Attribution is a word that gets used on the data side of online marketing, but unlike other engineering-speak, this is not a term that should be dismissed. After all, it’s very important. “Attribution” is the process by which you are told which ads (the stimuli) lead to which conversions (the results).
Without attribution, how are you to know which ads, which sites, and which marketing offers are most compelling to your potential customers? Because this process is the foundation of the reporting that governs multimillion dollar media-buying decisions, it’s worth a closer look.
In the early days of online advertising, the attribution methodology (i.e. the logic by which ad X is given credit for causing purchase Y) was fairly straightforward. The most recent click that a user made on any of that advertiser’s ads was given full credit. Doubleclick, Inc. made this the de facto standard for reporting success of online marketing campaigns. In the world of tracking sporadic ads across a sea of sites, this made sense and worked fairly well.
It was not long before people realized the limited insight allowed by basic click-to-conversion metrics. Thus, the next evolution of attribution methodologies incorporated the “post-impression” counts. This is a measure of when users saw but did not click on an ad and then converted. When this was not enough, advertisers started asking for customized limits on the time span between the user’s click (or view) and their subsequent purchase, along with other custom processing requests. These efforts were little more than groping around in the dark, but they were all symptomatic of the same problem: Attribution methodologies were still inadequate. As we’ll see, this wouldn’t change.
Payment Structures for Media
Moving ahead in time, the Internet bubble and struggling economy choked the flow of advertising dollars. Advertisers wanted guaranteed results before they would commit money to online spending. Cost-per-acquisition (CPA) cost structures for online media buys began to dominate the landscape, which created an ecosystem wherein each media outlet jockeyed to get the credit for each customer purchase. If a site gets the credit for a conversion, they get the commission. Because advertisers wouldn’t agree to pay media sites based on their own reporting, third-party systems remained necessary adjudicators of reports. And as we know, the attribution methodology is the foundation of these reports. Cleverly, media networks realized that they could exploit the algorithms of these third-parties in order to get the most credit for themselves. They correctly knew that in order to get credit, they had to be the most recent ad stimulus for the user prior to their conversion. Consequently, there was a direct financial incentive for media companies to use a shotgun approach to show ads to as many people as possible, on the chance that it was the one right before they made a purchase, so they would get credit even if there was no causal relationship.
Stepping Back to Move Forward
Consider the basic problem. The goal of attribution methodologies is to determine the cause of a desired effect. Because we’re talking about humans and not machines, it is an exercise in determining the most likely cause that drove a human action. How many things in your life do you do because of a single stimulus, much less a single piece of advertising? I’m guessing very few. Most people act based upon multiple events that occur over some stretch of time. Even a basic decision, such as trying a new restaurant, might be caused by the combination of seeing a colleague with a takeout bag from there, reading an article about the chef, and having a friend mention that they enjoyed the food.
The process of assigning credit for a conversion to some advertising interactions should reflect how people work, not how the log files work.
The process needs an overhaul. For starters, let’s abandon the outdated notion of assigning credit on an all-or-nothing basis and give partial credit to multiple small influences. Furthermore, let’s apply some weights to how much credit each interaction receives. Once you accept the possibility of partial credit, it reveals subtle influences in the success of your advertising campaigns. It also minimizes the effect of click fraud and conversion stealing when it comes to evaluating media sites and networks.
The attribution methodology drives the reporting data. The reports drive the evaluations of media and ad effectiveness. The evaluations drive where advertisers and agencies invest their future online dollars. Considering that advertisers spent $22.7 billion in online advertising in 2009, they should demand a better system.
In the new age of real-time bidding for ad impressions, where convenient roll-ups, such as page, site, and network lose their prominence, having a methodology to correctly attribute conversions to the appropriate stimuli is going to be even more important in determining which audiences are worth re-buying.
It may be unrealistic to expect every reporting system to overhaul its fundamental algorithms overnight, but this analysis can be done independently, given the raw data. Admittedly, issues such as scalability, data storage, and computing power all come into play when dealing with the massive amounts of data generated within even a single advertising campaign. When these obstacles are overcome and the attribution methodology is another lever that can be manipulated by marketing and analytics teams, that is how you gain a real advantage over your competition.