Dur­ing and fol­low­ing a recent Effi­cient Fron­tier webi­nar I hosted there were a lot of rel­e­vant ques­tions. Mar­keters are inevitably wrestling with the prob­lem of attain­ing the opti­mal media mix. Understanding what each channel’s value is and the opti­mal mix for the goals at hand adds an extra layer of com­plex­ity for already busy mar­keters. The days of the “last-click wins” men­tal­ity and one-off online strate­gies have drawn to a close. Attri­bu­tion analy­sis enables a more accu­rate under­stand­ing of what leads to con­ver­sions, and these insights can lead to dra­matic improve­ments in ROI. I’ve summed up some of the per­va­sive themes of the webi­nar and com­mon ques­tions amongst the group of lead­ers of how Effi­cient Fron­tier can aid in their mar­ket­ing decisions:

Q: Tell me what attri­bu­tion is exactly.
A. Attri­bu­tion is a heuris­tic method that quan­ti­fies the influ­ence of all media chan­nels on conversions.

Q: This is for DR (direct-response) only leads? What about brands?
A: Our attri­bu­tion method­ol­ogy applies to both DR as well as brand-focused adver­tis­ers. How­ever, the key dif­fer­ence in our method is that it pro­vides a way to mea­sure as well as opti­mize to an advertiser’s goals. While it is true that brand mar­keters are not directly focused on rev­enue met­rics they do care about reach and engage­ment met­rics. Fur­ther, in every orga­ni­za­tion brand mar­keters are held account­able to the amount of busi­ness they gen­er­ate. The key to proper attri­bu­tion for brand mar­keters involves three steps:

  1. Met­ric and objec­tive selec­tion, e.g. the goals you really care about. It could be total engage­ment on Face­book, or total web­site vis­its, or a com­bi­na­tion of sev­eral met­rics such as page views, engage­ments, email cap­ture etc.
  2. Find­ing the true value of these met­rics in rela­tion to the busi­ness or rev­enue they generate.
  3. Apply­ing attri­bu­tion and opti­miza­tion algo­rithms on top of these metrics.

Step 2 is tricky because the con­nec­tion between online brand­ing met­rics and offline rev­enue gen­er­at­ing activ­ity is often ten­u­ous, and the best one can typ­i­cally do is apply econo­met­ric mod­els. If this is not fea­si­ble then a blended met­ric can be cre­ated, which is a com­bi­na­tion of each indi­vid­ual engage­ment metric.

Q: So it’s a click model? But most peo­ple don’t click on dis­play. It does not seem to mea­sure adver­tis­ing it seems more of a click response opti­mizer.
A: This is not a click model. This is a model + algo­rith­mic approach that mea­sures the effect of every mea­sur­able inter­ac­tion in the user’s con­ver­sion path. In dis­play, view through impres­sions can be mea­sured and attri­bu­tion can be applied to impres­sions. In the webi­nar, I men­tioned that the key con­cept to attri­bu­tion is to mea­sure the jump in con­di­tional prob­a­bil­ity of con­ver­sion (or another goal) over a pre­ced­ing sequence of actions. This applies to dis­play view-throughs, too. One can set up con­trolled exper­i­ments to mea­sure the value of dis­play impres­sions to under­stand their value in the con­ver­sion path. Effi­cient Frontier’s attri­bu­tion plat­form can help adver­tis­ers attribute and opti­mize to dis­play impres­sions too.

Q: What is the inter­ac­tion?
A: In the webi­nar, I men­tioned that chan­nels are con­stantly inter­act­ing with each other and that effect has to be cap­tured. Exam­ple: Let’s say that the search bud­get is $ 1 MN and dis­play is $ 2 MN and that the con­ver­sion rate on search clicks is very high. As a result, you decide to shift the bud­get from dis­play to search where search bud­gets are now $2 MN and dis­play is $ 1 MN. This might cause the con­ver­sion rates on search clicks to drop more than you expect because the dis­play impres­sions that helped the search clicks before are being served a lot less than before. The search and dis­play chan­nels are inter­act­ing all the time and this cross elas­tic­ity has to be cap­tured in a good model.

Q: Does Effi­cient Fron­tier have tag­ging that works across social and dis­play?
A: Social impres­sions can’t be tagged or tracked due to Facebook’s pol­icy, but we can track social clicks and dis­play impres­sions and clicks.

Q: Isn’t Google Ana­lyt­ics “last click” attri­bu­tion?
A: Yes, it is, but Google does pro­vide first and last click reports. (http://​www​.google​.com/​a​d​s​/​i​n​n​o​v​a​t​i​o​n​s​/​s​e​a​r​c​h​f​u​n​n​e​l​s​.​h​tml)

Q: What is the dis­tri­b­u­tion of Effi­cient Fron­tier clients using each of the mod­els?
A: We have a vari­ety of adver­tis­ers using these mod­els span­ning, but not lim­ited to: finance, auto, travel, retail and more.

