Exploring alternative attribution models – it’s probably somewhere on your list of 101+ things to do. So, here is a summary of the different attribution models available.

1. Last-click

Regardless of what happens earlier in the conversion path, 100% of the revenue credit goes to the last click prior to a transaction or conversion. The hard work done to get the user on-site that first time, or coax them back for a 2nd or 3rd visit before conversion is ignored. It’s all about the last action taken.

2. Last-click non-direct

A modification of the previous model, this assumes that a direct visit which leads to a conversion was a mere formality. The decision was made and the conversion was inevitable. The user just needed to get back to the site to wrap things up. So, 100% of the credit goes to the previous channel (Google Analytics uses this as its default model of attribution).

3. First-click

The inverse of last-click – this gives 100% of the credit to the first-click. This ignores any work done by other channels to nurture and eventually convert a lead – it’s all about that first interaction.

4. Linear

Increasing in sophistication, Linear gives some credit to all platforms involved in a conversion. A slightly more nuanced approached, but one that is still based on assumptions.

5. Time-decay

Another step up in sophistication, time decay gives to credit to multiple channels, but weights the credit based on recency. So a paid search ad that leads to an immediate conversion will generate a large amount of credit for the Paid Search channel, but other channels involved will receive some credit, which will reduce the further back those interactions were.

6. U-shaped / position-based

U-shaped typically gives 40% of credit (each) to the first and last channels, with the remaining 20% split between all other channels. Intuitively, it’s a model that seems to make sense – getting leads into a funnel and eventually converting those leads are two critical stages. However, reasonable as it sounds, it is still a model that relies on assumptions. The 40-20-40 split is a completely arbitrary weighting. Why not 42-16-42, or 38-24-38?

7. Algorithmic

Algorithmic is the only one of these 7 models that is rooted in data and statistics. Algorithmic attribution models factor-in converting, and non-converting paths, to statistically determine the true impact of every channel in generating a conversion. There are a few different models of algorithmic attribution – we use one based on Markov Chains. The result is a custom model which accurately measures the impact of each channel. These models require maintenance (as the data changes, so does the model).