Recently, I shared a blog post out­lin­ing the details sur­round­ing our vision, strat­egy, and cur­rent capa­bil­i­ties in pre­dic­tive ana­lyt­ics, which have earned Adobe Ana­lyt­ics a per­fect score in this cri­te­ria in the lat­est For­rester Wave: Web Ana­lyt­ics Q2, 2014 report. Adobe Ana­lyt­ics received a per­fect eval­u­a­tion in 66 out of 75 cri­te­ria. Among these cri­te­ria, we received the high­est pos­si­ble score in the Cor­re­la­tions cat­e­gory. How’d we do it? Let me share our cor­re­la­tions capa­bil­i­ties and how this rein­forces our vision in help­ing build the smarter enterprise.

Defin­ing Correlations

In an ear­lier blog post, “Cor­re­la­tion Defined: Solv­ing Mys­ter­ies,” I laid out an in-depth def­i­n­i­tion of sta­tis­ti­cal cor­re­la­tions (e.g., Pear­son product-moment cor­re­la­tion coef­fi­cient) and some of the tips, tricks, and best prac­tices in using cor­re­la­tions as you explore and ana­lyze your data. As a quick review, cor­re­la­tion (or sim­ply “co-“ mean­ing “together” and rela­tion, with an extra “r” thrown in for some rea­son) mea­sures how closely related two sets of data are to one another. While cor­re­la­tion does not mean “cau­sa­tion,” it is still use­ful in dis­cov­er­ing hid­den rela­tion­ships across all your data and gen­er­at­ing ideas for analy­sis and opti­miza­tion. Cor­re­la­tion is also used as a method for iden­ti­fy­ing vari­ables to include as pre­dic­tors in more advanced data min­ing and machine learn­ing algo­rithms. If ana­lyt­ics were auto mechan­ics, cor­re­la­tion would be the diag­nos­tic machine hooked up to the engine of sta­tis­ti­cal analysis.

illustrations8

Adobe Ana­lyt­ics KNOWS Correlations

Most ana­lysts have too much data and not enough time to ana­lyze it all. A dis­pro­por­tion­ate data-to-time ratio can stymie analy­sis and over­whelm ana­lyt­ics prac­ti­tion­ers. Cor­re­la­tions are an ana­lyt­ics tool that can help alle­vi­ate some of these chal­lenges.  Adobe Ana­lyt­ics gives the user the abil­ity to cor­re­late unlim­ited dimen­sions and met­rics across all avail­able data. Data from dif­fer­ent touch points can be cor­re­lated based on cus­tomer ID, vis­i­tor ID, time, geog­ra­phy, or any other attrib­ut­able dimen­sion the user desires to use as the anchor for the cor­re­la­tion matrix. For sit­u­a­tions with­out a customer/visitor ID that helps con­sis­tently stitch across devices and chan­nels, using time or dimen­sions such as cam­paign track­ing codes are use­ful as a basis for the cor­re­la­tion. These cor­re­la­tions can be applied across web­sites, apps, and other user touchpoints.

Besides pro­vid­ing a rich, visual expe­ri­ence for under­stand­ing the strength and direc­tion across cor­re­lated data sets, Adobe Ana­lyt­ics enables a user to derive new met­rics and dimen­sion val­ues on the fly based on cus­tom def­i­n­i­tions set by the ana­lyst using it.

illustrations9

Users can incor­po­rate built-in binary fil­ters to derive new met­rics (0,1). For exam­ple, the ana­lyst may want to cor­re­late orders with video views across peo­ple. Per­haps they may want to know if there is a rela­tion­ship between an order being placed and whether a per­son ever watched a spe­cific video (not the total num­ber of video views). The ana­lysts would then use binary fil­ters to define a new met­ric called “Video Viewer” (Video Views ≥ 1). To cor­re­late across time, a user may apply a binary fil­ter to “logins” to mea­sure the rela­tion­ship between days with at least 1,000 logins and orders. The new met­ric (“Mucho Logins”) may be defined as logins ≥ 1,000.

Adobe ana­lysts can also use binary fil­ters to cre­ate new dimen­sion ele­ments by drag­ging and drop­ping them onto the met­rics within the cor­re­la­tion matrix. For exam­ple, they may believe that “Microsoft” within the Browser Type report influ­ences the cor­re­la­tion between logins and rev­enue. To prove or dis­prove this belief, sim­ply drag and drop Microsoft within the Browser Type table onto logins within the cor­re­la­tion matrix. The next step would be to select a com­par­i­son value and label the newly derived metric.

Cor­re­la­tions demon­strate our com­mit­ment to reduc­ing the time to insight for the mar­keter and ana­lyst. Adobe Ana­lyt­ics cou­ples that with invest­ments aimed at bet­ter data, sim­pler expe­ri­ence, smarter analy­sis, and stronger inte­gra­tions. Did I men­tion we received a per­fect score?

1 comments