Is Your Retargeting Incremental or Just Credit?

Your retargeting campaign reports the highest ROAS in the account, so it gets more budget every month. But ROAS is an accounting entry, not a measure of cause. The uncomfortable question every operator eventually has to answer is whether that campaign is creating sales or just claiming sales that were already coming. Much of the time, nobody checks — because checking is harder than reading the dashboard.

For the neighboring economics, compare Reporting Window vs Optimization Window, Honestly and use Express Checkout Without Going Blind to Your Meta Pixel to validate the measurement decision.

Attributed ROAS measures credit, not cause

When someone adds to cart, browses your site, then buys two days later, a retargeting ad almost certainly touched them along the way. The attribution model sees that touch and assigns the conversion to the campaign. The campaign’s ROAS looks spectacular. None of that tells you the person wouldn’t have bought anyway.

This is the core confusion behind the question “is retargeting incremental.” Attributed revenue answers “which campaign was nearby when the sale happened.” Incrementality answers “which sales would not have happened without the spend.” Those are different numbers, and on warm audiences they can diverge enormously. A campaign can post a 12x attributed ROAS and have a true incremental ROAS close to break-even, because it spends most of its budget reaching people who had already decided.

The mechanism is simple once you name it. Retargeting pools — site visitors, add-to-carts, past purchasers — are pre-selected for intent. You built that intent with prospecting, organic, email, and the product itself. Retargeting then serves an ad to those high-intent users and harvests the conversion. The delivery system is doing exactly what you asked: find the least expensive conversions. The least expensive conversions are the ones that were going to happen regardless. That’s not the algorithm cheating. It’s the algorithm optimizing toward credit because credit is what you rewarded.

Why the optimizer compounds the problem

Meta’s auction optimizes for the event you select, and it finds the lowest-cost path to that event. For a warm audience, the lowest-cost path is the user already three-quarters of the way to checkout. The system has no concept of “would have converted anyway” — incrementality is invisible to it. So the better your optimization gets, the more efficiently it concentrates spend on inevitable conversions, and the better your ROAS looks while your true contribution flatlines.

Advantage+ Shopping (or any broad, signal-rich campaign) sits in the same trap, just less visibly. It blends prospecting and retargeting under one objective and reports a single blended ROAS. Because it can freely reallocate toward whoever is least expensive to convert, it naturally leans into existing demand unless you constrain it. The headline number stays strong. What you can’t see from the report is the mix: how much of that revenue is genuinely new versus harvested. A high-performing Advantage+ campaign and an over-credited retargeting campaign fail the same test for the same reason.

The tell-tale symptoms

You don’t need an experiment to get suspicious. Watch for these patterns:

• ROAS that scales suspiciously well as you cut budget. If you halve spend and ROAS rises while total conversions barely move, the campaign was harvesting, not generating.
• Retargeting revenue that tracks your prospecting volume with a lag. When warm-audience sales rise and fall as a function of how many new visitors you sent in, retargeting is downstream demand capture, not a demand source.
• Frequency climbing with flat incremental volume. Showing the same near-converters more ads doesn’t manufacture new buyers; it just re-bills the same ones.
• A blended MER that won’t move no matter how you reallocate inside the funnel. If shifting budget between prospecting and retargeting leaves marginal efficiency roughly unchanged, you’re rearranging credit, not changing outcomes.

These are signals, not proof. Proof requires withholding the ad and watching what happens.

How to actually test incrementality

There is only one honest way to measure incremental lift: deny the ad to a comparable group and compare. Three practical methods, in rough order of rigor:

1. Platform conversion-lift / holdout study. The cleanest in-platform option. A randomized share of your eligible audience is held out from the ads, and the platform compares converters in the exposed group versus the holdout. The output you care about is incremental conversions and incremental ROAS — not attributed ROAS. Run it long enough to clear your typical purchase cycle, and size it for enough conversions that the lift estimate isn’t noise (treat a few dozen incremental conversions as a rough floor for a readable signal, not a assurance).

2. Budget step-down test. If you can’t run a formal holdout, cut the retargeting budget by a meaningful amount — 40 to 50 percent — and hold it for two to three full purchase cycles. Watch total account conversions and blended MER, not the campaign’s own ROAS. If total sales hold roughly flat while the campaign’s reported revenue drops, the missing revenue reattributed elsewhere — it was never incremental. If total sales genuinely fall, you found real lift. This is crude, vulnerable to seasonality and other moving parts, but it’s directionally honest and costs you nothing but patience.

3. Geo or audience split holdout. Suppress retargeting for one matched slice of your audience and keep it live for another, then compare conversion rates between the two. Harder to run cleanly, but it isolates cause better than any attribution report ever will.

Whatever method you choose, change one thing at a time and give it enough conversion volume and enough calendar time to read. An underpowered test that “shows no lift” after four days has told you nothing.

Reading the result honestly

When the numbers come back, hold yourself to the incremental figure, not the comfortable one:

Signal	Likely read
Attributed ROAS high, incremental lift strong	Real demand generation — fund it
Attributed ROAS high, incremental lift near zero	Re-billing existing demand — cap or cut
Total conversions fall when paused	Genuinely incremental
Total conversions hold when paused	Credit, not cause

The goal is not to kill retargeting. Some of it is genuinely incremental — abandoned-cart recovery, reactivating lapsed buyers, countering a competitor’s late pitch. The goal is to know which slice earns its budget and to stop letting one inflated ROAS number set your allocation.

This is exactly the kind of question worth pointing an analysis layer at. Bach reads the same delivery and unit-economics signals you do, but frames performance around contribution rather than attributed credit — and it stays read-only until you approve any change, so a flagged “this looks like harvested demand” is a prompt for your judgment, not an automatic budget cut.

The takeaway

Treat attributed ROAS as a reporting convenience, not a verdict. Before you pour more budget into your highest-ROAS retargeting or Advantage+ line, run one honest test: hold the ad out from a comparable group, or step the budget down hard, and watch your total conversions and blended MER instead of the campaign’s own number. If the account doesn’t notice the cut, you weren’t buying sales — you were buying credit for sales you’d already won. Fund the lift you can prove, and stop paying twice for demand you already created.