Digital Advertising

Everything You Wanted to Know about Attribution Modeling but were Afraid to Ask

Scott Ashby  In my last contribution to The Makegood, I touched on some of the challenges that we face within the programmatic industry that are impeding our transition from efficient advertising medium to effective advertising medium. Today I’d like to highlight another important data-focused lever within that dynamic: Attribution Modeling.

The importance of determining how credit for sales and conversions is applied based on the influence of different touch points within the conversion path is widely acknowledged. It’s something that we’ve been talking about within the industry for a while, but, on the face of it, no one really seems to have it figured out. Because no one seems to have it figured out we’ll continue to debate the correct implementation of a universal attribution model until the cows come home. This seems futile to me (having witnessed and participated in such debates), as there is no such thing as a universal attribution model. The reason that there is no universally applicable solution to the problem is that the multiple variables that constitute the Attribution Funnel are dynamic and non-linear.

Let me explain what I mean by outlining some high level steps in the creation of an attribution model:

  1. Identify the channels that you want to track (Sponsorships, Display, Email, Search etc.)

  2. Map out the consumer conversion path

  3. Determine how much ‘value’ to assign against each touch point

At first glance these 3 steps seem simple enough to implement, but in practical application steps 2 and 3 become very complicated. Mapping out the consumer conversion path could have infinite iterations given that path to purchase is non-linear, and assigning value against each touch point within the conversion path should differ for each individual conversion path itself. These challenges are further exacerbated when you take into consideration other contributing factors when determining attribution such as decay rate and ad timing. When you then start to think about additional external factors of uncertainty (like the influence of media exposure that you cannot track) it wouldn’t surprise me if you either booked yourself a two week vacation in the wilderness to escape the problem, or you went on-line to see if there were any marketing events that you could attend in the vain hope of finding that universal attribution model. The Silver Bullet as it were.

Here’s where my experience of working in a data-driven business can help – attribution modeling is not a Silver Bullet, it’s a process by which marketers can gain a deeper understanding of consumer behavior and educate themselves on which elements of the media mix are most effective at driving a specific action. If you don’t have an attribution model and you’re reading this article then that’s a start. If you do have a working attribution model that’s producing even a modicum of conversion insight then you should give yourself a round of applause. It’s all well and good to look to the market for inspiration and expose yourself to case studies which can help you develop your own model, but expecting that some of the third parties who have a view on attribution modeling will ride in on their white horse and save the day isn’t going to happen. From my experience these third parties are working to rule-based models which function on certain assumptions that probably won’t transpose effectively into your model. It’s important to acknowledge that developing a robust attribution model is an iterative process, and that tweaking variables like conversion path, media channel, weighting and attribution value through each iteration is the only way you’ll arrive at a successful model. Develop, Implement, Test and Repeat should be your mantra.

The two most important factors in the accuracy and success of custom Attribution Modeling are the methodology used to process data and the quality of the data itself. As programmatic / automated buying starts to extend out of Web, Tablet, Mobile and into TV, Radio and Outdoor, the data produced across multi-channel will grow in quality, whilst becoming increasingly more accessible. As the process takes place the industry will grow in maturity and its analytical capability will not only increase, but analysis will become the pivot of the majority of media-centric conversation. As things stand at the moment, the media industries over-use of last-click attribution due to its ubiquity and convenience is making performance media and search advertising look significantly more effective than brand activity in regard to driving conversions.

As technology accelerates and data quality increases, that perception will steadily start to erode, and it will be those companies that invested in the development of robust attribution models that will be at the forefront of that movement.

Scott Ashby is the Managing Director of MEDIA iQ, North America. After an eight year stint at Microsoft, he joined MEDIA iQ in 2012 and is responsible for building, launching and developing their performance trading operation in the US and Canada. This is his second contribution to The Makegood.



  • Jay Friedman

    This is great in theory. The problem is that current attribution companies make it all but impossible to implement their product, with “data scientists” requiring near perfect amounts/kinds of data to release their results. I think we can all agree that almost any model that credits more than last click/touch is better than last click/touch alone. The current attribution companies don’t see it that way. A client of mine recently joked, “I’m going to start an attribution company. I’m going to charge half of the lowest priced provider out there and then not deliver results, claiming the data isn’t sound. Clients will get the same results but only pay half!” In the end, we built our own. It’s not nearly as sophisticated as the best ones out there, but it gives us instant insight into something better than “awful”, which is better than 99% of the industry right now.