How is Brand Lift Calculated?
Upwave uses an exposed-control methodology to calculate brand lift.
This begins by gathering a sample of respondents who were exposed to a campaign and then finding comparable control respondents based on basic demographic factors like Age and Gender.
Upwave then use propensity score methods to weight the control sample to be comparable to the exposed sample across all demographic and behavioral factors targeted explicitly, or implicitly, by the campaign. We use state of the art machine learning methods to build a custom propensity score model for balancing the exposed and control samples within each cut. These methods automatically determine which factors need balancing for each sample, allowing us to estimate accurate baselines for all cuts, no matter how different the baselines vary from cut to cut.
This ML-based methodology delivers significant benefits over manual, "bucket" weighting methodologies (RIM weighting, raking) common in traditional research.
Greater Accuracy: Using ML methods to model and weight the control results in a more accurate and precise measure of lift that avoids erratic outcomes that result from manual weighting methodologies.
Daily Automated Reads: ML methods are automated, and eliminate the need for "mid-campaign reports" by automating the full re-modeling and re-weighting of all cuts every night. When manual weighting methodologies use dashboards, the "weights" are updated nightly, but the features (the "buckets") being weighted only change when there is a "mid-campaign report".
Cut-Level Controls: ML methods enable to creation of cut-level controls for cuts that are based on media exposure, such as audience, frequency and creative message. Manual weighting methodologies rely on overall controls for these cuts that make it impossible to measure the causal impact of a specific audience, frequency band or creative message.