Why your retargeting ROAS is lying to you: the truth about lift

Are you paying for sales you would have received for free? Many B2C brands find that up to 60% of retargeting conversions happen organically, even without ad spend. If you rely solely on platform metrics, you are likely overestimating your impact and wasting a significant portion of your budget.

To scale effectively, you must understand the difference between incrementality and ROAS. While return on ad spend (ROAS) often looks impressive for retargeting, it rarely reflects the true causal impact of your media investment. Understanding this distinction is the difference between scaling a sustainable business and simply subsidising existing demand.

What is retargeting incrementality?

Incrementality measures the causal lift in sales directly attributable to a specific marketing activity. It answers a fundamental question for media buyers and marketing strategists: Would this purchase have happened anyway? In a B2C context, retargeting campaigns often target high-intent users who have recently abandoned a cart or visited a pricing page. Because these users are already close to converting, platform attribution models credit the ad for the sale.

However, a deep base versus incremental sales analysis frequently reveals that many of these users would have returned to the site via organic search or direct traffic regardless of the ad. When you fail to account for this baseline, you risk over-investing in channels that intercept existing demand while under-funding the channels that create new demand.

The over-attribution trap in B2C marketing

Most digital platforms suffer from self-attribution bias. By using long view-through windows, they claim credit for any user who saw an ad and later purchased. This creates a significant problem for CEOs, CFOs and CMOs who need to justify marketing spend based on real growth.

Frequentist approaches to measuring lift

The Frequentist methodology relies on controlled experiments to establish ground truth. These tests provide a snapshot of performance under specific conditions and are essential for measuring advertising effectiveness with statistical rigor.

User-level lift tests

Platforms like Meta and Google offer native lift studies that split your target audience into a test group exposed to ads and a control group that is held out. By comparing the conversion rates between these two groups, you can calculate the lift. This is currently the gold standard for proving causality at a specific point in time.

PSA and ghost ads

In environments where native lift tools are unavailable, marketers use Public Service Announcements (PSAs) or ghost ads. In this setup, one group sees your actual brand ad while the control group sees a generic PSA. This ensures that both groups are equally biddable, providing an accurate comparison of user behavior without the bias of differing auction dynamics.

Geo experiments

For brands with significant offline presence or those facing privacy-related tracking limits, geo experiments are highly effective. You hold out specific geographic regions from retargeting and compare their performance against similar regions where ads are running. This is a common strategy for brands operating across Scandinavia and the Baltics to validate regional media impact.

Bayesian methodologies for deeper insight

While Frequentist tests give you a clear answer on lift, Bayesian approaches provide a more nuanced view of uncertainty. This is particularly useful when developing a marketing budget allocation strategy within the context of a marketing mix modeling (MMM) framework.

Bayesian models do not just provide a single point estimate. Instead, they produce probability distributions that allow you to quantify risk. For example, rather than stating a static ROI, a Bayesian model might conclude that you are 90% confident the incremental ROI is between 2.8 and 4.2. This approach allows you to incorporate results from previous lift tests as informative priors, making the model more stable when dealing with the noisy data typical of smaller markets. For technical teams, building for example PyMC marketing models allows for the integration of these sophisticated statistical techniques into your measurement stack.

Reducing ad waste with econometrics

To eliminate over-attribution, retargeting metrics must be viewed through econometric and data science methods. By using a multivariable model that accounts for seasonality, pricing, and macro-economic factors, you can isolate the true incremental contribution of every euro spent.

This process involves identifying diminishing returns in marketing to find the saturation point where the next euro spent on retargeting delivers less value than the last. Once these curves are established, you can use media budget scenario planning to simulate what would happen if you shifted 20% of your retargeting budget to awareness-driving video ads. If your platform reports a 5:1 ROAS but your econometric model shows a 2:1 incremental lift, you apply this correction factor to ensure your tactical decisions align with reality.

Achieving a hybrid measurement framework

The most sophisticated B2C organisations do not choose between attribution and incrementality. Instead, they use a MMM versus multi-touch attribution (MTA) hybrid approach. You use MTA for daily tactical optimisations and creative tweaks, while MMM provides the strategic, high-level, budget allocation and causal proof needed for long-term planning.

By reconciling these two views, you can slash ad waste and ensure your marketing budget drives genuine business growth. This holistic approach ensures that decisions are backed by over 90% predictive accuracy, bringing together isolated data points into a clear strategic vision. To move beyond the limitations of standard platform reporting, consider how a managed software approach can pinpoint where every penny boosts growth most. Analytical Alley offers mAI-driven media strategies that combine AI computing power with human insight to help you achieve your business goals through smart, calculated decisions.