# Control Group **Category:** metrics **Short Description:** A segment of users or data points that receive no treatment or intervention, serving as a baseline for comparison in experiments. **Last Updated:** 2026-07-11T00:00:00Z ## Definition A control group is a randomly selected segment of users or data points that receive no experimental treatment, serving as the baseline against which test groups are measured. In marketing experimentation, control groups enable marketers to isolate the true causal impact of campaigns, creative changes, or other interventions by comparing outcomes between exposed and unexposed audiences under otherwise identical conditions. ## Formula **Formula:** `Lift = (Test − Control) / Control × 100` **Result Unit:** % Percent change in the metric between exposed and unexposed groups — the core incrementality read. Example: a 2.6% conversion rate in the exposed group vs 2.0% in the control is a 30% lift. ## Calculation **Formula:** `Lift = (Test - Control) / Control` **Explanation:** Calculates the relative performance difference between test and control groups. Positive values indicate the treatment had a beneficial effect, while negative values suggest the treatment underperformed the control. ### Components - **Test Group Metric**: Performance metric from the treatment group - **Control Group Metric**: Performance metric from the control group ## Examples - Withholding 10% of users from a campaign to measure true incremental impact - Using geographic holdouts to measure regional campaign effectiveness - Implementing PSA (Public Service Announcement) tests as active controls - Creating persistent holdout groups for long-term incrementality measurement ## Key Points - A control group receives no treatment and provides the baseline that reveals what would have happened anyway. - Comparing an exposed test group to a randomized control isolates the causal, incremental impact of a campaign — not just correlated outcomes. - Random assignment is what makes the comparison valid; without it, pre-existing differences between groups masquerade as treatment effects. - In marketing, the control is usually a holdout — a slice of the audience deliberately withheld from ads, or a set of geos left untreated. - Incrementality measured against a holdout is often lower than platform-reported, last-click results, because attribution credits conversions that would have happened without the ad. ## Best Practices - Ensure sufficient sample size for statistical validity - Use true randomization for group assignment - Maintain clean separation between test and control groups - Document all external factors that might influence results - Consider using multiple control group types for robust validation ## How AdSights Helps **Tracking Control Group:** A control group only works if exposure is clean and the comparison metric is stable. AdSights helps teams design holdouts by surfacing which creatives and audiences have enough volume for a valid split, then tracks variant-level lift against the control once the test runs. When platform-reported ROAS diverges from holdout-measured lift, AdSights makes the gap visible — so budget shifts follow incremental impact, not attribution artifacts. ## FAQs ### What is a control group? A control group is a randomly selected set of users, or markets, deliberately kept from receiving an experimental treatment — an ad campaign, a creative change, or an offer — so it can serve as a baseline. By comparing outcomes in the treated (test) group against this untreated control under otherwise identical conditions, you can isolate the true effect of the treatment from everything else that influences results. ### Why are control groups important in marketing measurement? Because they answer the question attribution can't: what would have happened without the ad? Much of the conversion volume platforms credit to advertising would have occurred anyway — loyal customers and in-market buyers who'd have purchased regardless. A control group reveals the incremental lift a campaign actually caused, protecting budgets from being justified by conversions the ads didn't drive and exposing where spend is genuinely additive. ### What is a holdout group? A holdout is the marketing form of a control group: a portion of the target audience (or a set of geographic markets) intentionally excluded from a campaign so their behavior shows the no-ad baseline. Comparing the exposed audience against the holdout gives a clean, experiment-based read of incremental lift. Holdouts can be short-term for a single test or persistent, maintained over time to continuously measure a channel's true contribution. ### What's the difference between a control group and A/B testing? They're closely related. A/B testing compares two or more active variants (A vs B) to see which performs better, and the control-group principle is what makes it valid — one arm often serves as the baseline. A pure control group specifically receives no treatment at all (versus a variation of it), which is what enables incrementality measurement: A/B testing asks 'which version is better?', while a true control asks 'did the treatment do anything at all versus nothing?' ### What makes a control group valid? True randomization (so the only systematic difference between groups is the treatment), sufficient sample size for statistical power, clean separation so the control isn't accidentally exposed, and stable, comparable conditions across both groups. Documenting external factors that could influence results — seasonality, other campaigns, market shocks — also matters. A control group compromised by self-selection, contamination, or too little volume produces lift estimates that look precise but aren't trustworthy. ### What is a control group in advertising? In advertising, a control group is the slice of your target audience — or a set of matched geographic markets — deliberately withheld from a campaign so its behavior shows what would have happened without the ads. Comparing conversions in the exposed group against this holdout isolates the campaign's true incremental effect, something attribution models can't do because they credit conversions that would have occurred anyway. ### How do you set up a control group for an ad campaign? Randomly withhold a slice of the target audience — commonly 5–10% — or hold out a matched set of geographic markets, then run the campaign normally for everyone else. Check that the holdout is large enough to detect the effect size you care about, and keep the separation clean for the full test window so control users are never exposed. When the test ends, compare the two groups with the lift formula: Lift = (Test − Control) ÷ Control. For full methodology, including geo-holdout design, see the Incrementality Testing Guide linked in the resources on this page. ## Related Terms ### Similar Terms - **[A/B Testing](/resources/glossary/creative/ab-testing)**: Experimental methodology requiring control groups - **[Incrementality](/resources/glossary/general/incrementality)**: Measurement approach using control groups to isolate true impact ### Component Terms - **[Statistical Significance](/resources/glossary/metrics/statistical-significance)**: Determines reliability of test vs. control comparisons ## Related Resources - [Incrementality Testing Guide](/resources/guides/incrementality-testing-guide) - How to structure geo and audience holdouts for causal measurement. - [Marketing Incrementality Calculator](/resources/tools/calculators/marketing-incrementality-calculator) - Estimate incremental lift from test vs. control conversion rates. ## Featured in topic hubs - [Experimentation & Statistics](/resources/topics/experimentation)