Pre­dic­tion is very dif­fi­cult – espe­cially if it is about the future.” Niels Bohr, Dan­ish physicist

Today Adobe Dig­i­tal Index released its pre­dic­tion for daily online sales this hol­i­day sea­son. Everybody’s got a pre­dic­tion.  Why is Adobe’s any bet­ter?  3 reasons:

  1. Purely data dri­ven: Based on sta­tis­ti­cal model, not informed conjecture
  2. Rel­e­vant: Mar­keters can dig into the spe­cific seg­ment that mat­ters to them
  3. Action­able: Daily pre­dic­tions explain key sales dri­vers (vis­its, con­ver­sion, aver­age price paid)

Purely data driven

We’ve ana­lyzed more than 150 bil­lion vis­its to 500+ retail web­sites and found shop­ping pat­terns that are star­tlingly con­sis­tent and sur­pris­ingly robust. The regres­sion model used to pre­dict total US online sales looks only at nor­mal­ized online sales over the last six years and is able to pre­dict this year’s sales with an R2 value higher than .95. Roughly trans­lated, 95% of the vari­a­tion in day-to-day online sales can be explained by fac­tors in the model. It’s truly impres­sive that the model can have such a high degree of accu­racy with­out rely­ing on impre­cise exter­nal data and eco­nomic indi­ca­tors. Peo­ple ask whether hur­ri­cane Sandy or the recent elec­tion will dra­mat­i­cally affect online shop­ping. The only response I can offer is that there was an elec­tion in 2008, an eco­nomic reces­sion in 2009, stag­nant high unem­ploy­ment in 2010, a debt ceil­ing cri­sis in 2011 and the data shows that through it all online shop­ping pat­terns have been per­sis­tent. If some event in 2012 were to change those pat­terns it would truly be “sta­tis­ti­cally significant.”


We’ve cre­ated an inter­ac­tive web­site that will allow mar­keters to slice and dice the data in 168 dif­fer­ent ways and dig out the most rel­e­vant insights. Mar­keters always have to won­der whether their actions are hav­ing the desired effects. Did our Cyber Mon­day deals cause peo­ple to buy or would they have any­way? Did we start our hol­i­day pro­mo­tions too soon or are peo­ple just not ready to shop yet. His­tor­i­cal data only tells the mar­keter what hap­pened, not what would or could have hap­pened. By ana­lyz­ing the indus­try as a whole, Adobe is able cut out the noise to reveal the under­ly­ing con­sumer behav­ior to help mar­keters know what to expect. For exam­ple, while typ­i­cal retail­ers should expect that con­ver­sion rates will jump to 60% above nor­mal on Cyber Mon­day, apparel retail­ers can expect a much higher jump up to 215% of nor­mal just because shop­pers will be in the mood to buy clothes. Retail­ers who see dif­fer­ent results than their seg­ment will know it’s because of their atyp­i­cal actions.


Big data also gives us the abil­ity to pro­vide a daily pre­dic­tion.  Why is a daily pre­dic­tion impor­tant?  Because it enables mar­keters to over­lay their own pat­terns against the aver­age and dis­cover how their pro­mo­tional strate­gies are cre­at­ing sep­a­ra­tion between their web­site and the rest of the cat­e­gory. Besides know­ing that sales will be up again mar­keters need to know what will drive sales day-to-day. For exam­ple, this Fri­day online sales are going to jump to 200% of nor­mal for the typ­i­cal retailer. Why? Not because peo­ple will be shop­ping that much more. Web­site vis­its will be up only 10% above today. Online sales will spike because con­ver­sion rates will jump up to 40% between today and four days from now. By under­stand­ing these shop­ping pat­terns retail­ers can tai­lor their pro­mo­tions to the mood and pur­chase propen­sity of their customers.

You’d never fin­ish read­ing this blog post if I tried to tell you the rest this data says about Black Fri­day. And that’s just one day. So go ahead and find the answers your­self here. Let me know what you learn and what you’re inter­ested in. And fol­low me @tyrwhite if you’re dying to hear more.