Truth about Digital Attribution: Statistically inferred attribution trumps first and last click models

ScreenHunter_52 Jul. 29 08.14
Facebooktwittergoogle_pluspinterestlinkedinmail

Marketers are obsessed with identifying the winning marketing touchpoints so they can distribute the budget accurately between the channels. Attribution is important because it gives us the intelligence to improve the return on the campaigns. Attribution makes it easier for us to justify the spend in marketing without over or under spending.

I don’t think there is any question, whether attribution should be used or not. The key issue is choosing the right type of attribution model. We all love choices, and no one understands this better than grocery store chains. They give us hundreds of choices even if for something as simple as drinking water. There is spring water, distilled water, zero calorie water, flavored water, vitamin water and many more options.

Attribution was not invented by grocery store analytics team, but sometimes I feel it must have been their secret strategy. It is extremely important to choose the right attribution for digital marketing measurement. So let’s first understand what are our options when it comes to attribution.

Let’s assume a website Bluewidgetsforsale.com sells blue widgets. On July 9th, a visitor Mr.X landed on Mashable.com and saw a display ad by bluewidgetsforsale.com. Mr. X clicked the display ad and visited bluewidgetsforsale.com. He did not make any purchase and left the website. Next day on July 10th, Mr. X conducted a search for “best blue widgets” on Google. He noticed the website bluewidgetsforsale.com is the #2 organic listing after Wikipedia. He clicks the listing and landed on the bluewidgetsforsale.com site. After spending fifteen minutes browsing and reading content Mr. X closed his browser window. Later in the day, Mr. X opened his browser and typed “bluewidgetsforsale.com”. This time he ended up placing an order for one blue widget.

Here is how the bluewidgetsforsale.com digital marketing team will analyze the sale and apply the attribution models to improve the conversion cycle.

1. Last click attribution – Last click is the most widely used attribution and majority of the measurement tools use this as a default. Last click attribution simply means the last channel or touchpoint before the sales/purchase/order gets full credit.

In our case of Mr. X, Direct channel touchpoint will get the full credit for the order because direct was the last touchpoint.

last click attribution

 

2. First click attribution – Marketers and analyst soon realize giving full credit to the last channel is injustice, and the practice needs to be stopped before the leadership cuts the marketing budget. First click attribution came to rescue, and all analytical tools made first click touchpoint as the new defaul. In this case, Display channel touchpoint will get a full credit for the one order for the blue widget.

first click attribution

 

3. Equal credit or linear attribution – Linear attribution is the simplest of all attribution models. In linear attribution, all of the marketing touchpoints in the conversion cycle gets equal credit irrespective of the influence.

In case of the generous equal/linear credit model, all three touchpoints (display, SEO and direct) will get equal credit for the blue widget order.

equal credit attribution

4. Time lapse or Time decay – Time lapse is relatively newer compared to the traditional attribution models. With the time-lapse model, the marketing channels or the touchpoints nearest to the time of the conversion will get the most credit.

Notice for our example, direct was the closest to the sale so it will end up getting 0.5 followed by SEO with 0.4 and display 0.1 credit.

time lapse attribution

5. Proximity model – Proximity model gives equal credit to the touchpoint which is in the close proximity of the previous touchpoint. To explain this model further, here is a screenshot of Mashable.com. There are two ads by Legalzoom.com on this page (728×90 and 300×250). If a user clicks the 728×90 first followed by 300×250 before making a purchase on legal zoom with proximity based modeling the credit with be equally divided between the two ads due to the close proximity of each other.

ScreenHunter 43 Jul. 27 15

6. Statistical attribution model – The popular attribution models mentioned above do not tell the complete story. You may end up giving more credit to one touchpoint over the other in almost all cases, or you will have to rely on the equal credit model. Instead of relying on manual selection of the attribution model statistical algorithm can be used to assign credits. In fact, some of the best attribution models in the market are based on machine learning instead of relying on manual selection. IBM’s SIRA (statistically inferred response attribution) modeler does a great job in assigning accurate credits to the marketing touchpoints ruling out all the human-induced discrepancy. This leads to true cross-channel attribution and improves marketing efficiency and ROI allocation.

statistical attribution model

In our example of bluewidgetforsale.com, in addition to assigning statistically accurate credits SIRA will also decide the winning touchpoint. Blue widget marketers can immediately identify the winning touchpoint and invest more dollars on the SEO campaign.

Here is a screenshot of an actual SIRA experiment. Row 1 shows in 208 cases, “Natural Search” (SEO) was the last click/touch winning offer but SIRA’s winning offer is “Referring sites” with 95% confidence. Similarly, the “Natural search” channel trumps “Email marketing” in 26 cases or orders placed.

SIRA model

Share your thoughts, feedback, comments and tweets below!

Leave a Reply

Be the First to Comment!

Notify of
avatar

wpDiscuz