Q: Does the idea of attri­bu­tion mod­els only apply to agen­cies that sup­port mul­ti­ple dig­i­tal chan­nels?
A. No, it also applies within one chan­nel. An exam­ple: in search where non-branded key­words often help branded keywords.

Q: Does Effi­cient Fron­tier take into account each company’s over­all strat­egy into this exper­i­men­ta­tion?
A. Yes, and this can be chal­leng­ing when doing the attri­bu­tion. For instance, if the attri­bu­tion sug­gests that the bud­gets need to be dra­mat­i­cally changed then it might not be fea­si­ble to do so as it can put the line of busi­ness at risk. In such a case the spend rec­om­men­da­tions com­ing from the attri­bu­tion should be used direc­tion­ally. For instance, if search is more effi­cient than dis­play and the media mix rec­om­men­da­tion shows that search bud­gets need to increase by 25% one can start with a 5% increase and observe the over­all per­for­mance. If per­for­mance improves then more bud­get can be shifted to search. Still, these approaches can be chal­leng­ing as the effect of slow bud­get shifts can be hard to mea­sure due to per­for­mance volatil­ity, sea­son­al­ity, mar­ket­place changes etc.

Q: How much exper­i­men­ta­tion is needed to find opti­mal point?
A. The­o­ret­i­cally, the exper­i­ment is always on. Opti­miza­tion requires con­stant exper­i­men­ta­tion because things are always changing.

Q: Does the Effi­cient Fron­tier attri­bu­tion sys­tem take into accounts both con­verted traf­fic and non-converted traf­fic?
A. Yes, as long as non-converted traf­fic is cap­tured via one or more metrics.

Q: Do adver­tis­ers need to rely on clicks? How does Effi­cient Fron­tier track impres­sions ver­sus clicks across chan­nel?
A. Only dis­play impres­sions can be tracked. Social and search impres­sions can­not be tracked due to the rea­sons men­tioned above.

Q: What are the tech­nol­ogy hur­dles that need to be over­come to achieve the vision? Where are the capa­bil­ity gaps today?
A. First the plat­forms such as Face­book need to allow adver­tis­ers to track view through impres­sions. Sec­ond, math­e­mat­i­cally there is no method that can accu­rately pre­dict per­for­mance for the entire media mix. The meth­ods are only locally accu­rate in their pre­dic­tions and this neces­si­tates an exper­i­men­ta­tion exploita­tion approach. I believe that, ulti­mately, attri­bu­tion is not the solu­tion for media mix mod­el­ing because it does not cap­ture the inter­ac­tion between chan­nels very well. Attri­bu­tion is a more effec­tive solu­tion for ad and con­ver­sion optimization.

Q: What are the lim­i­ta­tions of only track­ing dig­i­tal touch points and not offline media points? Are we just mak­ing dig­i­tal even more siloed in its account­abil­ity? Are there any tools that track online and offline touch points together?
A. The lim­i­ta­tions are that it is hard to directly mea­sure the online to offline effect and vice versa. There are some tools that track offline phone calls but the best one can do for TV is use panel data. Apart from that, one can build econo­met­ric mod­els, but using these is not a scal­able solu­tion because it requires a lot of human input and the qual­ity of the model depends strongly on the per­son who built it. It is also a look back into a snap­shot of time in the past and may not reflect the dynamic adver­tis­ing envi­ron­ment look­ing forward.

Q: What impact do you think Google’s move to secure searches by signed-in users will have on attri­bu­tion mod­el­ing?
A. Lit­tle apart from paid searches because these can still be tracked. How­ever, for organic searches some esti­mates say that up to 30% of traf­fic will not be accu­rately cap­tured. I don’t think this is a big prob­lem because not cap­tur­ing the organic traf­fic has poten­tial for mis­at­tribut­ing the effect of organic searches. Adver­tis­ers have lit­tle imme­di­ate con­trol over the posi­tion they show up for an organic search, so when it comes to media mix allo­ca­tion lit­tle will change.

Q: How does Effi­cient Fron­tier track TV cam­paigns?
A. Effi­cient Fron­tier doesn’t, but we use indi­ca­tor met­rics pro­vided by our adver­tis­ers such as spend, GRPS and TRPs to build econo­met­ric models.

Q: Is there a scal­able solu­tion that can track the dis­play impres­sions in the cross chan­nel track­ing?
A. Effi­cient Frontier’s solu­tion can track dis­play impres­sions in cross-channel tracking.

Next Effi­cient Fron­tier Webi­nar: Tues­day, Decem­ber 13, 2011 at 9 A.M. PST / 5 P.M. GMT
Mak­ing Face­book Mar­ket­ing Work
How to Pro­vide Value to Your Fans and Real Busi­ness Results to Your Brand

Join Effi­cient Fron­tier and fea­tured guest For­rester Research Inc. as we dis­cuss the find­ings in Forrester’s Novem­ber 2011 report, “It’s Time to Make Face­book Mar­ket­ing Work”, and pro­vide you with spe­cific next steps and con­crete exam­ples for suc­cess­ful Face­book mar­ket­ing with tan­gi­ble ROI.