As we mentioned in a previous blog, Predictive Analytics is More than a Magic Act, one of the impediments to predictive analytics success are company-wide misconceptions. These misconceptions can arise before a predictive analytics roll out even begins.
A perfect example of a common misconception is the act of perfecting data quality prior to hitting the analytics stage, an act we previously discussed in reference to Computer World’s 12 Predictive Analytics Screw-ups article. Taylor dug deeper into the topic, highlighting the need for purified data in his “four traps of predictive analytics”:
“It is not true that companies need good data to use predictive analytics,” Taylor said. “The techniques can be robust in the face of terrible data, because they were invented by people who had terrible data.”
Expanding further on that point, (as Computer World also noted), data scientists deal with imperfect data all the time. Their job is to turn messes into squeaky-clean outcomes. They rely on time-tested methods to work around the issues to continue driving meaningful results.
The problem is that once an organization believes that its data isn’t quite ready for big time analysis, they pull back on analytics completely.
Taylor acknowledges that good data would certainly be preferable to bad data. But companies still need to start somewhere with these projects. Regardless of the condition the data is in, if it is still useful to drive a decision, that’s really all that matters.
Misconceptions are also born at the modeling stage. One of Taylor’s traps deals with organizations that think they need all possible analytical power to answer their questions build out models that are too complex. These models usually aren’t scalable, which leaves the organization with an oversized model instead of a one-size-fits-all algorithm.
Larger sized models aren’t always deployed because organizations simply don’t have the internal resources to make them work. That can leave a company high and dry on the return on investment scale.
Taylor recommends that organizations should refrain from investing such enormous upfront capital into their models. Instead, they should start small and pilot algorithms that can be scaled up based on the types of decisions that need to be made.
Do you agree with Taylor’s predictive analytics traps? Give us your thoughts.