Increasingly, savvy search marketers believe that the traditional approach of giving conversion credit to the last ad click is a flawed attribution method. Therefore, many of you have investigated or invested in technology that allows you to track beyond the last ad that is clicked so that you can perform attribution over the team of ads that lead to a conversion.
With this new tracking in place, you then need to determine the correct attribution models, and it is at this point that if you are like most marketers, you get stuck. Why do we get stuck? We quickly realize that the argument could be made for many different attribution models.
For example, consider the following purchase path: a user clicks on an ad for “running shoes,” then clicks on an ad for “Nike Shox,” followed by clicking on an ad for “Nike turbo 7″ and makes a purchase. One way to attribute in this scenario would be to give equal credit to all three ads. Another could be to give 20% of the credit to first ad, 30% to the second ad and the final 50% to the final ad. Or a third model could be the reverse, where 50% goes to the first ad then 30% to the second and 20% to the final ad. The point is that we can go on for a while with many different models.
So Which Model Is Correct?
Marketers can find attribution very frustrating when observing the models above because there doesn’t seem to be a clear cut mathematical solution to what the right answer is. At this point, most marketers then begin to question if attribution is all that it’s cracked up to be given that there doesn’t seem to be any accurate way to solve the correct attribution models. Attribution models take some very complicated mathematics to develop.
Recently, we conducted a webcast titled ‘Measuring the Immeasurable’ that featured our partner, Vetra Analytics. Vetra is a statistical consultancy made up of PHDs in statistics and mathematics that can utilize advanced mathematical modeling to create attribution models. When solving for attribution, one needs to determine the “influence potential” of each ad click, impression and site visit.
To determine this potential, one needs to consider many factors, including but not limited to the timing of the ad, decay rate of the ad, if a conversion was made, what products were sold, the amount spent, was it a first time buyer or repeat buyer, etc.
To determine the influence potential, a model or models need to be built which will help to predict the consumer decisions as accurately as possible. The statistical models account for one of the most difficult things for marketers to grasp: uncertainty. Uncertainty is all of the factors that may go into a buying decision that are not possible to measure, even with your advanced technology.
For example, did a friend recommend this product to the consumer? Did they see a billboard or TV commercial? Or did a sales person in the store influence their purchase decision? By accounting for the uncertainty and building a model that incorporates these factors, we can test the model on a go-forward basis.
If the attribution in the model matches the actual results over a sufficient sample size and period, we then know the model is mathematically sound. If the reality does not match the model, we then know the model is not optimal and should be recalibrated. There are solutions to attribution management that will change the ways in which you manage campaigns.
If you are serious about solving attribution because you recognize that accurately attributing credit to your ads will allow you to make more effective media buys, which ultimately lead to greater profits for you and your clients, then you will need to utilize a tool set that implements advanced modeling. Always remember that a technology is only as good as the people behind it.
When implementing an attribution solution, make sure you have the staff that understands how to calibrate it or that the vendor you choose offers services to assist you in building sound models. Your success with attribution is solely dependent on your ability to employ accurate models.
Opinions expressed in the article are those of the guest author and not necessarily Search Engine Land.