Predictive Monthly: The Role Of Predictive In Your Analytics Strategy

Ashish Morzaria

Two of the hottest terms in the business world right now are “predictive analytics” and “data science.” We know that they are the keys to making better decisions and ultimately improving business performance, but how do they actually do that? And if they are so crucial, why isn’t everyone implementing them?

Part of the problem is that people bring their own preconceptions that lead to wildly different definitions and expectations. Another problem are the terms themselves: are other forms of data analysis (such as business intelligence) not “science” but just “data art”? These terms draw an artificial distinction between descriptive and predictive analytics that creates a very real barrier when businesses develop an analytics strategy.

Are descriptive and predictive analytics that different?

Let’s take a step back. The core mission of any business intelligence system is to analyze data, extract insights, and enable the user to make the best decision possible. I bet if you asked a data scientist, they would argue that predictive analytics has the exact same mission. Obviously these two analytical approaches are not replacements for each other – they are in fact yin-yang complements.

Most of you have seen the “infamous” analytics maturity curve that characterizes business intelligence as “descriptive” and data science as “predictive” analytics. This is based on the notion that BI is all about analyzing the past and predictive is all about the future – but how does predictive analytics determine what the future looks like? Yup – by analyzing the past.

The real difference between descriptive and predictive analytics is simply in how they achieve their goal of surfacing insights. BI tools make heavy use of visualization techniques and rely on human interaction to drill, slice, dice, and draw data, whereas predictive tools leverage mathematical algorithms guided by humans to look at data in ways and at speeds not humanly possible.

Barriers to predictive analytics

The real barrier preventing many enterprises from adding predictive analytics to their existing analytics landscape is the perception that it will require a mountain of cash, a library of knowledge, and mythical data scientists that everyone knows are scarce. This is where those nasty preconceptions come in. We’re trained to consider everything as a cost, to de-risk projects so far they are no longer innovative, and to prove tangible value before we even start. Now is a good time to start unlearning those bad habits!

Predictive analytics is not a TCO (total cost of ownership) discussion – it’s an ROI (return on investment) one. How? Consider that you had to make a very important business decision, and a magical genie popped out of a lamp to tell you with great certainty which option will make you more money. Would you want to know how much money you could make? Or would you ask how much the lamp costs first?

That said, predictive analytics isn’t a genie in a lamp – but neither is a data scientist. Even if your business employs data scientists, they likely do not have the domain knowledge for the business problems you are trying to solve. This is where that nasty barrier between descriptive analytics (BI) and predictive analytics (data science) is especially detrimental. By considering these two disciplines as separate, it is very difficult to have a coherent analytics strategy that leverages the best of both worlds.

Adding predictive to your analytics strategy

Fortunately, integrating predictive into your existing plans doesn’t have to be difficult, and best of all, it doesn’t always require a data scientist (although they can be useful if you have them). The key for determining where to start is by focusing on a defined business problem that is small, yet important. It may be a problem you don’t know how to solve, or one that you are currently struggling with by using a BI solution.

For example, the question “Why are my customers canceling their service with me?” is a common problem that gets to a BI analyst’s desk. The BI approach is to slice, dice, drill, and graph characteristics shared by those who have already cancelled to filter at-risk customers and reach out to them before they call us to cancel. We rely on the analyst’s knowledge of the market, customer base, data set, and personal experience to choose how to successively narrow the data set.

The predictive analytics approach is to “train” a predictive model by feeding it data on customers who have and haven’t canceled so the model understands each characteristic of the customer and how much each would contribute to their decision to leave us. We can then use the predictive model to “score” each customer on their propensity to leave and even understand the key influencers to their potential decision. The BI analyst can then filter on the score – a weighted variable that takes all of the customer’s characteristics into account, resulting in a far more accurate prediction than cascading filters in a BI report.

Making it happen

You may be wondering how to “train” and “score” and do all that predictive stuff without a data scientist. Automated predictive algorithms have come a long way and encode many of the steps that data scientists currently do by hand or through scripting today. Think of these solutions as a “Data Scientist-in-a-Box.” At the core are automated predictive algorithms that been refined with data science over many years, both in the mathematics lab and in the real world, with hundreds of customers who rely on them for mission-critical decisions every day.

While it may sound like automated predictive analytics is only for the non-data scientists, it’s quite the opposite. Automated predictive technologies make quick work of the boring, tedious, and error-prone tasks that make up “data science.” A skilled data scientist can tweak and optimize automated algorithms to make them even better, and at a fraction of the time it would take them to do even a basic predictive model by hand. This frees up the data scientist’s precious time to work on more projects and provide value to more parts of the business.

Regardless of where you are in your analytics journey and whether you have data scientists on staff or just really smart business intelligence experts, there’s a 100% chance that predictive technologies can either remove some analytical obstacles – or at least make them much easier to conquer.

Get a complimentary Forrester paper and read more on how an emerging class of software – insight platforms – can better connect data to action, which is what your business really wants.


Ashish Morzaria

About Ashish Morzaria

Ashish C. Morzaria, Director of Advanced Analytics for SAP, is responsible for SAP’s Predictive Analytics portfolio strategy and driving its go-to-market efforts. He has been with SAP (and BusinessObjects before it) since 2005 in a variety of Product Management roles across the Analytics portfolio including the SAP BI Suite and Platform, SAP Lumira, SAP’s BI Cloud strategy, and numerous acquisitions. Prior to his career at SAP, Ashish spent almost a decade as a product manager, development team manager, and software engineer.