Adobe’s Dig­i­tal Mar­ket­ing solu­tions are now open­ing doors to pre­dic­tive mar­ket­ing opti­miza­tions that were pre­vi­ously only avail­able to huge com­pa­nies that spend mil­lions of dol­lars con­struct­ing com­plex eco­nomic mod­els to explain the reach of their mar­ket­ing spend.  These large mod­els typ­i­cally include sur­veys, stud­ies, pan­els, and hosts of other expen­sive out­side resources to deliver sta­tis­ti­cal opti­miza­tion to your mar­ket­ing bud­get that will max­i­mize your return on spend.

With Adobe’s recent acqui­si­tion of Effi­cient Fron­tier and recent advance­ments in pre­dic­tive con­sult­ing, Site­Cat­a­lyst cus­tomers are able to unleash the full poten­tial of their web ana­lyt­ics data in ways that have not been pos­si­ble before.

One par­tic­u­lar cus­tomer Adobe Con­sult­ing recently engaged with wanted to deter­mine how to best allo­cate their dig­i­tal mar­ket­ing bud­get based on sev­eral years of his­tor­i­cal per­for­mance.  Their dig­i­tal mar­ket­ing chan­nels included SEO, SEM, affil­i­ates, media adver­tis­ing, social media, net­work part­ners, and email cam­paigns.  Because the per­for­mance of these chan­nels had been thor­oughly recorded across sev­eral years, Adobe Con­sult­ing was able to use this data to form sta­tis­ti­cal per­for­mance mod­els around each of these mar­ket­ing channels.

Accom­plish­ing this analy­sis required sev­eral impor­tant steps:

1.  Build­ing the right attri­bu­tion model that incor­po­rates cross-channel interactions

Attri­bu­tion can be a very chal­leng­ing prob­lem in its own right.  Part of the prob­lem that com­pa­nies face (includ­ing our cus­tomer) is that the total online rev­enue is not usu­ally com­pletely due to dig­i­tal sources, so attribut­ing all online rev­enue to dig­i­tal sources can be a big mis­take that will lead to over­es­ti­ma­tion in a company’s return on spend.

To tackle this prob­lem, Adobe Con­sult­ing Ser­vices cre­ated a spe­cial dataset based on Site­Cat­a­lyst data that com­bined sev­eral years of user inter­ac­tions across each cam­paign chan­nel with the total rev­enue that each user gen­er­ated in the same time period.  Using this spe­cial dataset and sev­eral sta­tis­ti­cal mod­els, we were able to assign the mea­sured rev­enue to 3 cat­e­gories: rev­enue that was derived from non-digital sources, rev­enue that was derived from non-paid “earned” dig­i­tal sources, and rev­enue that was derived from paid dig­i­tal sources.

2.  Assign­ing mar­ket­ing spend amounts to each chan­nel and campaign

This step might seem the most obvi­ous, but it is actu­ally quite chal­leng­ing.  Many orga­ni­za­tions do not have a cen­tral loca­tion to look up the mar­ket­ing spend on each cam­paign chan­nel, so this infor­ma­tion has to be hunted down from var­i­ous groups across the com­pany.  Addi­tion­ally, some mar­ket­ing chan­nels do not have an imme­di­ately appar­ent cost such as cer­tain social cam­paigns or SEO.  With help and addi­tional data from our cus­tomer, we were able to finally assign a cost to each chan­nel or cam­paign in a way that the cus­tomer felt was most logical.

3.  Build­ing mean­ing­ful mod­els to char­ac­ter­ize what each dol­lar spent means to each mar­ket­ing channel

After each chan­nel has been assigned a cost and return, we built appro­pri­ate mod­els for each chan­nel that map spend within a chan­nel to a cer­tain return based on each channel’s indi­vid­ual cam­paign per­for­mance, search key­word per­for­mance, or ad impres­sion per­for­mance.  These mod­els describe the the­o­ret­i­cal return of any given cam­paign spend.

4.  Com­bin­ing all of these mod­els and form­ing an opti­mum mar­ket­ing blend

Finally, once each chan­nel has been given an opti­mal per­for­mance model, a sim­u­la­tion is run that will locate the opti­mal media mix to max­i­mize an aggre­gate mar­ket­ing return using an over­all bud­get that reaches across all of the mar­ket­ing chan­nels together.

After the entire process had been com­pleted for our cus­tomer, the model found that there was a larger oppor­tu­nity in SEM than they had pre­vi­ously thought.  We also found that media adver­tise­ments were being weighted too heav­ily and had a much smaller ROI than the cus­tomer ini­tially believed.  This was incred­i­bly use­ful infor­ma­tion to our cus­tomer who had already sus­pected that this was the case but just needed some solid analy­sis to val­i­date his claims to upper man­age­ment.  Most impor­tantly, Adobe Con­sult­ing found that by real­lo­cat­ing the mar­ket­ing bud­get, our cus­tomer was able to make an addi­tional $600k in rev­enue that would have been left on the table otherwise.

Using Site­Cat­a­lyst to con­struct media mix mod­els is an awe­some way to take full advan­tage of the mas­sive amount of data that you may already be col­lect­ing.  The result is not (as Jared Lees has already pointed out) a mag­i­cal out­put con­jured from a crys­tal ball, but more of a guid­ing com­pass to inform and direct a dig­i­tal mar­keter in how improve­ments can likely be made.

If you’re inter­ested in build­ing your own media mix model with your Site­Cat­a­lyst data, Adobe con­sult­ing can help you set up, cre­ate, and max­i­mize your mar­ket­ing spend with­out hav­ing to rely on a multi-million dol­lar prod­uct.  Con­tact your Adobe sales rep or account man­ager to learn more!