Articles on: Modeling

Model Components

Basics



The media mix models trained by Polaris include several components. Each component captures a different performance factor. Those performance factors combine to make up your total country-wide performance.


Organic Factors



Organic performance factors include:

Seasonality (day of week, month of year)
Trends (country-wide, app-wide, category-wide, industry-wide)
Promotions (discounts, product releases, events, specials)
Organic activities (non-paid social media posts, viral product features)

Seasonality and trend components are automatically input into the model. Promotions and organic activities must be input into the model via the Custom Export data sharing method in the marketing dataset.

If the organic demand output by the model seems higher than expected, you may have unmodeled performance factors.


Marketing Factors



All marketing factors are rolled into a single model component and each marketing source is assigned a coefficient when a model is trained. Marketing factors must always be input into the model via the marketing dataset, which consists of daily aggregates of spend and/or impressions for each marketing source at each level of available granularity.


Components File



The components file contains a breakdown of your model’s components. If you were to use the date column to plot the various component columns over time for any specific channel or campaign and metric, you would see how each effect contributes to the overall country-wide performance, like the example below.



In this example, the country is the US and the metric is installs. The yellow line at the top is the relative effect of marketing (called the spend effect in the components file). It seems to have a pretty significant if variable impact on the country-wide US installs. There is also a pronounced, but fairly small day of week seasonality effect (the red line), while the trend effect (blue line) detected a large spike in installs that seemed to be uncorrelated with marketing (spend/impressions) late in the time period (likely some sort of promotion, holiday, or other external factor).

Updated on: 22/08/2022

Was this article helpful?

Share your feedback

Cancel

Thank you!