Attribution Part 5 - Multi-touch Attribution Models

In my last post I talked about the simplicity of single source attribution models and how that simplicity leads to bad decision making and ignores the other touch points, making it very difficult to justify marketing budget for those channels.

As the name suggests multi-touch attribution looks at more than one touchpoint along the customer journey. While more accurate than the single touch models, their added complexity can make them harder to implement, harder to use and harder to explain.

U-shaped attribution models are a combination of the first and last touch models. They split 80% the revenue equally across the first and last touch and divide the remaining 20% of the value across the rest of the touch points.

If two points are good, then three must be better is the thinking that goes into the W-shaped attribution model. This model captures both the first and last touchpoint as well as an important intermediate point and splits 75% of the value between these points. Typically used in B2B, the first point is when the lead is captured, the second when the lead becomes marketing qualified and the third when it becomes an opportunity or is sales qualified.

Many businesses find they have more than three important points in the customer journey so add these additional points to the W-shaped model and use the Zig-Zag Model. All of the important points are captured and split 90% of the value. Another good model for B2B companies, it works well in Account Based Marketing (ABM) environments.

Should you not want to exclude any of the touch points, there are models that cater for that too. The first of these is the Linear Model where all touchpoints in the journey get an equal share of the revenue value. This is certainly the most egalitarian approach and is very easy to calculate and explain.

We all know from our own experience however, that not every touch in a customer journey delivers the same value. In fact, things like paid search towards the end of the journey has negative value. You have just paid Google to get an existing lead to your website after spending quite a bit to nurture that lead. Surely, that has a negative value.

The linear model’s strength is also another “supposed” weakness in that it gives the same weight to both recent and distant events. Most marketers would say that the most recent touches are the most important in leading to the sale. The Time Decay Model was created to better capture this. It gives the most weight to the most recent interactions and discounts those further back.

Before using this model, we need to make sure that the most recent interactions actually have the greatest impact on our sales, and we don’t have a recency bias in our thinking. Like last click attribution, this model is great for identifying what is getting customers to convert. However, what if the most recent touches are not the most important but those early interactions are – where the lead has come from, your welcome email program and your lead nurture. If this is the case, flip the model around.

Multi-touch attribution models are more accurate than single touch models but still leave a lot of room for misinterpretation. Also, with more touchpoints to measure and the variability in each customer’s journey, the calculations involved are more complex.