Overview

Broadly, there are two types of data imports:

Historical: 3 to 12 months of historical data, aggregated daily
Ongoing: only new data since the last sync, aggregated daily

There are also two categories of data.

Marketing: spend and impressions
App events: performance metrics

The marketing and app events datasets are quite different in terms of dimensions and metrics. It should be noted that each data category can be shared using a different method.

Considerations

There are some important points to consider when importing data for the purposes of media mix modeling and incrementality measurement.

Country Consistency

Our media mix models are partitioned by country so it is the most critical dimension. First, country values must be consistent across the marketing and app events datasets. For example, if Great Britain is represented as “GB” in the marketing dataset, but “UK” in the app events dataset, the model will not be able to allocate the "UK" app events to any campaigns since the spend and impressions are in a different country called "GB".

Country Allocation

It is important to scrutinize any data that cannot be allocated to a single country. This can apply to both the marketing and app events datasets, but is most common for marketing data. Some channels, in certain situations, will not be able to report spend or impressions at the country level. They could be completely unreported or allocated to a "blank" or "unknown" country.

If the spend or impressions are significant, they must be split per country in some logical, repeatable way. This often means the custom export method must be used so that country splits can be applied to the data prior to the import into Polaris. Some common methods to split spend and impressions by country include:

Allocate fully based on targeted country (this only works for campaigns that target a single country)
Allocate proportionally by estimated reach (this works by assigning static "reach scores" to each country or country tier and evaluating countries targeted by each campaign)
Allocate proportionally by last touch installs (this introduces an implicit dependency on last touch)
Allocate linearly (split spend and impressions evenly across countries targeted by each campaign)
Allocate fully to each country that could possibly be targeted (this duplicates the spend and impressions across all countries targeted by each campaign, which to the model, will be effectively the same as the linear option, but this method can be easier since there's no splitting involved, but total spend and impression metrics in Polaris will look high)

Brand Marketing

In this context, brand marketing includes any digital and offline marketing activities such as TV, influencer marketing, etc. These activities must be represented in the marketing dataset just like any other marketing activity in terms of spend and/or impressions. When brand marketing is not platform-specific (e.g., TV ads normally promote both iOS and Android versions), it should be included in both app's marketing datasets since it can impact the performance of both. It can be included fully in both or 50% in each so that spend totals are still accurate across apps.

When spend/impressions are not available on a daily basis, they should be divided evenly across the campaign period or modeled in another more accurate way if possible. For some types of brand media, like TV, exact impressions are unknown, but another reach metric is available like gross rating points (GRPs). If so, any reach metric can be supplied as impressions. Brand marketing data can classified under any channel or campaign names desired. For example, a channel could be called "Influencer" while campaigns are named according to the influencer and contracted time period.

There are 2 ways to share brand marketing data:

Data Partner Method - Ensure complete brand marketing data is available in your data partner's platform (e.g., MMP or cost aggregator service) and use the data partner method to share marketing data (see the Data Partner subsection of the Methods section below for more information)
Custom Export Method - Join all marketing data including brand and performance/direct response (often pulled from a data partner) into a single marketing dataset (ideally via an automated process/script) and use the custom export method to share marketing data (see the Methods section including the Custom Export subsection below for more information)

Please note that all marketing data must be shared via a single method for each app. For example, performance marketing data can't be shared from an MMP while brand marketing data is shared via custom export. In that scenario, it is usually easiest to utilize the Custom Export Method for sharing brand marketing data, which is detailed above.

Methods

Both historical and ongoing data can be shared with us in one of two ways: through a data partner such as an MMP or via custom export from an internal system. It is possible to share each dataset via a different method. A dataset is defined as a type (historical or ongoing) and category (marketing or app events) of data. When sharing data via multiple methods, always keep the following considerations in mind.

Each dataset must be shared using a single method (e.g., you can't share ongoing marketing data via both custom export and a data partner, but you can share historical marketing data via custom export and ongoing marketing data via a data partner)
Each dataset must be complete (e.g., you can't share historical marketing data partially via custom export and partially via a data partner; you'd need join the data from those two sources yourself and then share the full dataset via custom export)
Each dataset should have matching dimensional values (e.g., if you share marketing data via custom export and app events data via a data partner, you should ensure your custom export contains channel and campaign names that match your data partner; this is optional, but if the dimensions don't match, marketing data and incrementality metrics will always be separated from last touch data)

Data Partner

Sharing data with us via data partners is the easiest option when it is possible. It only requires you to share the proper API keys or tokens with us. Our existing integrations will take care of the rest of the work automatically for both historical and ongoing data imports. See the articles below for the exact API keys or tokens necessary for each category of data and data partner.

AppsFlyer
Adjust
Singular

Custom Export

In cases where sharing via data partner isn’t possible or preferable, data can also be custom exported from your internal systems and shared with us via a secured S3 bucket. This method allows for the most flexibility in terms of transforming data prior to the import.

Find out more about Custom Exports.

Data Validation

After the data is shared and imported into Polaris, it is important to validate that the data in Polaris was imported successfully, is accurate according to your source of truth, and adheres to the considerations above. More details are available in the Data Validation help article.

Updated on: 28/07/2022