Marketing attribution analysis part 2: Embrace the complexity

Kent Stones | December 6, 2012

In October, I wrote about the changing world of marketing attribution analysis and why it is important to invest time and resources in this complex form of analysis. I’ve received some great questions and comments, many of them along the lines of “Are you making it more complicated than it needs to be?” or “Are we overthinking it?” My answer to these questions is an unqualified “No!”

In my morning scan just today, the tweets, blogs and articles written about marketing fragmentation were too numerous to count. Consumers have more ways than ever to interact with a brand, frequently using multiple channels on the same shopping journey. The thousands of shopping path permutations that can now occur have far outstripped our ability to quickly and easily understand and assess them, let alone quantify the influence each of the touch points (or a combination of them) has on the overall path. And yet we are expected to demonstrate a return on our marketing investment.

I made the suggestion last month that the way to garner the knowledge and insights we need, given these fragmented paths, is by employing a more advanced analytical approach than is being used today. Easy to say, I know, and if your company is like ours, you may not have that skill set residing in house. So how do you accomplish this? In our case, the solution was to partner with the Data Scientists at Analytic Marketing Innovations: Kurt Seemar, Trich Schultz and Jill Giele. And since I don’t have the knowledge to be able to explain these new multi-touch modeling approaches (yet!), I asked Kurt, Trich and Jill if they would answer a few questions about how these more advanced multi-touch models work.

Can you recap why a multi-touch model is better than a single-touch model?

Jill: Marketers have historically relied on “single-touch attribution,” an attribution method that assigns credit for a sale to either the first touch or the last touch. The method was never perfect, but is now considered by most to be seriously flawed because there are so many more potential touch points that may be contributing to a sale.

Kurt: All of this led to a recent push in the marketing analytics world to find a way to assign a value (credit) to various touch points along the pathway, based on a “multi-touch attribution” algorithm. A common approach to these newer multi-touch models has been to assign a weight to each step on the pathway (either equally or subjectively), but this is still problematic because it doesn’t really tell us what we need to know – the true difference of influence among different touches (or combination of touches) – or predict what the next best touch point should be for the individual consumer who interacts with a brand.

That’s why at Analytic Marketing Innovations, we designed a unique multi-touch attribution model that tells us what we need to know by employing an analytical method called the Markov Chain Model.

So, just what is a Markov Chain Model?

Trich: The Markov Chain Model (named after mathematician Andrew Markov) is a method for assigning value to any number of steps along the customer journey (as long as that number is finite).

Markov Chain Model Diagram

When data scientists employ a Markov Chain Model for multi-touch attribution, we can see the likely chains of events along the various consumer pathways to a sale. The Markov Chain Model assigns credit to each touch along the path to conversion while also sequencing and predicting each step.

All of this, of course, can be quite complicated, because with the increasing fragmentation in how consumers interact with a brand, there are more than just a few pathways to a sale. Using the Markov technique, it is possible to find and understandevery statistically significant path, which can add up to thousands of models (paths) for a marketing campaign. That would have been difficult to accomplish using traditional tools and techniques, but is very possible with this approach due to the way the data is managed, integrated and analyzed.

Why is the Markov Chain Model better than other models?

Trich: The Markov Chain Model is more effective because it uses superior, proven statistical methods to derive the value for each touch on the customer journey, as well as what combination of touches will be most effective. That knowledge allows more informed, objective decisions (rather than just assigning equal weights or subjective judgments), which in turn leads to marketing efforts that are more effective and quantifiable.

Kurt: Importantly, the models can be used to optimize marketing throughout the execution cycle. Attribution results can be applied in real time because they are relevant at the individual level. Because of this, we can truly optimize the customer journey by predicting the best next step for each interaction an individual may have. For example, if we know that a person who has received an email and then clicked on a banner ad is much more likely to buy if s/he receives another email, the marketing team can then make sure that an email is sent. In other cases, if we can see a certain marketing path is creating fatigue, the team can pull back that type of marketing for some customers.

Are there particular types of brands that are better suited to this kind of approach than others? 

Jill: Any brand that is executing multi-channel marketing and is committed to learning more about how to make their campaigns more efficient and effective will benefit from this type of marketing attribution analysis. But commitment is key, because it does require an ongoing devotion of time and effort to establish and then use the models.

Typically, it’s best to have three months of data to start the analysis. The first step is setting up tracking on a campaign and working with various vendors, including a Data Management Platform (DMP) vendor, which helps combine the online and offline touches. The second step is gathering data for three months and then building models. The final step is analyzing the resulting output to identify the top paths leading to conversion and allowing the development of marketing heuristics for the remainder of the campaign.

In an environment of shrinking marketing budgets, there is considerable pressure on marketing departments to demonstrate quantifiable results for each channel, and good multi-touch marketing attribution analysis can be invaluable in this type of assessment. The insights not only improve results overall, but they also help in shifting budget dollarsduring a campaign from unproductive channels into those that ARE contributing to sales.


Want to receive our weekly blog insights? Subscribe now.