Why Your Data Efforts Fail

Mar 03, 2017

BY NICK MARKS – DATA SCIENTIST

For millennia, kings, generals, scientists, and explorers led and navigated the world with scant information and general approximations. Still, they employed a familiar methodology: “What has already occurred? What do I want to happen? What do I expect will occur if I maintain current course? How can I change my future? Now, act.”

Notwithstanding oceans of data pouring into companies, many business leaders today are still using approximations and “gut” to make their decisions—though not for lack of effort. Companies spend a fortune on Big Data and analytics, but too often with little return beyond utterances such as “that’s interesting” and “hmmm, we’ll need to consider this.” The problem lies in a lack of business-data alignment. Let’s start at the beginning.

There are three major buckets of analytics a company can and should engage in.

DESCRIPTIVE ANALYTICS

Descriptive analytics is the most basic kind of data analysis (although it can become quite advanced). Almost all companies have some form of it going on: dashboards with sales figures, weekly updates on performance, etc. Unfortunately, many companies fail at this very first step. Good descriptive analytics not only inform a business leader as to what happened, but also give him or her the “why” that enables action. For a firm with new products targeting baby boomers, sales growth by age bracket might be the important “why” question to ask. Meanwhile, a conglomerate diversifying overseas might want greater focus on differences between geographic regions. An analytics department operating in a silo will only answer the questions they ask themselves, and these may not always align with the “why” questions a business leader needs answered.

PREDICTIVE ANALYTICS

Predictive analytics is the next logical step once you know what’s going on: what will happen? With a solid understanding of the past and present, data scientists can build models to predict the future. Airlines need forecasts of passenger demand to know how many planes to buy. Hospitals predict daily patient inflow to plan staffing. Restaurants forecast food consumption to optimize food supply purchases. With a plethora of modeling options, and incredibly granular and accurate data, businesses have never been better positioned to plan for the future. And yet, as with descriptive analytics, pulling this off takes more than a shout down the hall for “Models!” to the data science team. A model always needs a raison d’être —a business case— otherwise it’s just a fancy math problem. If business leaders can clearly articulate the model’s purpose, data scientists will be much better positioned to select an appropriate and effective methodology.

PRESCRIPTIVE ANALYTICS

It’s apparent that without strong business guidance neither descriptive nor predictive analytic efforts will succeed. While the exact definition is debated, prescriptive analytics boils down to developing an analytic feedback loop that effectively uses descriptive and predictive analytics, plus human input, to inform real business decisions.

Human input—that’s the key— is where the loop occurs. After digesting the outputs of descriptive and predictive analyses, a business leader must act, and then just as crucially, measure the outcomes of their actions. By acting, and then measuring, a business links their actions back to their descriptive analytics framework and the whole process begins again—improved models, improved actions, and continual measurement.

In many ways, prescriptive analytics resembles the classic “test and learn” framework, applied on a grand scale. It requires a systemic approach to business and analysis; not one driven solely by the individual brilliance of modelers or the business savvy of executives, but rather by the organizational determination of leadership in both areas. If done right, the data scientists and business leaders will become more closely linked than ever before, leading to better decisions, better data, and most crucially, better outcomes.