• by David Baker , Op-Ed Contributor, March 2, 2015

I was watching a short video by Babson College Prof. Thomas Davenport on the Harvard Business Review website recently and found it to be a good, simplified explanation of analytics. 

(Stop! Don't swipe the screen to another email just yet. This isn’t another quant geek's discussion on the merits of causation and correlation, and it won't make your eyes glaze over.  I’m writing this because the term "analytics" often gets generalized in industry surveys and analyst reports, as in "80% of marketers will invest in analytics next year."  What does that mean?)

Where does the Big Data discussion fit into your analytics discussion? Analytics generally is the hardest thing to sell to a budget-conscious marketer, because it's vastly ill defined in companies and hard to translate to terms or just things we do every day.

I loved how Davenport described the three basic types of analytics:

1. Descriptive analytics is mostly characterized in many Big Data discussions today. You have a lot of data but don’t know how to define it, organize it or tabulate it. We also interchange this term with "reporting." It's valuable information as a foundation, but it doesn’t tell you much about why the result happened or what might happen in the future.

2. Predictive analytics is what's getting so much attention today. Here, you “use data from the past to predict the future,” according to Davenport. Don’t we all wish we could do this?

Skeptics hammer this concept because they insist correlation doesn't mean causation.  You might have the propensity to buy a hot dog from 7-Eleven at the same time you're buying a Gatorade and a lottery ticket, but the reality of you buying a hot dog on your next visit, or any visit, is likely zero.

As Davenport says, "You don’t need to imply causation to apply predictions." You're simply predicting a likelihood of an action. For example, a certain type of customer might respond better to a certain type of email or product recommendation.

This is an important concept to understand whether you are using a model, a recommendation engine or simple business rules based on past behaviors. Predictions are just that -- choices  --and in a transient world moving faster, you’ll need to rely more and more on these.  How far you stretch them is your RISK model.

3. Prescriptive analytics is what I think about on long walks. This is often where cause-effect analysis meets the real world. We mask this as "testing" in the marketing world. We all know the No. 1 rule of testing: You must have a hypothesis to test against. 

Think of this area like a doctor writing prescriptions: fine if you are treating a common cold, but if you are trying to ascertain the relationship between increased purchase, profit and number of ad exposures over a given period of time by channel and segment, this becomes almost unfathomable operationally without a quant-geek team spending months on it in the back room.  

So, what does it mean for you tomorrow?  

Many discussions on the future are rooted in the questions you want to answer, how often you want the answer, from whom you want the answer and how you want to apply the answer to your business.  The importance of knowing this is what I’ve harped on for years:  We have more data, more transient consumers, more tools to choose from, and are expected to be faster and smarter than we were yesterday, all with the same budget. 

Ten years ago, it would take two to three months to build a predictive model. Now we have hundreds of thousands in play, even some that self-learn. 

Five years ago, compiling a simple cross-channel report was a multi-functional effort. Now we have real-time dashboards that we can configure on the fly. 

Last year we dreamed of automation and optimization with high degree of confidence in the outcomes. Now we can apply machine learning to in-market programs to take potential error out of what we know works.  Tomorrow is here.  As this quote from engineer and author W. Edwards Deming implies: "In God we trust; all others must bring data:"