Method and system to account for timing and quantity purchased in attribution models in advertising
Within the field of advertising, a data-driven solution is proposed that enables attribution to use a two-stage model, to account for (1) the timing of a purchase occasion, and (2) a prediction of the quantity purchased. This allows for an accurate forecast of quantity purchased, which is meaningful to media teams and advertisers. Attribution is the process of assigning credit to marketing activities for their contribution towards driving a desired marketing result.
1. A computer-implemented method for accounting for the timing of and quantity of consumers' purchases in attribution models, the method comprising:
identifying consumer activities that represent purchase events with quantity variation;
obtaining media event data and consumer action data;
merging the media event data and consumer action data to create an attribution dataset, the attribution dataset including a path to a purchase event including timestamped media touchpoints and one or more of tracked consumer activity and conversion events;
calculating a quantity variation metric based on purchase data along at least one dimension of interest;
splitting the attribution dataset into a training portion and a testing portion;
fitting an attribution model on the training portion based on the quantity variation metric and a forecasted variation metric produced using the attribution model;
determining accuracy of the attribution model using the testing portion by assessing a difference between the forecasted variation metric and the quantity variation metric;
re-fitting the attribution model if the difference between the forecasted quantity variation metric and the quantity variation metric exceeds a threshold; and
using the attribution model to forecast purchased quantities and generate a recommended media campaign budget.
2. The method of claim 1 , wherein the at least one dimension of interest comprises an individual.
3. The method of claim 1 , wherein the at least one dimension comprises a household.
4. The method of claim 1 , wherein the at least one dimension comprises a product.
5. The method of claim 1 , wherein the at least one dimension comprises a timeframe.
6. The method of claim 1 , wherein the at least one dimension comprises a combination of at least two dimensions.
7. The method of claim 1 , further comprising removing outliers in the purchase data based on their distribution.
8. The method of claim 1 , further comprising removing outliers in at least one calculated metric.
9. The method of claim 7 , wherein the removing outliers is done according to criteria comprising confidence intervals, percentiles, or numerical thresholds.
10. The method of claim 7 , wherein the removing outliers is based on a distribution of quantity variation metrics or their significance.
11. The method of claim 1 , wherein the fitting the attribution model on the attribution dataset includes using an accelerated failure time survival model.
12. The method of claim 1 , wherein the attribution model includes a linear regression forecast model.
13. The method of claim 1 , wherein
the media campaign budget includes a plurality of campaigns, and
the generated recommended media campaign budget includes changing a proportion of spend on the respective campaigns.