Note: The example in this post uses synthetic, illustrative data. The numbers are constructed to show the pattern clearly. They do not represent any specific client account. The pattern itself is real and common.
Here is a scenario that plays out in performance marketing accounts more often than the platforms would like you to know.
You are running Meta ads for a DTC ecommerce brand. Your retargeting campaign is showing a 6x ROAS. Your prospecting campaign is showing a 1.8x ROAS. The account manager recommends shifting budget from prospecting to retargeting. The numbers are obvious. Budget shifts. Retargeting ROAS climbs to 7x. The case for prospecting keeps getting weaker. Budget shifts again.
Six months later, your account-level ROAS looks fine. Your new customer acquisition rate has dropped by 40%. Your retargeting audience is shrinking because fewer new users are entering the funnel. The 7x retargeting ROAS is now claiming credit for users who would have purchased anyway. They were already deep in the consideration cycle when your retargeting ad appeared.
Last-click told you retargeting was working. It was lying about why.
The mechanics of the lie
Last-click attribution assigns 100% of conversion credit to the last ad interaction before a conversion. For most e-commerce and lead-gen accounts, that last interaction is either a retargeting ad (remarketing to past visitors) or a branded search ad (capturing users who were already searching for the brand by name).
The problem is the counterfactual: what would have happened if that retargeting ad or branded search ad had not appeared?
For many users in the retargeting pool, the answer is: they probably would have converted anyway. They visited the website twice. They added to cart. They responded to an email. They were going to come back. The retargeting ad that appeared right before their conversion did not cause the conversion. It was present when the conversion happened, which is not the same thing.
Last-click cannot distinguish between these two populations: users who converted because of the retargeting ad (incremental converters) and users who would have converted without it (organic converters who happened to see an ad).
When you scale a retargeting budget based on last-click ROAS, you are scaling spend on a population that increasingly skews toward organic converters. The ROAS looks great. The incrementality is low.
An illustrative worked example
Take a hypothetical account with the following structure (numbers are synthetic and constructed for illustration):
Setup:
- 10,000 website visitors per month
- 2,000 of those are in the retargeting pool (visited in last 30 days)
- Natural conversion rate from retargeting pool without any ads: approximately 5% (based on email and direct traffic behavior) = 100 conversions
- Retargeting campaign reaches 1,800 of the 2,000 (90% reach)
- Average order value: $100
What last-click reports:
- Retargeting conversions attributed: 150
- Ad spend: $1,500
- Attributed ROAS: 10x (150 conversions × $100 AOV ÷ $1,500 spend)
What an incrementality holdout reveals:
- You hold out 200 users from seeing any retargeting ads (10% holdout)
- Holdout group converts at 4.8% = approximately 10 conversions
- Exposed group converts at 6.0% = approximately 108 conversions
- Incremental conversion rate: 6.0% - 4.8% = 1.2%
- Incremental conversions attributable to ads: 1.2% × 1,800 = ~22 conversions
- Incremental ROAS: 22 × $100 ÷ $1,500 = 1.5x
Last-click reported 10x. The incremental truth is 1.5x.
The gap is not unusual. In retargeting-heavy accounts running on last-click, incremental ROAS is often a fraction of attributed ROAS. The gap depends on how intent-saturated your retargeting audience is: the more you are re-showing ads to users who were already very likely to convert, the bigger the gap.
Where the distortion compounds
The scenario gets worse when you trace the second-order effects.
The account manager, seeing 10x retargeting ROAS vs. 1.8x prospecting ROAS, scales retargeting and cuts prospecting. Less prospecting spend means fewer new users entering the retargeting pool. The retargeting audience shrinks. To maintain retargeting volume, the audience definition broadens, including lower-intent visitors. Conversion rates in retargeting drop. Budget expands to compensate.
Meanwhile, new customer acquisition rate declines. The retargeting pool is now mostly composed of recent visitors, not warm prospects. Last-click keeps showing reasonable ROAS because it is still claiming credit for the organic converters in the audience. The account manager keeps the budget allocation.
Twelve months in: customer acquisition cost for genuinely new customers has doubled. The retargeting pool is a fraction of its original size. The brand has been farming its existing warm audience while starving the awareness funnel that keeps that audience replenishing.
Last-click was never measuring effectiveness. It was measuring presence at the moment of conversion.
A 3-step sanity check
You do not need a full attribution stack rebuild to detect this distortion. Three checks take 30 minutes and will tell you whether last-click is steering you wrong.
Step 1: New customer rate trend. Pull your last 12 months of orders and separate new customers from returning customers. If your new customer rate (new customers ÷ total orders) is declining while overall revenue is flat or growing, you are farming the installed base, not acquiring. Retargeting-heavy last-click accounts show this pattern clearly.
Step 2: Run a holdout test on your best-ROAS retargeting campaign. This is the direct test. Hold out 10–20% of your retargeting audience for 2–4 weeks. Compare conversion rates. The gap between holdout and exposed is your incremental lift. If it is small, your retargeting ROAS is mostly claimed credit, not caused conversion.
Step 3: Track retargeting pool size. Most ad platforms show you the size of your retargeting audience over time. If your retargeting audience is shrinking or not growing despite sustained traffic levels, your prospecting channels are underperforming relative to what the account needs. The pool is the canary.
If all three signals are unfavorable, you have a last-click distortion problem. The fix is not to stop retargeting. Retargeting is real and valuable at the right budget level. The fix is to rebalance the channel mix toward prospecting, using incrementally-tested ROAS to set the budget split, not last-click reported ROAS.
What to do next
The three-step sanity check above is a diagnostic, not a solution. If you find distortion, the next moves are:
- Run holdout tests on your highest-ROAS campaigns to establish incremental baselines.
- Rebalance the attribution model in your analytics stack. GA4 now uses data-driven attribution by default, which is better than last-click (though still not a substitute for incrementality testing).
- Set channel-level budget targets based on incremental ROAS, not platform ROAS.
This is not simple to implement, especially if your team or agency is used to reporting on platform numbers. The conversation is uncomfortable: "the 6x ROAS you were celebrating might be closer to 1.5x in incremental terms." But that conversation is the one that produces a sustainable growth model.
The alternative is a budget allocation model that optimizes for the appearance of performance while quietly degrading the engine that produces it.
How the operating system reads this
Inside the Dynamic Ad operating system, the Analytics & Backoffice layer ships an attribution audit alongside every cycle close: incrementality-tested ROAS, deduplicated server-side conversion source, and a written read of where last-click distortion is shaping the spend ledger. The make-good guarantee runs on retained outcome, not on platform-reported lift, so the comparison stays defensible across the seasonality and attribution-model shifts that the platforms themselves cannot correct for. The flagships (Dynamic Lead OS and Dynamic Commerce OS) carry the read end-to-end against the signal contract frozen at engagement start.
Amit Harari is one of the founders of Dynamic Ad. He writes about AI-driven performance marketing operations at dynamicad.ai/insights.