The short answer is no.
Incrementality measurement using causal inference for continuous marketing requires a multi discipline approach, combining several analytical methods to make smart strategic decisions. Measuring incrementality in an actionable way offers marketers with operational and tactical insights to improve marketing efficiency.
Incrementality testing methods typically need the marketer to perform an action specifically for the sake of the test – creating split audiences, turn off advertising, or pay providers to run surveys, or pay media vendors for serving PSA ads.
INCRMNTAL uses a method that requires no changes.
We are using an algorithmic approach using machine learning and AI. At the core of our approach is a method called causal data science.
Causal Inference uses a reactive scientific approach to find the causality in the marketing activities as they run, while they run.
Causal Inference enables continuous incrementality measurement.
Causal Inference creates a branch in time examining what happened against what should have happened.
The prediction range acts as confidence (%) where anomalies in the marketing data beyond the bounds of the prediction show if certain marketing activities are producing incrementality or cannibalization.
Marketing is a dynamic field where change is the only constant. Separating external factors (“features”) with internal factors requires an algorithmic approach to measurement, unlike attribution.
After denoising the data, a causal inference model will create a prediction range for every marketing activity tracked, as well as the overlapping influence between one activity and another. Adding the need to consider all external and inventory factors, leads to the creation of millions of predictions in order to increase the confidence level of the predictable range.
(Example: New Inventory Source Cannibalizes Organic Conversions)
While the model itself is reactive – the marketer’s strategic response (i.e. pausing the campaigns with the new inventory source) provides the model with a new input, thus further calibrating the model itself. i.e. continuous.