Incrementality testing: the only honest way to measure retargeting.

Last-click attribution flatters retargeting. View-through attribution flatters it even more. The only measurement that tells you whether your retargeting is actually generating revenue — instead of just being credited for revenue that would have happened anyway — is a properly designed holdout test.

Every retargeting vendor will show you a ROAS number. Most of those numbers are wrong. Not because the math is wrong — but because the math is answering a question other than the one you should be asking.

The question retargeting reports almost always answer: "Of the people who saw a retargeting ad, how many converted, and what's the revenue from those conversions divided by the spend?"

The question you actually want answered: "Of the additional revenue we're generating beyond what we would have generated without this retargeting layer, how much is attributable to it?"

The first question has a comfortable answer. The second question is the one that decides whether your retargeting budget is earning its place.

Why last-click flatters retargeting structurally

Retargeting, by definition, targets people who have already demonstrated intent. They visited your site. They added to cart. They're already on a path to conversion — sometimes a very high probability path. When you serve them an ad and they convert, last-click attribution credits the ad with the whole sale.

But many of those conversions were going to happen anyway. The retargeting ad was the last billboard before the destination, not the reason for the trip. Last-click can't distinguish between the two, and so it systematically over-credits the bottom of the funnel — and especially over-credits retargeting.

What a holdout actually does

A holdout is a deliberately under-served control group. You take your retargeting audience, randomly assign (say) 10% of it to never see a retargeting ad, and let the campaign run as normal for the other 90%. After a meaningful window, you compare conversion rates and revenue per user between the two groups.

The difference is incrementality. Specifically:

• Lift = the % difference in conversion rate between exposed and holdout groups
• Incremental revenue = (exposed revenue per user − holdout revenue per user) × exposed audience size
• Incremental ROAS = incremental revenue ÷ retargeting spend

Incremental ROAS is almost always lower than last-click ROAS. Sometimes dramatically lower. That's not a problem — it's the truth. The retargeting was generating real value, but less than the last-click report suggested.

Designing a clean test

A few things go wrong frequently enough that they're worth flagging upfront:

• The holdout has to be random. Not "users in markets we're not active in" — that's a different audience. Random user-level assignment within the same audience pool.
• The window has to be long enough. Conversion windows for retargeting are typically 7–14 days; the test window should comfortably exceed that.
• Cross-channel leakage has to be acknowledged. A holdout user can still see your search ad, your email, your push. A holdout isolates the incrementality of retargeting, not of marketing as a whole.
• Statistical significance is not optional. A 2% "lift" that's within the margin of error is not a result. Test until you have enough volume to see a real signal, or accept that you can't prove the result either way.

Beyond holdouts: ghost ads and geo splits

Holdouts are the gold standard but not the only option. Ghost ad testing serves a different message to the holdout group, controlling for the placebo effect of being served any ad. Geo splits turn retargeting on in some markets and off in others — useful when user-level assignment isn't possible, but messier because markets aren't truly comparable.

Each method has tradeoffs. The right one depends on volume, identifier coverage, and how much measurement rigor your stack supports. The wrong one — the one to never accept — is "trust the last-click report."

The hard truth most agencies won't tell you

If you've never run a holdout against your retargeting, the most likely outcome of doing so is that your reported ROAS will fall. Sometimes by a lot. That's not a reason to skip the test. It's the reason to run it.

Knowing the real number lets you make real decisions. You can spend more confidently where the incremental ROAS justifies it. You can cut spend where it doesn't. You can stop the bleed on campaigns that look healthy on the dashboard but aren't actually generating new revenue.

The vendors that will resist this conversation are the ones whose business model depends on you not running it. The vendors worth working with will design the test for you and tell you what to do based on what comes back.

The takeaway

If your retargeting can't survive a holdout, it isn't working — it's just visible. Run the test. The version of your performance marketing that comes out the other side will be smaller, sharper, and honest about what it does. That's the version that compounds.