Machine Learning and Business Are Made for Each Other

Mike Romeri, Mar 18, 2018 8:06:29 PM

Data science has been around for a long time. What has changed is the vast increase in the volume of available data.  As more data becomes available, analytics perform better, which mostly means that we can better predict the outcomes of alternative decisions and actions than ever before.

In past projects, we often found that decision makers made high-impact decisions, while knowing that their ability to predict decision outcomes accurately was, in fact, very poor. Using analytics effectively can increase predictive accuracy (i.e., % of accurate predictions) by as much as 1/3. Better predictions enable better decision making, because decision makers can more reasonably make decisions based upon their predicted outcomes.

Leading Companies Have Developed Reliable Best-Practice Techniques for Applying Advanced Analytics 

They launch projects to investigate the relationships between internal and external data drivers and their targeted objective(s) (e.g., revenue, profitability, working capital, etc.). Here is how they do it:

  • Usually, the drivers reflect some insights about human behavior, such as:
    • Consumers spending patterns change with the weather, the time of day and the day of the week
    • Price changes usually influence buying decisions
    • Changes in the economy or unemployment levels drive changes in consumer spending
  • Historical data sets are used to test and optimize machine-learning predictions; this technique is most often called “training the model”
  • Real-time predictions can be made by considering current values of the known performance drivers

Once decision outcomes can be predicted more accurately, it is worthwhile to vastly increase the number of decision alternatives that are considered. Using advanced predictive analytics, it is not uncommon to systematically consider thousands or even millions more decision alternatives. The most important thing to remember is that analytic predictions are almost always far more accurate than any other kind of prediction. The predicted outcomes of decision alternatives can then be used confidently to select the superior decision alternatives. 

Delivering analytically-optimized decision recommendations at the time a recurring decision must be made is the essence of augmented analytics. The recommendation made reflects best-practice insights, as they apply to the decision being made and all the current operating conditions.