Experience Personalisation: Data and Content

Expe­ri­ence Per­son­al­i­sa­tion Series, Part 2 : Content

This arti­cle is the sec­ond in a three-part series, you can read part 1 here: Expe­ri­ence Per­son­al­i­sa­tion: Get­ting the Data Right

“Data + Con­tent = Per­son­al­i­sa­tion.” This is the sim­ple for­mu­la we use here at Adobe to express the impor­tance of know­ing who you’re serv­ing con­tent to, and when and where to serve it. The bet­ter you’re able to per­son­alise that content—and its delivery—the more effec­tive­ly each piece of con­tent will help dri­ve conversions.

In this series of three arti­cles, I’m explor­ing all three aspects of this equation.

In the first arti­cle of the series, we exam­ined two cru­cial ways of gath­er­ing data on cus­tomers: build­ing pro­gres­sive pro­files from mul­ti­ple inter­ac­tions with the same users; and pulling in action­able pro­file data from oth­er sys­tems through­out your organ­i­sa­tion, such as your cus­tomer rela­tion­ship man­age­ment (CRM) sys­tem, and inter­nal sys­tems like enter­prise resource plan­ning (ERP) and data ware­house (DWH).

Now, in this sec­ond arti­cle, we’ll be tak­ing a deep­er dive on content.

Automa­tion

In the most basic terms, automa­tion is the use of algo­rithms and machine learn­ing to mod­el cus­tomer behav­iour. One of the most pow­er­ful uses of automa­tion is auto­mat­ed behav­iour­al tar­get­ing—lever­ag­ing large amounts of cus­tomer data and machine learn­ing to deter­mine which vari­ables in a cus­tomer pro­file will be most pre­dic­tive of a purchase.

Any mar­ket­ing sys­tem you use should pro­vide a com­pre­hen­sive solu­tion for mod­el­ling cus­tomer behaviour—a solu­tion that takes both the actions and pref­er­ences of the gen­er­al pop­u­la­tion into account, as well as those of the indi­vid­ual vis­i­tor. This way, changes in user behav­iour on your site caused by exter­nal factors—for exam­ple, a new prod­uct is released, a star is pic­tured wear­ing your cloth­ing or using your prod­uct, or the sea­son changes—can be picked up quick­ly; and when an indi­vid­ual exhibits strong pref­er­ences that over­ride the respons­es of the gen­er­al pop­u­la­tion, this can also be captured.

For exam­ple, say you run a site for a pay-per-view TV oper­a­tor. You need to make cus­tomers aware of the tele­vi­sion chan­nels you pro­vide, as well as your sport and movie offer­ings. You see a par­tic­u­lar vis­i­tor mul­ti­ple times—and this vis­i­tor always brows­es the crick­et pages in the sports sec­tion, as well as look­ing at their account online. From this infor­ma­tion, you can con­struct a pic­ture of who this cus­tomer is, based on the individual’s account pro­file as well as on the types of con­tent that they like to consume.

Then, when the new foot­ball sea­son starts, and users of your site start respond­ing in great num­bers to the foot­ball con­tent you’re serv­ing up on the home­page, your sys­tem will know that this par­tic­u­lar user is more like­ly to respond to crick­et-relat­ed content—and rather than wast­ing a serve, your sys­tem will present that user with a home­page tai­lored around his inter­est in crick­et. Mean­while, vis­i­tors for whom we don’t have this lev­el of detail will see a foot­ball-relat­ed cre­ative, because that’s what the gen­er­al pop­u­la­tion of the site are engag­ing with.

Sky TV in the UK recent­ly used this exact approach to deliv­er rel­e­vant con­tent and mes­sag­ing to users engag­ing with the Sky Shop, in order to dri­ve incre­men­tal sales and upgrades on desk­top and mobile.

Real cus­tomer pro­files, of course, include many more ele­ments than this—and the more robust your cus­tomer pro­files, the more use­ful and time-sav­ing vari­able tar­get­ing will be.

Opti­mi­sa­tion

Adobe Target’s auto­mat­ed behav­iour­al tar­get­ing capa­bil­i­ties inte­grate seam­less­ly with the oth­er con­tent devel­op­ment and deliv­ery tools through­out Adobe Mar­ket­ing Cloud—meaning that as soon as the sys­tem dis­cov­ers a high­ly pre­dic­tive vari­able, it can imme­di­ate­ly begin assem­bling the ide­al cre­ative to tar­get that vari­able, and serve that ide­al cre­ative to the cus­tomer across chan­nels and devices.

Anoth­er aspect of automa­tion comes through the use of some­thing we call auto-allo­ca­tion. In a stan­dard A/B test, you have to run the test, col­lect enough data in terms of respons­es or con­ver­sions, read the results and then imple­ment the win­ning variation—which will often be dif­fer­ent per seg­ment or audi­ence that you have on your dig­i­tal properties.

With auto-allo­cate func­tion­al­i­ty, the solu­tion auto­mat­i­cal­ly deter­mines which expe­ri­ence is work­ing best for your giv­en mea­sures of success—then serves that expe­ri­ence auto­mat­i­cal­ly to your vis­i­tors, so that you max­imise the expo­sure of that best per­form­ing mes­sage, ban­ner, or creative.

Even bet­ter, oth­er expe­ri­ences are also being con­tin­u­al­ly tested—so if any sig­nif­i­cant vari­able changes, the solu­tion can react and change the expe­ri­ence for you, with­out the need for man­u­al intervention.

Rec­om­men­da­tions

We’ve all seen the rec­om­men­da­tions at the bot­toms of pages on sites like Amazon—similar and relat­ed items that cus­tomers viewed and bought. Those rec­om­men­da­tions pro­vide a tiny win­dow into a vast col­lec­tion of crowd behav­iour, which the solu­tion is using to make rec­om­men­da­tions to you, the indi­vid­ual user – who may not even have a pro­file in that data­base yet.

These types of rec­om­men­da­tions are a pop­u­lar exam­ple of the use of behav­iour data from the gen­er­al pop­u­la­tion to pre­dict how a user may behave. In oth­er words, the more peo­ple in the world are engag­ing in a cer­tain activ­i­ty at a cer­tain time—going car­a­van­ning in the sum­mer, for exam­ple, and read­ing about it on your site—the more like­ly your cus­tomer is to imi­tate the trend. How­ev­er, a great rec­om­men­da­tions solu­tion will also enable you to bring indi­vid­ual pref­er­ences and behav­iour into account—just as we saw with auto­mat­ed behav­iour­al targeting.

When you bring these two types of auto­mat­ed analysis—automation and recommendations—together, you’ll gain insights not only on what your cus­tomers want and need, but also on what they plan to do soon, as well as what they’re like­ly to do in the near future. This lev­el of insight enables you to serve per­son­alised, rel­e­vant cre­ative like nev­er before.

In the third and final arti­cle of this series, I’ll be delv­ing into the per­son­al­i­sa­tion spec­trum, and explor­ing how data and con­tent come togeth­er to dri­ve increas­ing­ly per­son­alised cre­ative. See you there!