Introduction:

We've significantly improved the conversion attribution logic for personalizations in Spotler Activate. These changes result in faster loading times, more accurate attribution, and a more intuitive user experience. This documentation should provide more insights into the changes we have made

Previous Implementation (Before 5.14.0):

• Attribution was based on PersonalizationView and PersonalizationClick events within the selected date range in the overview. This meant the date period directly affected the attribution window.

• Main statistics were calculated only for visible personalizations in the overview.

• The "Last Interaction" attribution model was implemented in a way that could attribute the same conversion multiple times to different personalizations if a user interacted with several before converting. This led to inflated conversion counts and was one of the main pain points our clients had.

• The underlying code was complex.

Problems with the Previous Implementation:

• Slow Performance: Calculating attribution based on the selected date range led to long loading times, especially for larger date ranges or datasets.

• Inaccurate Attribution: Attributing conversions multiple times led to inflated conversion metrics and made it difficult to accurately assess the true impact of each personalization.

• Confusing Logic: The complex code and the way the "Last Interaction" model was implemented made it difficult to understand how attribution was being calculated.

Visual presentation of Previous Implementation according to the “Last Interaction” Model:

This implementation gets all views/clicks and closes of personalisations and then fetches the Attribution events that are in the future. This diagram shows when the Attribution window for PersonalizationClick is set to 7 days, PersonalizationView 1 day, and we have selected a period of 7 days:

Open

New Implementation (After 5.14.0):

• Attribution is now based on the specific attribution event type set in the personalization settings (e.g., Purchase). This provides a clear and consistent attribution window based on the actual conversion event.

• A separate function was created for calculating main statistics, decoupling it from the attribution logic.

• Conversion attribution and click/view statistics are now handled separately, reducing confusion and improving performance.

• The "Last Interaction" model is now implemented correctly: only the last interacted personalization before the conversion is attributed.

• The code has been significantly refactored for improved efficiency.

Benefits of the New Implementation:

• Dramatically Improved Performance: Loading times have been reduced by up to 20x. This allows for faster analysis and quicker insights.

• Accurate Attribution: Conversions are now attributed correctly, providing a more accurate understanding of personalization performance. This leads to better decision-making about which personalizations are most effective.

• Clearer and More Logical System: The simplified logic and separation of concerns make the system easier to understand and use.

Visual presentation of New Implementation according to the “Last Interaction” Method:

This implementation gets all the Attribution events of the selected date period (7 days) and then fetches the (last) PersonalizationClick 7 days prior, PersonalizationView 1 day prior:

Open

Example:

To illustrate the difference between the old and new methods, we have the following example:

• Old Method: A user clicked a personalized banner on Monday. If the reporting period was set to Monday-Wednesday, a purchase on Thursday would not be attributed to the personalization, even if it was a direct result.

• New Method: A user clicks a personalized banner on Monday. A purchase on Thursday will be correctly attributed to the personalization.

Conclusion:

These changes represent a significant improvement to the conversion attribution system in Spotler Activate. We are confident that these enhancements will provide a better experience and more accurate insights for our users.

Explanation of all attribution models implemented:

• Last interaction: Gives 100% credit to the interaction occurring most recently before conversion (this interaction still has to occur within the attribution window).

• First interaction: Gives 100% credit to the interaction first done in the attribution lookback window.

• Participation: Gives 100% credit to all unique interactions with personalizations. The total number of conversions is inflated compared to other attribution models. Participation deduplicates personalizations that are interacted with multiple times.