Attack of the Killer Data Scientists: How to Prepare for the Invasion

by | Sep 7, 2017

This blog is inspired from a topic I presented this year at the Observe­Point User Groups in New York and Boston. The audi­ence thought it was pretty cool and I hope you do to.

The last five years of my career have been built on a shift in the way we “do” digital analyt­ics.

I entered this space back when most compa­nies had two groups of analyt­ics people: one group of tech­ni­cal people who installed analyt­ics tools (usually a web dev team), and the marketers who used said tools (usually me).

I was leading the digital analyt­ics team at the Amer­i­can Cancer Society in 2012 when we became early adopters of a tag manage­ment tool called Satel­lite (devel­oped by Search Discovery, now Adobe DTM).  It was love at first sight, and I quickly became the world’s biggest evan­ge­list of this tool. This is not a shame­less plug, I really was passion­ate about it and told anyone who would listen about how it would revo­lu­tion­ize the way we do digital analyt­ics. Even­tu­ally, I left my job at ACS to join Search Discovery, where I’ve spent the last four years helping other digital analyt­ics teams harness the power of tag manage­ment.

Tag manage­ment changed a lot about how we do our jobs, and more impor­tantly the types of skills we need to be success­ful. Analyt­ics prac­ti­tion­ers provide the same value today that we did before tag manage­ment, but the way we do our jobs has shifted dramat­i­cally.  I believe that data science is bring­ing the next shift.

The Status Quo

What Makes A Success­ful Digital Analyt­ics Team

If you read enough job descrip­tions online for a Digital Analyst, you’ll notice a theme: analysts are expected to solve busi­ness prob­lems.  Here are a few exam­ples I pulled this week:

  • Guide the use of data measure­ment to drive the strate­gic roadmap”
  • Busi­ness-savvy and action-oriented strate­gic leader”
  • Self-starter with outstand­ing analytic skills, strong busi­ness acumen and the ability to synthe­size data into mean­ing­ful insights”

And I don’t disagree. In fact, I summa­rize the senti­ment like this:

Digital analyt­ics teams provide lead­er­ship on inter­pret­ing data to advance the busi­ness.”

I’ve consulted a lot of analyt­ics teams over the years, and although every company is differ­ent, they all do four things on a daily basis to achieve this goal:

  1. Collect data — Own the tag manager, or work closely with the dev team.
  2. Oper­a­tional­ize data — Distrib­ute base­line infor­ma­tion through dash­boards and reports.
  3. Reac­tively inter­pret data — React to busi­ness ques­tions, such as “Why was revenue down last month?”
  4. Proac­tively inter­pret data — Iden­tify oppor­tu­ni­ties to drive new busi­ness, such as “A blogger drove a ton of traffic last week, we should reach out to them about an affil­i­ate oppor­tu­nity.”  Or, “This list of prod­ucts have a poor conver­sion rate from users who view the product detail page, so we should update that content to make it more appeal­ing.”

Just about every task that we consider to be digital analyt­ics falls into one of these four cate­gories (I would even stick A/B testing into #3 or #4 since it involves making a hypoth­e­sis, inter­pret­ing data, etc).

Getting Good Data

If our goal is to provide lead­er­ship on inter­pret­ing data to advance the busi­ness, then #4 is where we really show our value.  Unfor­tu­nately, most of my clients strug­gle to find time for proac­tive inter­pre­ta­tion because they’re consumed by trying to make sure they’re collect­ing the right data or are just churn­ing out KPIs to the busi­ness.

Items #1 and #2 are just about keeping the lights on. Tagging websites and mobile appli­ca­tions are thank­less tasks that are more complex than most people realize. No one is impressed when they are done well, but they get very upset when there is a problem. After all, no job descrip­tion says, “seeking someone to work through the weekend to figure out why revenue in Adobe Analyt­ics differs from revenue in our account­ing tool”!

Change is Coming

The world we operate in today requires us to dispro­por­tion­ately invest our time in data collec­tion and other tasks that offer little busi­ness value. But I have a predic­tion:

The problem of collect­ing data is going to be resolved

It won’t be tomor­row and it won’t happen overnight, but tech­nol­ogy will even­tu­ally reduce the amount of effort required to collect and main­tain good data. We’ve come a long way from collect­ing server logs to the way we work tag managers today, and we are moving quickly in the direc­tion of stan­dard­iz­ing and automat­ing data collec­tion. When the problem of collect­ing data from websites and mobile appli­ca­tions is resolved, the demand for highly skilled imple­men­ta­tion special­ists will inevitably decline.  

