August 14, 2013

2011: Equalizing Customers

Five popular songs from the year that was 2011:
  • "Rolling In The Deep" - Adele
  • "Party Rock Anthem" - LMFAO featuring Lauren Bennett and GoonRock
  • "Firework" - Katy Perry
  • "E.T." - Katy Perry featuring Kanye West
  • "Give Me Everything" - Pitbull featuring Ne-Yo, Afrojack, and Nayer

You can't look at customer behavior without thirty-nine vendors screaming at you about attribution.
  • "Brands leveraging last-click attribution are destined for the scrapheap of obsolescence."
  • "Our complex and proprietary algorithms enable sophisticated brands to thoroughly understand which marketing channels caused orders, and which marketing channels influenced orders."
By 2011, you were behind the times if you didn't employ some level of attribution.  In catalog marketing, matchback (first-touch) was/is used most often.  Online marketers typically used last-touch attribution.

Then you have the 99% of attribution vendors who will not let you behind the curtain to see the Wonderful Wizard of Oz.

In other words, you're doing it wrong, and the experts won't say a peep about how to do it right - just trust them with your money.

Worse, if you hire 100 different attribution vendors/experts, you get 100 different outcomes. Would you get 100 different outcomes in your merchandise reporting?

And the experts frequently disavow testing ... the only methodology that actually tells you the impact of marketing channels.

So you're stuck, aren't you?

Here's a problem I was asked to analyze.  "What is the value of an email click-through?" In other words, if a customer clicks through an email campaign and then does nothing, does that click add to the lifetime value of the customer?

The secret, of course, is to compare equal customers ... some who just clicked-through an email campaign and did nothing ... compared to others who did not click-through an email campaign.

One of the tactics you use is customer scoring.  Folks score customers, predicting annual customer value based on historical RFM (recency, frequency, monetary) value plus channel overlay plus mobile/social overlay.  The customer receives a predicted annual purchase rate (say 40%), and the customer receives a predicted spend amount (say $200).  By multiplying the two factors together, you get a customer with $80 of next-twelve-month demand value. Each customer in the database is scored.

Next, you take a group of customers with, say, $75 to $85 of next-twelve-month value (generally equal-value customers), and you segment them ... some clicked through an email campaign in the past week and did not buy, most did not click through the email campaign.

Once customers are segmented - you measure spend in the next 30 days, or 90 days, or 180 days, or 365 days ... whatever you feel is the appropriate measurement window. Because the customer segments have been "equalized", the theoretical (and I repeat - 'theoretical') difference in the two customer segments is an email marketing click ... therefore, any future value is attributed to the email marketing click.

Many attribution vendors use derivations of this methodology ... instead of applying segmentation strategies, they use statistical modeling to accomplish a more sophisticated outcome.

Here's the thing - it's easy to explain what I did to an Executive.  It's terribly difficult to say that you have a sophisticated methodology that you can't share for proprietary reasons, but your methodology is better than the segmentation/equalization technique I described so you, the Executive, just need to trust everything to the vendor.

Segmentation/Equalization became terribly important in 2011, because of the explosion of mobile and myriad social channels.  Beginning in 2011, most of my projects had some element of Segmentation/Equalization, used to explain to Executives what is happening to customers.  Segmentation/Equalization led to "comp segment" analytics that helped illustrate that merchandise issues (not channels) were killing businesses.  That led to Merchandise Forensics, a series of tools that now comprise more than half of my projects.

Your attribution vendor is executing what the market wants - analysis of a bunch of orders, parsed into the theoretical marketing channels that caused the order to happen. And we need attribution, no doubt about it.

That being said, your executive team realizes that the business succeeds / fails via merchandise productivity. As you go down the attribution path, you're going to realize that attribution leads you to question merchandise productivity. It has to.