Both Site­Cat­a­lyst and Test&Target are amaz­ing and pow­er­ful tools that allow a dig­i­tal mar­keter to bet­ter under­stand the traf­fic on their site and test their mar­ket­ing ideas among groups of users.  How­ever, busi­ness users often won­der how to make sense of the titanic amount of data that is col­lected and actu­ally use it to increase con­ver­sion on their site.  What I’m going to show you today is that you can actu­ally build a sta­tis­ti­cal model to find the users who are most likely to com­plete a con­ver­sion using Site­Cat­a­lyst, and then tar­get those exact indi­vid­u­als using Test&Target to really get some amaz­ing con­ver­sion lift!

To explain how this works, let me break the process down into the fol­low­ing steps that I’ll explain in more detail:

1.  Define your most impor­tant con­ver­sion event

First, you’ll need to define what con­sti­tutes your most impor­tant con­ver­sion event.  This can be a pur­chase, or per­haps a sub­scrip­tion, or a form that you’d like your users to fill out.  This event needs to be mea­sured in Site­Cat­a­lyst and can be mea­sured using a suc­cess event, eVar, or prop.

2.  Con­struct a sta­tis­ti­cal model to find those users who are most likely to convert

Next, Adobe Con­sult­ing can help you cre­ate a spe­cial dataset that can be used to build a machine learn­ing model that can fore­cast like­li­hood of con­ver­sion.  This model will use the con­ver­sion event you defined ear­lier with all the addi­tional Site­Cat­a­lyst data col­lected around each vis­i­tor to your site.  Basi­cally, the dataset will look some­thing like this:

Each row in the dataset rep­re­sents the sum­mary of an indi­vid­ual user’s behav­ior over a given date range, and can include any data that you’re record­ing as inputs to the model.  The model (in our case a deci­sion tree model) will use all of these users’ attrib­utes as inputs, and it will spit out the prob­a­bil­ity that a given user will con­vert in the future.  If you’ve never heard of a deci­sion tree before, you can read all about deci­sion trees here.  The out­put of a deci­sion tree looks some­thing like this:

This deci­sion tree is fab­ri­cated, but it’s easy to see how it cre­ates groups of users that are more or less likely to com­plete a con­ver­sion event based on their behav­ior.  The rea­son that a deci­sion tree is a great model is because it results in a set of out­put rules that can be used as seg­ment def­i­n­i­tions in Site­Cat­a­lyst or for tar­get­ing pur­poses in Test&Target.  For the exam­ple above, we could set up a seg­ment “Likely Con­vert­ers” that had the fol­low­ing seg­ment definition:

This Site­Cat­a­lyst seg­ment would allow you to report on users who were most likely to com­plete your con­ver­sion event.  Like­wise, you could also setup a Site­Cat­a­lyst seg­ment to report on users that were unlikely to convert.

3.  Adobe con­sult­ing can import these likely con­vert­ers into a Test&Target profile

Often­times, valu­able vis­i­tor seg­ments are defined by things they did not do over a long period of time. This makes tar­get­ing very dif­fi­cult using Test&Target since a vis­i­tor pro­file expires two weeks after the last mbox they vis­ited.  To work around this prob­lem, Adobe Con­sult­ing has cre­ated a process to import these users’ IDs into a Test&Target pro­file.  This solu­tion opens new and excit­ing tar­get­ing pos­si­bil­i­ties at the indi­vid­ual vis­i­tor level that just weren’t pos­si­ble before.

4.  Cre­ate some tar­geted mar­ket­ing cam­paigns and test them against likely con­vert­ers to achieve max­i­mum lift

Now that we know the users who are most likely to com­plete your con­ver­sion event (or even those who won’t), you can cre­ate a mar­ket­ing cam­paign aimed specif­i­cally at those indi­vid­u­als to entice even greater con­ver­sion success.

An impor­tant point to note is that no model can guar­an­tee that a user will per­form or respond bet­ter to a cam­paign (even though our model sug­gests they will).  For this rea­son, it’s impor­tant to test this seg­ment to see if your pre­dic­tive model was effec­tive.  For exam­ple, give 50% of likely con­vert­ers your cam­paign and see how they com­pare to the other 50% of likely con­vert­ers who do not see the campaign.

To sum­ma­rize, here’s what the process flow looks like:

Sev­eral Adobe Con­sult­ing cus­tomers have begun using this process (or have a sim­i­lar one) and have already seen sub­stan­tial suc­cess.  If you’re inter­ested in using sta­tis­ti­cal mod­els to guide your test­ing efforts, con­tact your sales rep or account man­ager soon!