Explaining optimization

Business leaders often say they’re data driven. Conceptually, using data to drive decision making is easy to grasp. But what if the data doesn’t improve the desired outcome? I’ve built machine learning models to optimize outcomes, and I’ve learned better ways to explain this concept.

DEFINE THE RELATIONSHIP

What is the relationship between the data and business objective? For example, you may find yourself discussing health in the context of improving revenue. Someone wants to know the health of a sales opportunity or the health of a customer. Health can mean different things to different people. But if it doesn’t directly relate to the outcome that drives revenue, you may have unintended consequences. Once there’s understanding of how data relates directly to an objective, it can be optimized. 

REDUCE ERRORS

Broadly speaking, optimization comes down to reducing errors, especially when they’re costly to the business. Better decision making means less expensive errors. What can we expect from someone making “gut” decisions? Probably some mistakes, and certainly no understanding of a baseline for accuracy. On the other hand, imagine producing a data point that’s 100% accurate, removing all possible errors. In reality, there’s a solution between “gut” and “100% accurate”. If you’re a leader who claims to be data driven, do you know the current baseline of your decision accuracy and do you have a team working to improve it?

Using machine learning to optimize business outcomes, a model plays two roles. The output provides the data point for decision making, the inputs provide a path to prescriptive analytics, and errors are minimized along the way. Compared to the alternative, leaders with machine learning solutions in their business processes should feel confident in their data driven proclamations.