Marketing Attribution for 2024

We’ve written quite extensively about multi-touch marketing attribution using clickstream data. The topic is as relevant now as when we first wrote about it in 2019.

We still see a near complete reliance on last-touch attribution for assessing marketing channel performance and making investment decisions. The exception to this is when using the ad networks’ self-reported attribution. Certainly in the cases of Google Ads, Linkedin, and Meta, attribution data is often over-reported: the same single transaction could be reported as a whole conversion for all three networks.

Attribution & Google Analytics 4

The move to GA4 has advanced the cause of marketing attribution in two ways:

1. There is now a “data driven” option within the “attribution” section. This allows a direct comparison between last click and data-driven models. 

2. Raw GA4 data, at the event (clickstream) level, is now fully exportable into Google Big Query.

The opportunity from the former is to assess quickly a truer value of each channel’s value. For example in the screenshot example we can see differences between the models indicating a channel’s value higher up the conversion journey.

Example of Attribution Model Comparison in GA4

Using BigQuery to Export Attribution Data

Here at Deducive we use the Big Query exports as our main data source for attribution. Using these exports gives us the ability to analyze far beyond what is available in the GA4 interface.

• Inconsistencies in base GA4 data, especially with respect to UTMs, can be cleaned up

• More flexibility in grouping ad / source / medium / campaign parameters into channels

• Data enrichment by joining with ecommerce and CRM systems

• Different data-driven models including Markov and Shapley

• Predictive modeling of media budgets

• Deeper understanding of conversion journeys

The Deducive Attribution Approach

In engineering terms, our attribution system is written in R and Python and runs entirely on Google Cloud, with the analysis generated in the R & Python ChannelAttribution Pro package.

Results are as a simple table and can be visualized in any tool. We typically do this in Google Looker Studio.

Example of Channel Attribution in Looker Studio

Seeking a Single Source of Truth is Folly

Even with the quality of the data in the GA4 export, there is still a need to interpret the data in attribution reports — there is no single number that is an absolute truth.

But when different models and sources are presented together, better decisions can be made.