Wrapping this week up with another data and analytics guest blog post by Chichi Okeke, a Statistician on the Measurement and Analytics team at AMP Agency. Chichi provides some great information about attribution models and how they can be applied to strategic planning. The theme next month is UX and Design, interested in writing? E-mail me at taylor [at] mitx [dot] org.
Chichi Okeke is currently a Statistician on the Measurement and Analytics team at AMP Agency. He has been working on advanced analytics and modeling solutions for many clients including Hasbro, Philips, Coldwell Banker and Lifestyles. He has created statistical indexes, probability, attribution and cannibalization models as well as in depth high value behavior analyses. Chichi has a Bachelor’s degree in Economics from Massachusetts Institute of Technology (MIT). He is passionate about data and all things quantitative.
Depending on who you ask, there is a long-running, and possibly fabricated, statistic that consumers see hundreds of brand messages each day. Regardless of the actual numbers (some say 200, others say 20,000) we can all agree that each day we are bombarded by brand messages in all forms. So what is a media planner or a statistician on the Measurement & Analytics team at AMP Agency supposed to do when trying to determine just which one of those messages led to an actual purchase? Getting to an answer typically relies on developing complex attribution models that requires more math than I’d like to discuss. But before we get into any of these complex scenarios, let’s first discuss what attribution modeling is. In its simplest form, it’s merely giving credit where credit is due.
Let’s say you’re trying to buy a pair of sneakers. You first see a TV spot about a pair you are considering during your favorite TV show. Two days later you check out the website to get more information. The next day, you see a Google search ad about the sneakers so you end up clicking on it and purchasing that day. What gets the credit for your purchase? Do you give credit to the TV placement, the sneaker website or the search ad? How much credit do you ATTRIBUTE to each?
It’s a common marketing question that has several possible solutions. There’s last click attribution, where the last step gets all the credit (the search ad). There’s first click attribution, where the first step would get all the credit (the TV placement). There’s also linear attribution where you give equal credit to each touch point in the process. Avinash Kaushik put together a wonderful post detailing some of the various types of attribution models.
Each of those various types of attribution models has its benefits. They can be cost effective, easy to understand or easy to implement, but how accurate are they really? It’s almost as if you’re just arbitrarily giving credit based on nothing concrete.
This is where econometric attribution modeling comes in. It is a customized statistical approach in order to assign credit leveraging large amounts of data (some may even go as far as to describe it as Big Data… gasp!) using modeling techniques not unlike those seen in predictive or media mix modeling.
Now you’re saying “Hey Chichi, I’ve heard about that kind of stuff, but it’s too complicated for me. I want to use something I understand.” Don’t worry, I’ve got you covered. The title of this post implies I’m going to explain how it works, so let me give it a shot.
This econometric approach essentially looks at trends over a long period of time and can attribute impact based on fluctuations in both the channels and the final conversion. Have I lost you already? Let’s look at some pretty pictures with fake data. The graph below shows display investment, TV investment and POS data (Point of Sale or Sales data) trended over the days of a year. The goal of this exercise is to accurately attribute sales credit to TV and display investment. I’ve purposely broken the chart up into four quadrants to explain the basics of how econometric attribution modeling works.
Now, for my fellow statisticians and the like out there, this is merely an example for illustrative purposes. There are many more factors that can influence sales that are not accounted for in this image, but the methodology is still very similar. Channels that are heavily “correlated” with success should get the majority of the credit for that success. Just because a channel was first (or last) doesn’t mean that it is best and deserves all the credit. A statistical approach like this can accurately give credit based on real data.
Econometric attribution modeling has its limitations because you need a good amount of data on your customers to even begin. So here at AMP we use it with our clients more for strategic planning than in-market optimization. However, despite its limitations, it is still more accurate and actionable than most of the standard attribution approaches out there. So when your boss asks you who he should thank for improving his business by introducing econometric attribution modeling, don’t forget to attribute some of that credit to this post.