Predictive analytics is a powerful tool that requires strong management commitment in order to succeed. While it is tempting to say that everyone should use predictive analytics, the most successful users are companies that involve all departments in the process at an early stage, including operations, safety, HR, and technology.
“Successful users include companies that have traditionally very strong safety and turnover performance, as well as companies that want to improve in those areas,” said Chris Orban, technical services for FleetRisk Advisors. “The key thing is that all areas of the company have a desire to improve, regardless of the starting point.”
From FleetRisk’s perspective, predictive analytics is predicting the future with as much accuracy as possible.
“When we provide a model, it tells the customer with as much accuracy as possible what will happen in the next 28 days,” Orban said. “Not what has happened in the past, but what the most likely outcome for a specific driver is in the future, and why we believe that to be the case. While historical analysis can tell us a great deal about driver behavior, it cannot tell us with the same degree of accuracy what will happen next. That is our goal.”
To successfully utilize predictive analytics, a few key steps should be adhered to. “The fleet should have a good, solid set of data to work with, which can be accessed for the predictive modeling activity,” Orban stated. “It is critical that good quality data is available, which can be associated with a particular driver, in order to make the predictive modeling most effective.”
Unfortunately, many companies underestimate the quality of their data. It is important to balance the frequency of a predictive model with the resources available to take necessary action.
“A model that is accurate, but is run too often, can lead to liability when the fleet is unable to take action on those predictions,” Orban cautioned. “The best balance between accuracy and timeliness is to run the models every four weeks. This gives the safety and operational groups enough time to take action, while still accurately reflecting the changes in a driver behavior that indicate risk.”
Predictive analytics has been most successful in reducing accident frequency. “While our models more heavily weigh more severe accidents, and, thus, predict them at a higher rate than other, less severe accidents, our overall goal is to reduce accident frequency,” Orban said. “It is difficult to use cost as a model input because the same driver behavior that results in a $5,000 accident in rural Texas could be a $500,000 accident in Manhattan, N.Y. We focus on predicting and preventing as many accidents as possible, with the knowledge that we will prevent some of these high-dollar accidents, regardless of where they would occur.”
Fleets looking to utilize predictive analytics must buy into the process. Start by choosing the area of greatest need (e.g., safety, retention, or workers’ compensation). Next, a data assessment is required, identifying the key elements that will make up a customer-specific predictive model. Next, the necessary data is extracted for the models, which are built and validated by the fleet, to gain insight into what the data shows.
According to Orban, the best way to convince senior management that the technology is effective is to actually try it out. However, this does require commitment from the operations and IT groups, which will need to spend at least some time working to identify data sources and review the model results. “Once management sees the results, it is usually easy to move forward rolling the system out company wide,” Orban noted.
A fleet’s drivers are not often told that a predictive model is in place, but will see an increase in personal communication. “After using the program for several months, one client reported that drivers were now calling in to discuss safety proactively, looking for conversations when, in the past, safety had only been a source of punishment for a driver,” Orban noted. “This truly changed how safety was viewed by the driver population.”