Incrementality in Remarketing: Key Methods Â
Defining Incrementality and Its Importance for Remarketing
Incrementality in marketing measures the additional effect advertising campaigns generate beyond what would occur without advertising. Simply put, incrementality answers the question: How many sales or conversions occurred precisely due to advertising, rather than naturally?
Marketers often notice that simple attribution models don’t prove causality—correlation doesn’t equal causation. Incrementality, however, is measured through A/B testing: audiences are randomly divided into a test group (which sees remarketing ads) and a control group (which doesn’t see ads), and then outcomes are compared. Such tests are also known as uplift tests.
Why is this crucial for remarketing?
In remarketing campaigns (targeting previous users), it’s often hard to discern if customers returned because of ads or spontaneously. Incrementality testing isolates the true ad effect, showing real added value. It helps evaluate true ad ROI and excludes “free” conversions that would occur regardless of ads.
Understanding real campaign uplift (the percent increase of key metrics in the test group vs. control) allows marketing teams to optimize budgets and strategies based on proven effectiveness, accurately forecast results, and avoid ineffective spending.
Main Methods of Measuring Incrementality
Several methodologies exist for running remarketing incrementality A/B tests. They differ mainly in handling the control group and balancing measurement accuracy against cost. We’ll cover four primary methods: Intent-to-Treat (ITT), PSA, Ghost Ads, and Ghost Bids, their essence, pros, and cons.
Intent-to-Treat (ITT)—borrowed from medicine and sometimes called a holdout test—is a basic incrementality method. It’s simple: we randomly tag a percentage of users as control (no ads shown), while the test group sees ads as usual. Results from the entire test group (including both exposed and unexposed users) are compared against the ad-free control group.
• Easy to implement and understand. Simply divide and withhold ads from part of the audience.
• No additional costs for the control group, making ITT relatively inexpensive.
• Noise from unexposed users within the test group: Not all test-group users actually see ads (due to losing auctions or low activity). This dilutes differences between groups, complicating accurate effect measurement.
• Requires sufficient coverage (about 40-50% exposure) in the test group for statistically significant results, which can be challenging or costly.
• Temptation to compare only exposed users versus control leads to selection bias, distorting results.
Public Service Ads (PSA) Method
PSA (Public Service Ads)—the placebo approach—shows unrelated neutral ads (e.g., Red Cross banners) to the control group, while the test group receives targeted remarketing ads. Thus, both groups are exposed, but the control sees irrelevant content.
• Completely eliminates noise by comparing only truly exposed users from both groups.
• Easy to implement technically, using existing ad platform tools.
• High costs with no ROI from control-group spending on irrelevant ads.
• Risk of incorrect comparisons due to optimization differences between PSA and remarketing ads, potentially skewing audience segments.
• Doesn’t measure competitive displacement effects (competitors’ ads potentially influencing the control group).
Note: PSA tests are excellent for one-time high-precision incrementality measurement, if budgets allow.
Ghost Ads—a sophisticated “magic trick”—combines ITT and PSA strengths. The control group doesn’t see actual ads, yet incurs zero costs. Instead, the platform tracks users who would have seen ads if they were in the test group by simulating auctions without real impressions. Users in control groups who hypothetically win auctions are marked as “ghost-impressions”.
• Zero control-group costs and minimal noise, effectively providing perfect experimental conditions.
• Reduced selection bias due to natural randomized auction dynamics.
• Technical complexity and opacity (“black box”) for advertisers. Deep platform integration required, with advertisers forced to trust the platform’s auction-based reporting.
• Limited applicability in niche remarketing segments lacking competing bids, making ghost impressions challenging.
• Difficulty of independent verification: advertisers must trust platforms fully.
Ghost Bids, developed explicitly for remarketing incrementality (popularized by Remerge in 2019), adapts Ghost Ads principles specifically for narrow audience segments. Unreachable users (those not appearing in advertising channels) are first filtered out. Remaining reachable users are split randomly into test and control groups. The test group receives ads normally, while control users don’t see ads but have their potential impressions (“ghost bids”) recorded.
• Practically no additional control-group costs—no ad spending for the control group.
• Significant noise reduction through pre-filtering unreachable users.
• Well-suited for narrow remarketing audiences, not reliant on competing advertiser presence.
• Ideal compromise between ITT’s simplicity and PSA/Ghost Ads’ accuracy, making it suitable for ongoing incrementality tracking.
• Requires specialized technology or platform support (e.g., Remerge) to monitor ghost bids.
• Slight residual noise: some control users still wouldn’t have seen ads even if in the test group, slightly diluting effect measurement.
• Communication complexity to non-technical stakeholders unfamiliar with ghost bidding concepts.
Conclusions and Recommendations
Incrementality testing is essential for understanding remarketing’s true effectiveness, shifting from simplistic attribution toward scientific measurement. Methods vary significantly in complexity, cost, and accuracy, making careful method selection crucial:
• ITT: Ideal for initial, inexpensive incrementality checks or limited budgets. Provides general insights but risks statistical noise.
• PSA: Best for maximum precision when budget allows a one-time high-quality measurement; difficult for long-term use due to high costs.
• Ghost Ads: Theoretically ideal, practical only when working with large platforms capable of simulation. Perfect for broad campaigns, less so for narrow remarketing.
• Ghost Bids: Currently the most practical choice for consistent, accurate incrementality measurement in remarketing. Suitable for specialized platforms and narrow segments.
In summary, using any incrementality measurement surpasses none. Start simple, verify if your campaigns create uplift, then progressively adopt more precise techniques as your expertise and budgets allow.