Part 2: The many names and faces of digital attribution.

Following on from Part 1 - which covered the different tracking implementations available across digital channels - this post gets into the attribution models themselves, exploring how credit is assigned to different touchpoints in a user journey, and why the model chosen can produce very different pictures of the same campaign. 

Attribution Models

First click

Assigns 100% of the credit to the first interaction in the user journey. Useful for identifying which channels are generating initial interest and tend to work well as a measurement tool for brand or awareness campaigns, where the goal is introducing the brand rather than closing a sale. 

Last click

Assigns all of the credit to the final interaction before conversion. Still one of the most widely used models, largely because it's simple and produces clean-looking numbers. The problem is that it tells you nothing about what happened before that final touchpoint - if a user spent three weeks being influenced by upper-funnel activity before converting via a Meta retargeting ad, last-click gives Meta all the credit and everything else nothing. This information can then unjustly influence budgeting decisions for future campaigns, at the expense of the channels that actually built the intent to buy. 

Linear

Distributes credit equally across all touchpoints in the journey. Less commonly used, but provides a more balanced picture of the full funnel and gives upper-funnel channels a fairer share of the reported value. 

Time-decay

Gives more credit to touchpoints that occurred closer to conversion, based on momentum and the influence of repeated brand exposures over time, indicating that interactions immediately preceding a purchase are more significant than earlier ones. A reasonable model for shorter sales cycles where momentum matters. 

Position-based (U-shaped)

Assigns the highest credit to the first and last interactions in the journey, with the remaining credit distributed equally across everything in between. The logic here is that the first and last touchpoints are the most valuable in regard to building awareness and driving conversion, while mid-funnel activity serves primarily to maintain presence and reinforce intent. 

Data-driven

Increasingly the default model across major platforms, data driven attribution (DDA) uses machine learning to assign credit based on actual performance patterns rather than a fixed rule, providing a more nuanced view of channel impact. Requires larger datasets to be effective, so smaller campaigns may not produce reliable results under this model. 

Incremental (holdout testing)

Attempts to measure only the conversions that actually occurred because of campaign activity - excluding any that were likely to have happened anyway. Incremental attribution does this by excluding a portion of the audience and comparing conversion rates between the exposed and unexposed groups. Most commonly associated with Meta's Conversion Lift studies, though the methodology exists across other platforms and third-party measurement tools. 

Data-driven attribution is generally the most accurate model on the list - but it comes with a caveat worth sitting with. In most cases, the model is owned and operated by the platform being measured. Google's DDA is scoring Google's own channels. Meta's is scoring Meta's. There is an inherent and fairly obvious conflict of interest in that arrangement, and while it doesn't mean the numbers are wrong, it does mean they probably shouldn't be taken entirely at face value.  

Beyond platform-level attribution

One option worth knowing about, particularly for advertisers running activity across multiple channels, is paid for cross-platform attribution tools that sit outside the platforms themselves. These typically use server-side tracking to stitch together the user journey across channels - connecting an impression on one platform to a click on another and an eventual conversion elsewhere - and produce a single, consolidated view of performance that no individual channel's reporting is able to provide on its own. They're not cheap, and they require a solid tracking foundation to be useful, but for businesses with enough budget and complexity to justify it, they offer something the platforms can't - a picture of the full journey that nobody has a financial interest in making look better than it is.  

Part 3, the final part of this series, looks at how to assign actual value to each stage of the funnel - turning the tracking and attribution data from the first two posts into something that resembles a usable picture of ROI.