The ELD Mandate leveled the playing field, but now that the majority of carriers in the U.S. have electronic logging devices (ELDs), fleets are asking themselves what they can be doing with all the data that they’re collecting.
The first and most obvious way fleets can utilize ELD data is to create reports. Managers can see which drivers have the most violations, where violations occur, and then coach and make recommendations. Of course, everybody is doing that. So, what can fleets do to really take advantage of their ELD data and gain a competitive advantage and improve operations? The answer could be in predictive analytics.
Analytics are pervasive. We see them throughout our industry, and every industry, in all shapes and forms. The simplest form of analytics is insurance. Insurance companies want to know about your health, your family track record, your age — basically, they look at your past and try to predict the future from it. And everyone knows about car insurance and how a behavior like driving a red car could indicate reckless driving.
Predictive modeling works essentially in the same way. It looks at past events and tries to predict future events. Most fleet managers conduct reporting and analysis in some form, looking at data to determine what happened, why, and what to do to prevent it from happening again (if the event is negative). More and more fleets are using score-carding and dashboards today. Managers take that information and try to develop correlations to predict what will happen. But, correlation is not causation.
How, then, can fleets accurately predict when their drivers are going to resign, when they might get into an accident, and what to do about it? With predictive analytics, a large sample of data is collected in order to identify patterns in a fleet’s history that can predict the future. Active drivers are measured against these predictors.
Productive Driver Conversations to Reduce Accidents
The key to a predictive model is to be able to prevent something like an accident or driver resignation from ever occurring. If you can prevent them, you don’t have to worry about how to classify them or analyze the ‘why’s.
Looking at extensive data in Omnitracs’ predictive modeling warehouse, somewhere around 80 percent of all bad things that happen in fleets happen with 20 percent of their drivers. And in the top 10 percent of a fleet, 31 to 46 percent of all bad things happen. These numbers can help fleets zero in on that 10 percent of drivers and mitigate potential risk factors.
Mitigating potential risk factors, it could be argued, is both an art and a science. The modeling itself is a science that tells you which drivers are at risk and what to do about it — and how to go about remedying it is an art. After identifying the 10 percent of drivers, managers need to know how to have a productive conversation about the behavior. This may seem simple but these conversations are not always productive.
What likely sticks out in driver conversations are questions about where the driver is, when they will be empty, when they will get somewhere on time, etc. These transactional conversations are second nature to driver managers. Having a proactive and positive conversation is new to many managers and organizations. Proper training can help fleet managers have more meaningful, productive conversations with their drivers.
When using predictive modeling technologies, managers record a quick synopsis of their conversations with drivers. This is done for continuous feedback into the model itself for improvements as well as utilizing text analytics, which can be thought of like word clouds. For example, in Omnitracs’ predictive models, it was discovered that when the word ‘back’ was in the same message that is evidenced by a delay of an hour of more, there was a 600-700% increased probability of a driver having an accident.
Building an Accident Severity Model
To build an accident severity model, there should be ELD data across clients that gives a good amount of historical perspective. These datasets are too enormous for the human mind to comprehend. To segment the data, all accidents in a dataset can be reclassified into either distraction or loss of control.
Most conventional safety programs deal with driver behavior such as checking mirrors, keeping two hands on the wheel, and avoiding distractions like phone usage while driving. Loss of control, on the other hand, is physiological and can be defined as when the body is awake but the mind is asleep. When this happens, usually due to fatigue or sleep abnormalities, loss of control can occur.
According to Omnitracs’ Senior Director of Analytics and Modeling, Lauren Domnick, a root cause analysis found that it is the lack of responsive time that makes accidents so severe. The “big six” accidents that can be defined as severe and result in big dollar expenses are roll-overs, run-off-road, head-on, jack-knife, side-swipe, and rear-end. These are accidents where the driver is disconnected from the driving task, takes zero evasive action, could have seen the point of impact six to seven seconds prior if awake, and make no attempt to minimize damage at the point of impact.
Using ELD data to make business decisions
A good way for a fleet to understand how well it is doing is to compare it to other fleets. Big data analytics makes benchmarking possible. Fleets can be objectively compared using standardized Hours of Service data. Not only does this allow fleets to see how they’re doing compared to peers, but it also helps fleets uncover underlying issues.
A fleet could see through data that their refrigerated vehicles start the day earlier than other refrigerated vehicles in the same local time zone. At first glance, it might seem that the drivers are putting in more hours if they are starting an hour earlier than their peers. Digging in deeper, a fleet manager might observe that those same vehicles are also ending an hour earlier than their peers. This could lead to a discovery such as parking issues that drivers try to prevent by altering their hours. One this is discovered, fleets can take measures to implement parking programs or try to solve other issues. Issues like drivers spending more time driving and less time on-duty not driving. Or drivers leaving hours on the table that could be utilized more efficiently.
Analytics from HOS data allow fleets to gain greater insights into their operations and make better business decisions. Most fleets have already invested in electronic logs. These are data points that can be used for benchmarking and leveraged to gain deeper insights into safety, turnover, and other operational issues.