Mid last year, we here on the Adobe Media Opti­mizer (AMO) team announced the avail­abil­ity of an inte­gra­tion between Adobe Ana­lyt­ics and Media Opti­mizer. With this inte­gra­tion, cus­tomers were able to begin to gain more value from the rich ana­lyt­ics data they’ve accu­mu­lated in Site­Cat­a­lyst by feed­ing it into the pow­er­ful algo­rithms of our fore­cast­ing and auto­mated bid­ding sys­tems. No longer were sep­a­rate tags required to run mul­ti­ple prod­ucts within the Adobe Mar­ket­ing Cloud. Now your world is stream­lined and sim­pli­fied. One tag. One data col­lec­tion sys­tem. One set of uni­fied data.

SC-AMO Integration With dozens of users now set up with the inte­gra­tion, we’ve seen some excel­lent results. Not only have mar­ket­ing pro­grams lever­aged the data to drive incre­men­tal ROI from our auto­mated bid­ding sys­tems, they have also real­ized effi­cien­cies in work­flow and report­ing, with Site­Cat­a­lyst con­ver­sion met­rics all avail­able in Media Opti­mizer, and all search engine met­rics (Clicks, Cost, Impres­sions, etc.) passed auto­mat­i­cally back to Adobe Ana­lyt­ics for report­ing and analy­sis in Site­Cat­a­lyst, Dis­cover, and ReportBuilder.

How­ever, back in the sum­mer of 2012 we were just scratch­ing the sur­face. From our per­spec­tive: If some data is good, more data is even bet­ter. So the bril­liant minds of our data sci­en­tists con­tin­ued plug­ging away to uncover even more value from the gold mine of ana­lyt­ics data now avail­able to Media Opti­mizer. And – eureka! – they’ve done it again! Just weeks ago, the Media Opti­mizer prod­uct team announced the launch of beta for a site engage­ment met­rics integration.

But first, a lit­tle context…

One of the biggest chal­lenges for any bid­ding sys­tem – be it algo­rith­mic, rules-based, or a sum­mer intern play­ing man­ual “bid jockey” – is data scarcity. Con­ver­sions are great, and they are exactly what we hope our mar­ket­ing pro­grams lead to. Unfor­tu­nately, there are never “enough” con­ver­sions; you always want more. And we know that across all Adobe cus­tomers, the aver­age con­ver­sion rate is only around 3%. That means out of every 100 vis­i­tors you drive to your web­site, 97 of them don’t actu­ally com­plete the ulti­mate action you’d like them to perform.

Tra­di­tional bid opti­miza­tion focuses heav­ily on the 3% of con­vert­ing vis­i­tors, increas­ing spend to the ads and cam­paigns that drive that traf­fic. How­ever, as most mar­keters know, only a small frac­tion of dig­i­tal mar­ket­ing ini­ti­ates actu­ally account for this con­vert­ing traf­fic on a fre­quent enough basis to have suf­fi­cient data vol­ume needed for effec­tive bid optimization.

So what do you do about those key­words and ads that only con­vert once a month? Once a quar­ter? Annu­ally? These bid­d­a­ble objects don’t have nearly enough con­ver­sion data to make solid bid deci­sions day in and day out. And there are a lot of them! (That’s why they’re appro­pri­ately named the “long tail”.)

Under the tra­di­tional approach of Hier­ar­chi­cal Model Esti­ma­tion, key­words with sparse con­ver­sion data have their per­for­mance extrap­o­lated from pool­ing data higher in the account hier­arch. That means that if a given key­word only con­verts, say, once in the last 90 days, the sys­tem looks at the entire ad group the key­word is in to see if there is enough data to build an appro­pri­ate model. And if the ad group still doesn’t have enough data, the sys­tem looks at the cam­paign level. The logic is intu­itive enough: Key­words with sim­i­lar attrib­utes should be grouped together, there­fore per­for­mance for the group in the short term should be able to approx­i­mate the per­for­mance of the indi­vid­ual in the long term.

And the results we’ve seen over the years have indi­cated that this hier­ar­chi­cal mod­el­ing works fairly well – with Media Opti­mizer cus­tomers expe­ri­enc­ing notable per­for­mance improve­ments once employ­ing portfolio-style bid­ding, and the sys­tem gen­er­at­ing mod­els that approached 95% accuracy.

Going Fur­ther…

But why,” we asked our­selves, “should we con­tinue to esti­mate per­for­mance of indi­vid­ual key­words based on the col­lec­tive per­for­mance of other key­words when we now have access to expo­nen­tially more data about the traf­fic that each and every key­word is dri­ving itself?” We knew there was a bet­ter way, and we were dri­ven to push that 95% model accu­racy even higher.

Enter the next gen­er­a­tion of data inte­gra­tion between Adobe Ana­lyt­ics and Adobe Media Opti­mizer. Under the newly designed sys­tem, Media Opti­mizer algo­rithms can now eval­u­ate site engage­ment met­rics such as PageViews/Visit, Time Spent on Site, and Bounce Rate when mod­el­ing out bid rec­om­men­da­tions. This means that even when cer­tain key­words and ads don’t have a high vol­ume of con­ver­sion data (which is the major­ity of your account the major­ity of the time!), we con­tinue on to eval­u­ate what lev­els of engage­ment each indi­vid­ual key­word and ad drove to your dig­i­tal prop­er­ties as well as mod­el­ing out how well each of these engage­ment met­rics serve as lead­ing indi­ca­tors to even­tual con­ver­sion behaviors.

The beta pro­gram for rolling out this new cross-solution inte­gra­tion was announced just a few short weeks ago, and already we are see­ing impres­sive results for par­tic­i­pat­ing accounts. Across a sam­ple group of twelve port­fo­lios run­ning on the new mod­els incor­po­rat­ing engage­ment met­rics, we have seen long-tail model accu­racy improve from 82.26% to 92.03%. That’s a nearly 10% increase in fore­cast accu­racy for the most dif­fi­cult seg­ment to model!

Increased Model AccuracyIncreased Model Accu­racy = More Effec­tive Bid Automa­tion = Increased ROI!

Obvi­ously we are still just scratch­ing the sur­face of what is pos­si­ble when you merge the rich data of an industry-leading ana­lyt­ics plat­form with the power of world-class algo­rith­mic mod­el­ing for mar­ket­ing automa­tion. Whether it’s rapidly adapt­ing to an ever-changing dig­i­tal mar­ket­ing land­scape (think Google enhanced cam­paigns) or div­ing deeper into analytics-based seg­men­ta­tion (time-parting and device-/geo-targeting any­one?), the Adobe Media Opti­mizer team will con­tinue to push for­ward to help our cus­tomers real­ize increas­ing lev­els of automa­tion effi­ciency and ROI.