Viewing Attribution Results After Reloading Spot Logs

DescriptionThis article outlines why your attribution results may change after reloading pre- or post-spot logs to your XP platform.


Overview

Each time you reload previously uploaded data, the platform attribution results will change.

This happens because when you reload your data, your data is reprocessed and remodeled, and attribution is re-calculated based on the latest information.

For example, you may have uploaded pre- or post-log files to your platform and then received additional information on airings, but you want to reload files to include this information. Every time you reload previously uploaded data, the platform results will change.

Note: This does not mean you have lost any data, but results may change due to this reattribution. As we get more information, the response attribution can shift based on the new information.

Reasons for attribution changes

Spots can change in three different ways after reloading:

  • Extra spots: If additional spots are present in the spot log, these may generate additional attribution, but they may also share the credit previously attributed to other spots. For example, if you upload planned pre-spot logs and the networks do not run everything as initially planned, the reconciled spot logs will result in changes.
  • Fewer spots: If a spot is removed from the spot log, you can no longer attribute sessions to it. You may not lose everything associated with that spot, as other spots may gain some of the credit that was initially attributed to the spot.
  • Airing time of spot: If the time at which a spot is aired changes, the audience and behavior of that audience will be different, which will impact attribution.

Other factors that can affect results:

  • Upscaling: Because a sample is used for initial attribution, the changes resulting from combinations of the scenarios described above can be magnified when upscaling is applied.
  • Channel mapping changes: If any are applied, the results will be affected.
  • IP verification: If there are OTT impressions where an IP is not verified the first time and, therefore, is not attributed, we later get information that verifies the IP and include it in the attribution. This could also result in the appearance of a "lost" response, as it would then get the attribution it deserves.

Example Scenario

  • Initially, 100% credit for a response was given to an OTT impression.

  • Linear logs were then uploaded, showing three impressions that should have also been credited for the response.

  • That initial OTT credit is now fractionalized across the OTT impression and three linear impressions the platform did not have information on previously. The OTT impression did not "lose" credit, but the credit was spread to all the impressions that fell into that attribution window.

Note: This can also apply to a linear impression credited 100% for a response.

For example, one spot was uploaded for Network A, after which it received all response credit. Then three days later, a spot log was uploaded for Network B. Network A may have fewer attributed conversions because it is now sharing credit with Network B for some. The overall attributed outcomes may increase because Network B influenced new events.

Need more help understanding how we give credit when we attribute?

InnovidXP’s attribution approach uses a time decay model, which considers the influence of all exposures on behavioral response. This means we give higher weightings to the closest touchpoint, but we take into account all earlier touchpoints (i.e., impressions or ad exposures) and weight them algorithmically based on impact likelihood.

How our time decay approach works:

  1. Firstly, we link every web session to any impressions served to a unique household within the attribution window of 7 days to identify "view throughs".
  2. For platforms leveraging our control groups feature, we estimate the proportion of each view that should be considered incremental.
  3. We then weight each view through/incremental portion shared amongst multiple impressions based on a time decay weighting, where we assign more credit to impressions closest to the web session. See Apportioning Credit Across Impressions and Platforms for a deeper dive into how this works.

Related content

Approach to Linear TV Attribution

 

Was this article helpful?
0 out of 1 found this helpful