March 16, 2009

Mega-Metrics: Attribution Combinations

Oh those online marketers ... looking to attribute orders to the right marketing channel.

Allocation / Attribution is a largely futile effort. We simply cannot get in the mind of the customer, can we?

We can test. Testing works really well in the direct marketing world, but testing for attribution purposes works less well in the online marketing world.

So here's what you can try ... I've had success with this.
  • Create what I would call "combinations". When a customer visits your site on March 3 via paid search, and purchases on March 5 via an e-mail campaign, you create a "combination" that captures the first touch and the last touch. In this case, the combination is "PAID SEARCH / E-MAIL".
  • If the customer only had one visit (say from an affiliate marketer), then the combination is "AFFILIATE ONLY".
  • The pundits will jump in and suggest that we're missing everything that happens in-between first touch and last touch. Let them jump in and argue.
  • Pay CLOSE ATTENTION to what happens on the next purchase. What are the combinations that are most likely to happen NEXT? Does "PAID SEARCH / E-MAIL" lead to "E-MAIL / PAID SEARCH"? Or does it lead to "PAID SEARCH / E-MAIL"? Or does it lead to "ORGANIC ONLY"? The mega-metric to calculate is the percentage of folks who migrate to different combinations in the future.
We care about what happens next, because what happens next tends to dictate what the primary driver is. Over time, you'll learn that certain advertising channels tend to "drive" orders, while other advertising channels tend to "complement" orders.

Once you identify the channels that drive orders, you begin to understand how you might attribute orders to ad channels better --- and you'll make better investment decisions.

The combinations help you see how customers migrate over time ... helping you figure out "what to do next" as opposed to helping you figure out "what just happened". Ultimately, you'll plug the most popular combinations into a Multichannel Forensics simulation, and you'll see what is about to happen to the future trajectory of your business.


  1. Hi Kevin,

    I have been reading your posts on channel migration with great interest. I have a few questions on this one in particular. When creating combinations, how do you handle the following cases:

    1. You mention creating variables that capture the first touch and the last touch. So if a customer visits twice before purchasing (say paid search on March 3, then organic search on March 4, then purchase via email on March 5), you ignore the touch in the middle and the combination is "PAID SEARCH / EMAIL," correct?

    2. How much time is too much time between touch points? In other words, if you have a visit via affiliate and then a purchase via organic search, at what point do you decide too much time has passed for the two to be a combination? Would you recommend basing that on average time between orders for the company or segment being studied?


  2. Good questions, Diane!

    I honestly don't think there is a right or wrong answer to any of this.

    Honestly, I try to find combinations that have enough customers to follow. So, if there are only 27 customers who have three combinations, I don't worry about those combinations --- I'll default to first/last --- and in so many cases, there are only a handful of people that have two combinations, so I'll create one that has last touch and "other" to label the fact that there was a first touch without enough customers to follow.

    Touchpoints --- I try to keep things simple here --- for catalogers, I look within say 30 days, because catalogs usually live for between 21 and 60 days. For online brands, I tend to look at activities within one or two weeks.

    Good job!

  3. Thanks for your help, Kevin! Your blog has given me so many great ideas.


Note: Only a member of this blog may post a comment.

Best Practices? Probably Not. Sold Out? Yeah!

From time to time I share information with you from the fish monger in downtown Phoenix. And I know you'll tell me that they're a sm...