So, those of us who have built our careers on this demand must ask ourselves: how will our prior­i­ties shift as the chal­lenge of getting good data is resolved, and how do we prepare ourselves for this change? The answer is through data science.

How data science skills will make future analyt­ics teams success­ful

Imagine a world where analysts have excel­lent data at their finger­tips, taking data collec­tion and data oper­a­tional­iza­tion off their plates. and they still have the same respon­si­bil­ity: to provide lead­er­ship on inter­pret­ing data to advance the busi­ness.  What would their prior­i­ties be, and what skills would they need to do them well?

To answer that ques­tion, I look to the work being done today by people who focus exclu­sively on inter­pret­ing data, whom I am broadly defin­ing here as data scien­tists.  The defi­n­i­tion of data science is under debate, and not every aspect of data science applies to digital analyt­ics, but I find the follow­ing from the Data Science Initia­tive at NYU to be helpful:

One way to consider data science is as an evolu­tion­ary step in inter­dis­ci­pli­nary fields like busi­ness analy­sis that incor­po­rate computer science, model­ing, statis­tics, analyt­ics, and math­e­mat­ics.” (http://datascience.nyu.edu/what-is-data-science/)

Concur­rently, while we have been making progress to solve the problem of collect­ing good data, the historic barri­ers to entry into data science have been quickly disman­tled by the increase in free train­ing mate­ri­als and the decrease in cost for cloud computing/storage.  If we consider that we will be spend­ing less of our energy on data collec­tion in the next 5 or 10 years, and think of data science as the next evolu­tion­ary step in digital analyt­ics (which is becom­ing increas­ingly attain­able), it is helpful to explore how computer science, model­ing, statis­tics, and math­e­mat­ics are already being applied to digital analyt­ics prob­lems today.  

Here are three exam­ples that are occur­ring on the fringes of our indus­try today, but which I predict will be common­place in the next 5–10 years:

  1. Anomaly detec­tion: Alert the busi­ness to unusual activ­ity.  This is being used today to monitor user behav­ior, as well as to iden­tify poten­tial errors within your track­ing code.
  2. Calcu­lat­ing lift: Under­stand the return on invest­ment we gener­ate with the projects we tackle by measur­ing how a metric performs rela­tive to its fore­cast.  This is being used today to deter­mine if an adver­tis­ing campaign was able to increase revenue above our fore­cast or to measure if a site redesign effec­tively improved user expe­ri­ence.
  3. User Clus­ter­ing: Assign users into groups that share similar char­ac­ter­is­tics and/or behav­iors.  This is being used today to detect the users who are about to make a purchase, or users who are inter­ested in consum­ing certain types of content.

And these are only a few of the ways that data science capa­bil­i­ties are being applied to common digital analyt­ics prob­lems.

This Is How We Prepare

Today, our team at Search Discovery is making signif­i­cant invest­ments to expand our data science capa­bil­i­ties and solu­tions.  As part of this process, we’ve outlined three things that anyone working in our indus­try can do to start prepar­ing for the future we’ve outlined above:

  1. Take advan­tage of free online train­ing tools:  For SDI, this meant asking a few team members to iden­tify the best resources for devel­op­ing partic­u­lar skills, and sharing that list with the rest of the team.
  2. Join the commu­nity: Pay atten­tion to those members of the digital analyt­ics commu­nity who are currently apply­ing data science tech­niques in our field.  We recom­mend joining the Measure Slack commu­nity of analysts (join.measure.chat) and taking a look at dartistics.com.
  3. Exper­i­ment with ambi­tious projects: Seek out oppor­tu­ni­ties to apply what you’re learn­ing.  This might mean tack­ling a busi­ness problem or simply compet­ing in a Kaggle compe­ti­tion.

We have chosen a young, dynamic, and constantly evolv­ing field to work in with digital analyt­ics.  I hope you’re excited about that because it means that those of us who are flex­i­ble, paying atten­tion, and eager to learn have the oppor­tu­nity to drive all of us forward as a commu­nity.

If you’re inter­ested in seeing the full deck from my presen­ta­tion, or if you’d like to learn more about Search Discovery’s approach to digital analyt­ics, contact me here.