Skip to main content
Ensemble Model for Clicks
Isobel Hirst avatar
Written by Isobel Hirst
Updated over a year ago

Introducing: The Ensemble Model for Clicks

Gone are the days of using only one model to show you how your click channels are performing - click-based attribution at Fospha is now being measured using an Ensemble model, providing the most accurate attribution possible for click-based channels - and an even more accurate input for our impressions modelling!

The algorithm will continue to use a foundation of Google path and position based modelling from GA4, with multiple Fospha machine-learning models to give you an (even) more accurate view of the impact of click-based paid media throughout the funnel.

What are the benefits?

Though there's no such thing as the perfect model, some models are very valuable - especially when working together. An Ensemble model takes advantage of this fact, by running multiple valuable attribution models simultaneously, combining the results of each model together for a more robust and accurate evaluation of your marketing performance.

Think of it like taking 'wisdom from the crowds' - you're consolidating opinions from many experts before taking a decision rather than one.

Which models are included?

  • Last Click Attribution - Position based attribution which credits 100% of the sale to the last click in the customer journey that is not Direct - GA's standard model

  • Data-Driven Attribution - Click Frequency based attribution that looks back over the last 4 touchpoints in the customer journey, and assigns credit based on how important each click was in the estimated overall decision to purchase

  • Tree-Based Machine-Learning attribution that uses clicks to predict subsequent sales. Predictions are continuously improved as it learns from errors until it reaches the most likely scenario to explain the observed sales

  • Regularized Regression Machine-Learning attribution determines the value of a click by identifying patterns and assigning a weighting to each channel based on their impact on sales, whilst balancing the potential for over-weighting due to outliers

For each of the Machine-Learning models, we're modelling engaged sessions instead of overall sessions, which leads to a more accurate output as there are no bounce sessions being modelled, unlike other attribution providers.

What does an example model output look like?

Each model outputs a different importance score for the clicks for each channel, and we take an average of the outputs to result in the final click-based attributed value - see an example below:

As a reminder, we use the Ensemble model results as the foundational step to establish where click-media should get credit - and where converting clicks have been influenced by prior impressions so that we can re-allocate the credit to the influencing channels such as for any Paid Social, Display, Discovery, and Youtube.

Did this answer your question?