Designing a Multitouch Attribution Model for Online Marketing

Multitouch attribution models are replacing first-touch and last-touch methods of attribution.
Multitouch attribution is used to figure out how the different marketing touches contributed to a customer sale. Multitouch is growing despite the complexity since it both maps to the reality of:

• the customer shopping experience 
•  the vendor’s activities and expenses such as PPC, SEO, email marketing, retargeting, website materials and demos, and social media interactions. 

Multitouch is complicated and there are a choice of models. The choice should be based on what you believe best maps to the customer shopping experience and what decision the company is trying to make.

At the end of the day, each company needs to figure out how to handle attribution.

At the start of the process, a key step is a strategic think on what type of analysis or decisions is being made.

For example, a company might build the attribution model to consider increasing or decreasing the investment in different methods of making contact with customers such as PPC ads on search, PPC on social, banner ads, buying lists, or retargeting. In this analysis, the company’s marketing automation system website with demos and white papers is a fixed cost that is not really up for review.

Here's a quick overview of attribution models, thanks to Jimmy Shang

1.Linear attribution is the simplest. Each touchpoint gets an equal percent of credit. So in a simple example, if the customer, a) clicked on an ad, b) clicked on a retargeted ad, and c) clicked on an email and then bought, each of these three would get1/3 credit for the win. Of course, it’s a little more complicated than this since the interactions on the website could also be included.

2. Time decay gives more credit to the touchpoints closest in time to the conversion. For example, the last email before a purchase/conversion is given more credit than the first organic search*.

3. Position-based / U-Shaped is a hybrid between first- and last-touch attribution. This method puts more weight on the first and last touchpoints, assigning 40% credit to each, and splitting the remaining 20% between the touchpoints in the middle*.

4. W-Shaped credits the first touch, the point where a visitor becomes a lead, and the final touch each at 30%. It divides the remaining 10% among any additional touchpoints. Some advanced multi-touch attribution models leverage machine learning to assign partial or incremental credit to predict the value that each touchpoint added*.

*Thanks to Jimmy Shang of Ad-Roll for his  spectacular article on attribution models. The second through fourth models are directly quoted from him.