This is the sec­ond post in a 2-part series that exam­ines look-alike mod­el­ing in Adobe Audi­ence­M­an­ager.


My pre­vi­ous post on look-alike (or algo­rith­mic) mod­el­ing pro­vided a con­cep­tual overview of how mod­el­ing worked and the poten­tial ben­e­fits it pro­vides to our buy and sell-side cus­tomers.  In this post, I’m going to change our focus and walk through a buy-side use case. I want to talk about how to use look-alike mod­el­ing to solve a basic remar­ket­ing prob­lem and tie this back to the con­cept of data/modeling trans­parency, which is an impor­tant dif­fer­ence between Audi­ence­M­an­ager mod­el­ing and the some­what more opaque solu­tions pro­vided by other products.

Retar­get­ing is lim­ited in scope:  A buy-side use case

Just by its very nature, re-targeting is only reach­ing vis­i­tors who have pre­vi­ously been to a site; which means, it’s lim­ited in reach.  Re-targeting is proven to be quite effec­tive, by deliv­er­ing rel­e­vant offers to a set of users who are inter­ested in a spe­cific prod­uct or type of prod­uct.  How­ever, the crit­i­cal ele­ment in re-targeting is deliv­er­ing the right cre­ative to the right audience.

Take, for exam­ple, vis­i­tors to the Adobe​.com site.  We have a mix of cus­tomers who have bought prod­ucts or browse the site for infor­ma­tion or the com­mu­nity inter­ac­tion.  Since Adobe offers a wide range of prod­ucts, and it’s very likely that the same users who pur­chased or browsed con­tent for the Cre­ative Cloud or prod­ucts like Pho­to­shop, Illus­tra­tor, or InDe­sign are not the same type of buy­ers for Site­Cat­a­lyst, Test&Target, or Search&Promote.  Look­ing for cre­ative users (art, design, con­tent, etc.) among ana­lyt­ics cus­tomers reduces the effec­tive­ness of your retar­get­ing efforts along with wast­ing time and money.

There­fore, it is crit­i­cal to build mean­ing­ful seg­ments that can be tar­geted accu­rately with the right con­tent – which is the point of this blog.  And, it is also desir­able to extend the reach of a re-targeting cam­paign to become an acqui­si­tion cam­paign.  To do this, we begin with that highly valu­able set of anony­mous data cap­tured from our site.  It is a great start­ing point to cre­ate look–alike mod­els which will help our mar­ket­ing team build a retar­get­ing cam­paign that is accu­rate with respect to find­ing the right type of cus­tomer and has enough reach to help meet cam­paign goals.

Build a look-alike model in the Audi­ence­M­an­ager UI: Cre­ate a baseline

To build a look-alike model that helps find new users, our mar­ket­ing team would log in to Audi­ence­M­an­ager, select Mod­els from the Man­age Data sec­tion, and click Cre­ate New Model.  After pro­vid­ing basic infor­ma­tion (model name and an optional descrip­tion), our team selects an exist­ing trait or seg­ment that already con­tains the type of users they want to find more of.  Next, they would choose a 30, 60, or 90-day look-back period to set a time range for the model.  Together, the selected trait, seg­ment, and time inter­val form a base­line for the TraitWeight algo­rithm.  The base­line is the basic group TraitWeight looks for when it searches for new users in other data sources.

In the illus­tra­tion above, our mar­ket­ing team has selected an exist­ing trait, “Cre­ative Cloud Pur­chaser.”  This con­tains the type of users our mar­keters want to find more of in their retar­get­ing campaign.

Select data sources to find new users

In this next step, the mar­ket­ing team selects the first and third party data sources they want the algo­rithm to model.  In this case, we want to find users that we haven’t seen on an Adobe prop­erty before, so we’re also includ­ing data licensed from a third-party data provider.

Once we select a trait or seg­ment, a time inter­val, and an avail­able data source, the model is ready to start search­ing for new users sim­i­lar to those in the selected trait or seg­ment.  When fin­ished, you can cre­ate and tar­get new traits and seg­ments with this data in AudienceManager.

Data trans­parency

Pro­vid­ing our cus­tomers with an unmatched level of con­trol and trans­parency over the model results is cen­tral to our look-alike mod­el­ing fea­ture.  This makes itself evi­dent in the data users get after the model runs.  In the results, our UI exposes the spe­cific under­ly­ing data points we’ve col­lected about new audi­ences.  Addi­tion­ally, it orders the results accord­ing to how closely they model the type of cus­tomer you want to reach.  These results appear in Trait Builder in a table of influ­en­tial traits as shown below.

 In the results, Audi­ence­M­an­ager gives you access to a new set of users dis­cov­ered in your selected data sources.  These look-alike mod­el­ing results pro­vide you with real feed­back in terms of how impor­tant, accu­rate, or valu­able var­i­ous traits are to a campaign.

Com­pared to Audi­ence­M­an­ager, other prod­ucts just gen­er­ate results and say, “here’s a new set of users, good luck tar­get­ing them.”  Our mod­el­ing process goes beyond that and gives mar­keters the abil­ity to deter­mine how spe­cific they want to be with their tar­get­ing.  Is their goal to run an effec­tive direct response cam­paign?  Well, algo­rith­mic mod­el­ing lets them see the most accu­rate traits they have access to and build new seg­ments with those users.  When the goal is brand aware­ness, as it is with our Adobe exam­ple (we want to reach a lot of new users), mar­keters can build new seg­ments that con­tain a desired audi­ence size (up to 25 mil­lion).  Note, how­ever, it’s impor­tant to under­stand that accu­racy declines as you increase reach.

Data con­trol

With a given result set, a mar­ket­ing team can build their own seg­ments with this data.  Also, they can take this process to the next level and have Audi­ence­M­an­ager build the seg­ments for them, but use radio but­tons or a slider con­trol on the results graph (shown below) to retain con­trol over the reach and accu­racy of the model.

Final thoughts

Once you choose how many users you want in a new seg­ment (and you can cre­ate mul­ti­ple seg­ments with dif­fer­ent size thresh­olds from the same model), those users are now tar­getable in real-time.  With look-alike mod­el­ing, mar­keters have a new, pow­er­ful tool they can use as part of their acqui­si­tion strate­gies and brand objec­tives.  Also, our model pro­vides data trans­parency and choice by giv­ing you access to all the results from each model run.

For more infor­ma­tion about look-alike mod­el­ing in Adobe Audi­ence­M­an­ager, talk to your part­ner solu­tions representative.