September 11, 2007

E-Mail Marketing Challenge: Assigning Customers To Versions

Deciding who receives a version of an e-mail campaign is a challenge.

You seldom read common-sense logic that explains what you should do, partly because it is hard work, partly because vendors want to monetize this process by doing the work for you.

Let's review a simple example, one that clearly outlines the challenges we all face.

The table below illustrates seven e-mail subscribers, and the amount each customer spent on Mens and Womens merchandise over the past year.

Cust # Mens Womens
One $50 $0
Two $0 $900
Three $400 $200
Four $0 $0
Five $1,335 $1,335
Six $0 $600
Seven $150 $150

Based on the purchase habits of these seven individuals, we have to assign three customers to the Mens version, and four customers to the Womens version.

Method #1 = Prioritization: One of the easiest strategies is to prioritize one version of the e-mail campaign over another. In other words, the database marketer may decide that the Mens version receives top priority. In that situation, the three highest-spending Mens customers get the Mens version, with everybody else receiving the Womens version.
  • Mens Version = Customer Number Three, Five and Seven.
  • Womens Version = Customer Number One, Two, Four and Six.
Had the Womens version been prioritized first, the assignment would have looked like this:
  • Mens Version = Customer Number One, Four and Seven
  • Womens Version = Customer Number Two, Three, Five and Six
Notice the challenge with this method. While this is the easiest method to execute, the method results in the version with the highest priority receiving the best customers. This means that the version with the highest priority will generally perform best. The version with lower priority will generally not perform as well.

Method #2 = Versions Based On Spend: In this case, customers are assigned to versions based on which merchandise division the customer prefers. This works well in theory. Notice in our case that customer numbers four, five and seven spent equal amounts in each version. Therefore, we have the following situation:
  • Mens Version = Customer Number One and Three.
  • Womens Version = Customer Number Two and Six.
  • Ties = Customer Number Four, Five and Seven.
The database marketer could randomly decide which of the three customers with ties receives the Mens version, and then allocate the other customers to the Womens version.

Method #3 = Versions Based On Expected ROI: Experienced e-mail database marketers assign an expected ROI to each customer, for each version of an e-mail campaign. In this case, each customer is "scored" based on the demand per e-mail / ROI expected from the customer, if mailed either version of the e-mail campaign. The Womens version of the e-mail campaign is expected to perform twice as well as the Mens version of the e-mail campaign. Let's look at the "scores":

Cust # Mens Womens
One $0.106 $0.163
Two $0.085 $0.329
Three $0.164 $0.213
Four $0.085 $0.163
Five $0.397 $0.794
Six $0.085 $0.252
Seven $0.126 $0.187

When we score each customer based on expected sales per e-mail sent, we see a problem, don't we?

The Mens version is expected to perform half as well as the Womens version. So, when evaluating each version of the e-mail campaign, we come to the logical conclusion that we should only have one version of the e-mail campaign ... the Womens version ... right?

Now go sit down in the office of the Mens Merchandise Executive, and tell him/her that we won't be sending an a Mens e-mail to our customers, because customers don't respond to Mens merchandise. After you make that statement, cover your ears and duck!!!!

So, you're left with a tough choice. What you have to do is assign the best customers to versions, "subject to constraints". In this case, the constraint is that we have to send three Mens e-mail versions to customers, yet we want to maximize the total ROI for the campaign.

A good tool to solve this problem is "linear programming". I don't have enough room to go into all the math here, but if you use software from Unica (the folks who brought catalogers the "Affinium" campaign management tool), you can purchase an add-on that does this for you. The add-on isn't cheap, but you're likely to recoup your profit in short order.

Of course, there are many, MANY ways to solve this problem. Here's an opportunity for you to share your thoughts. How would you attack this challenge?