Utilizing Predictive Revenue Models in Paid & Organic Search

One of the more inter­est­ing con­cepts pre­sented at last week’s SES show in San Jose detailed the use of pre­dic­tive rev­enue mod­el­ing for SEO and paid search. A for­mer col­league of mine, Dave Roth of Yahoo!, runs search pro­grams on behalf of Yahoo!‘s prop­er­ties. He dis­cussed their use of pre­dic­tive data mod­els to gauge how they were per­form­ing against their competition.

In SEO it works like this:

For each key­word, cre­ate an “Oppor­tu­nity report” that looks at your cur­rent organic rank and click vol­ume ver­sus your top com­peti­tors, and mea­sure the gaps. Then map those gaps to Life­time Value (LTV) to quan­tify their value. If, for exam­ple, you know how much rev­enue you made in posi­tion #1 ver­sus posi­tion #2, you can quan­tify your rev­enue losses or gains based on rank. In other words, how much rev­enue are you los­ing by not rank­ing higher than your com­peti­tors? Noth­ing gets an executive’s atten­tion faster than lost rev­enue, espe­cially if you can see (at least roughly) how much rev­enue your com­peti­tors are earn­ing by out­rank­ing you.

In paid search it works sim­i­larly, but is a lit­tle more complex:

For each key­word, cre­ate an LTV model based on past rev­enue per­for­mance, click vol­ume, rank, CPC and click costs. By fac­tor­ing in click costs you are cap­tur­ing your effi­ciency over time because click costs account for your qual­ity score. At Yahoo! Dave explained that they refore­cast every month to stay on top of rapidly chang­ing search mar­kets, then com­mu­ni­cate the updated num­bers across all rel­e­vant inter­nal depart­ments so every­body knows what rev­enue gains or losses to expect for that month. At Yahoo! they use a “score­card” to mea­sure each Yahoo! prop­erty vs. another, all based on ROI (both imme­di­ate for that month and LTV). My favorite line in Dave’s pre­sen­ta­tion was a Yahoo! mantra: “If you can’t quan­tify it, it doesn’t exist.”

In a world where every­thing is quan­ti­fied, all of your actions can be based on data and rev­enue mod­els; and per­haps even more impor­tantly, you can com­mu­ni­cate poten­tial fluc­tu­a­tions in rev­enue BEFORE they happen.

Make sure your toolset can read your search per­for­mance data across both paid and organic so you don’t have to marry two dif­fer­ent reports together. Hav­ing that data in two sep­a­rate sys­tems leaves room for human error when com­bin­ing the data sets. Don’t set­tle for any­thing less than a holis­tic approach to ana­lyz­ing the per­for­mance of your search pro­grams, and don’t rule out the impor­tance of pre­dic­tive rev­enue mod­els for your organization.