Comparing five attribution models to understand channel contribution

01Business question

Many marketing teams judge channels using last-click reports from their ad platforms. That raises a practical question: how much of a channel's apparent performance comes from the attribution model rather than the channel itself?

02Dataset & context

Dataset: Synthetic / Demonstration Dataset. This portfolio analysis uses a structured synthetic marketing attribution dataset created for analytical demonstration. It contains 11,292 touchpoints from 3,500 customers across seven channels, with channel, timestamp, conversion and revenue fields. It does not represent a PA Data Analytics client or real client performance.

03Approach

Build five attribution models on exactly the same journeys, so that any difference in results comes from the model and not the data, then compare each channel's share of credit, its implied return on spend and the budget each model would recommend.

04Methodology

• Data quality treatment, including removal of 198 duplicate tracking events (pixel double-fires) before attribution.
• Journeys reconstructed from touchpoint data with Pandas.
• Rule-based models: Last Click, First Click, Linear and Time Decay.
• Shapley value attribution: each channel's average marginal contribution to conversion across the channel combinations observed in the data.
• Revenue-share, ROAS and budget-reallocation comparisons by channel and model.

05Modelling & analysis

In this dataset, 84% of journeys involved more than one touchpoint, so the choice of model affects most of the revenue.

Dataset finding: Paid Search received 14.4% of credit under First Click and 41.9% under Shapley, a 27-percentage-point spread on identical data. Email showed a Last Click ROAS of 163x but received no independent credit under Shapley. That pattern is typical when high-intent customers are already on the email list: email is often the last touch without being the reason for the purchase.

06Results within the dataset

Model output. The analysis demonstrated that different attribution methods can produce materially different budget-allocation scenarios. Under the project's annualisation assumptions, the difference between the Shapley-based and last-click allocation scenarios was approximately $4.84M per year.

Those assumptions are: the same weekly paid budget in both scenarios, each channel's revenue valued at its Shapley ROAS, ROAS held constant as spend changes, and 52 campaign cycles a year. Because real channels saturate, the figure is directional. This is a model-derived demonstration result based on synthetic data. It is not revenue generated by a PA Data Analytics client and should not be interpreted as an observed business outcome.

07What this analysis demonstrates

That attribution is a modelling choice, not a fact. The same data can make a channel look indispensable or marginal. Bottom-of-funnel channels such as email, retargeting and branded search tend to look strongest under last-click, while prospecting and awareness channels tend to be under-credited.

08How a business could use this analysis

• Compare at least two attribution views before reallocating marketing budget.
• Treat model outputs as directional, and confirm large shifts with holdout or geo tests.
• Reconcile ad-platform conversions with GA4 before comparing channels.
• Judge top-of-funnel channels on assisted conversions and new-customer acquisition, not ROAS alone.

09Related service

This project demonstrates methods used in our marketing analytics services.

10Apply this to your data

If you are facing a similar question, we can look at what your data can support and what a useful answer would look like.