Mastering descriptive data analysis yields better predictions

TTlogo 379x201 Mastering descriptive data analysis yields better predictions

Big data and business analytics may not be considered ubiquitous quite yet, but they are getting there.

Total big data revenues over the past several years have grown exponentially and will approach $ 50 billion by 2017, according to business technology consultancy Wikibon. Forbes cited a 2015 Capgemini global study predicting a 56% increase in big data investments over three years. And Computer Science Corp. estimates data production overall by 2020 will be 44 times what it was in 2009.

The analytics needed to work with all of that data is growing just as fast. But analytics comes in many flavors, with the descriptive and predictive varieties being the biggest and most useful. Yet, of the two, descriptive is embraced far more by businesses than predictive.

Today, 90% of organizations use some form of descriptive analytics, which includes methods of mining historical data as well as real-time streams to extract useful facts to explain the data. The functions employed by descriptive data analysis include social analytics, production and distribution metrics, and correlations between operational results and changes in process.

Peering into the crystal ball

Predictive analytics involves processing big data to forecast future outcomes beyond the simple trending produced by business intelligence. It allows the enterprise to game out complex what-if scenarios, create accurate models for future performance, identify correlations that aren’t intuitively obvious and perform more thorough root-cause analysis. With these capabilities, an organization can forecast customer/client behaviors, predict logistical failures, anticipate changes in purchasing patterns and make more accurate credit/procurement decisions.

Descriptive data analysis is considered fairly easy since it can be implemented with the standard aggregate functions built into most databases and knowledge of basic high school math. In contrast, predictive analytics calls for a strong command of statistics, college-level math (linear regression and so on) and often specialized software. Most organizations have the in-house resources to do descriptive analytics, while predictive analytics requires recruiting specialists and very often the purchase of new systems.

Yet the gap between descriptive and predictive ostensibly isn’t as great as it seems. The resulting data from both methodologies is gathered for the sole purpose of answering questions, albeit different questions. For descriptive data, it’s “What has happened?” and for predictive data, “What might happen next?”

An organization with a solid grasp of descriptive analytics is well on its way to embracing predictive. The reason is simple: Predictions are generally granular, focusing on the behavior or performance of one unit or individual among many—for example, “What will this person buy?” “Is this customer a credit risk?” Descriptive data analysis creates the rules and conditions that lead to accurate predictive analytics. You can’t have the latter without the former. With well-constructed descriptive models, predictive models are much, much easier.

Descriptive models take large amounts of data and employ well-crafted rules of classification for that data to organize many units or individuals into useful subgroupings. Descriptive models condense that data into factors that identify certain characteristics of the people or processes being assessed that predictive models require.

Descriptive analytics is generally context-free — no correlation to other data is usually involved — while predictive analytics is all context. The accurate assessment of what a customer might want or need is greatly enhanced by knowing what the customer is doing. Customers with similar profiles and shopping patterns may vary wildly if the context of their shopping changes. Teenagers, for instance, buy different clothes in different quantities if they’re about to leave home for college. When context is factored in via predictive analytics, the result is highly refined forecasting.

Decisions times two

Descriptive data analysis, therefore, is the first step on a journey that leads to predictive modeling power, an enterprise-transforming combination. The outcome of combining these two methodologies rests with the ability to create decision models.

A decision model incorporates all the information necessary to generate an actionable decision from the output of the descriptive model and the predictions, and past decisions, it generates. That increases the accuracy and efficiency of decision making by enabling optimization — the ability to fine-tune processes and institutional behaviors based on the successful implementation of analytics.

This decision model leads to the next level — prescriptive analytics — a methodology for choosing effective courses of action from available options, and a very different beast from descriptive and predictive analytics. The point is that no branch of analytics exists in isolation; each methodology feeds into the next and adds a new layer of sophistication and functionality to the process.

The ultimate goal is to view analytics not simply as implementing a new process or tool, but more as important steps in an enterprise’s evolution on the way to perpetual growth and change.

Let’s block ads! (Why?)


SearchCRM: News on CRM trends and technology