment).

📌 The generalization of ATT mistake in Retargeting

Marketers often take retargeting results and say:

“This channel has 5x ROAS, let’s scale it massively”

But mind you, this result is not for the whole population. This is just applicable for ‘high intent’ or ‘already exposed’ group.

📌 Final Takeaway

Appreciating the nuance of how in retargeting, the ATE changes to ATT leads to correct inferences, better decision making in the form reduced ad spend wastage.

One should not confuse Precision targeting with Population-level impact. Retargeting doesn’t just change who you target. It changes what you are estimating.

P.S: If you are interested in learning causality and causal experiments for marketing, we have an extensive course on the same. Pls check our website -> courses->causality course.