The art of decision-making
Why predictive data models aren’t enough
Adam Kamp | May 8, 2015
Our brains are constantly calculating probabilities, whether it’s deciding on what to have for lunch or to tell our wife what we really think of her new dress. Prior research on how we make decisions pointed to skills developed in school and inherent common sense, but today’s neurologists have shown that we actually make microsecond calculations based on a combination of factors, including experiences.
Paralleling this change in thinking, the best businesses are altering their strategies by investing in their data, reporting, and advanced analytics. Not only are these businesses leveraging internal subject matter experts to make their best guesses, they’re also empowering their employees to use data to reduce the risk of making poor business decisions.
Once an organization cultivates rich data and understands its importance, it begins to compete on a different playing field. In Competing on Analytics, authors Jeanne G. Harris and Thomas H. Davenport share how two widely recognized brands, Dell and Progressive, have risen above the competition by focusing on analytics.
The book notes that during the last recession, Dell was able to identify how industry sales were going to drop months before it happened thanks to predictive modeling. “They took preemptive action on prices and products, which resulted in better … financial performance during the downturn. After the recession ended, they were able to readjust, resulting in increased market share.” The authors go on to talk about how Progressive was the first insurance company to offer auto insurance online in real time, allowing consumers to easily compare rates. “The company is so confident in its price setting that it assumes that companies offering lower rates are taking on unprofitable customers.”
Allowing the brightest people to create data-based models minimizes risk, maximizing the possible returns. But predictive models aren’t foolproof, as illustrated in Nate Silver’s The Signal and the Noise. Silver shares a cautionary tale about a dam that protected a nearby village from flooding. The dam needed to be a certain height to accommodate the highest level of rainfall during the worst rainy seasons. A data model considered the likelihood of flooding based on climate and other foreseeable factors, but failed to factor in human experience. Unfortunately, a rainy season came along that overwhelmed the dam and the village was washed away. While the model was accurate most of the time, it only took one bad rainy season to overflow the dam.
Silver does not divulge how many years the dam successfully protected the village, but it only took a few extra feet of non-forecasted water to cause a catastrophe. Wouldn’t it be reasonable for someone to ask, “what if the model breaks?” Ideally, the village would have utilized the best models available, put them in front of the most experienced minds, and a contingency plan would have been implemented to avoid the town’s decimation. This is a dramatic example for business use, but the lessons learned apply.
Let’s look at another plausible example: a major food distributor forecasts standard probabilities for food needed in a given city for the next month. The model likely accounts for past delivery needs and factors in population change and recent economic forecasts, and is likely accurate (within a reasonable margin of error) the majority of the time. That said, wouldn’t it be helpful to know that next month a large sporting event is coming to town that would require larger shipments of certain goods (e.g., snacks, beverages, paper products, etc.)? The model doesn’t know this, but it would only take one person in the know to bring this factor into the equation.
Allowing people and data to work symbiotically maximizes your chances of making the best decision possible. There are no guarantees in business, but limiting potential risk and maximizing opportunity for gains comes down to optimizing your available resources.
Adam Kamp has been with Slalom’s IM&A practice for over two years. He brings cross-industry experience to his projects, with an emphasis on advanced analytics.