Fleets can use predictive analytics to anticipate — and eliminate — errors before they happen.
As the amount of data available to fleet managers and executives has grown, many have incorporated exception-based tools to view events or metrics – such as speeds, fuel economy, mileages and detention times – that depart from the norm or the acceptable. For some, however, that’s not good enough. They want to know how to prevent exceptions in the first place.
“The cost of operations today – fuel, personnel, etc. – are such that the errors can be quite costly,” says Kathi Laughman, vice president of business systems and services for SCG, a 600-truck carrier based in Houston. “By the time you see an exception, the damage has already been done.”
Preventing problems before they actually occur sounds like a tall order, but in many situations it’s possible to analyze existing data in ways that will predict future outcomes. And with intervention, fleet managers can stop unfavorable results before they happen.
Unlocking your data
Fleet managers typically have numerous reporting tools that summarize past events and transactions, allowing them to take corrective measures. Less common, however, are tools that identify real-time or future trends.
Consider cargo claims. Most trucking companies use a cargo claims ratio (expense to revenue) as an important measure of risk. But this is very much a trailing indicator. At less-than-truckload carrier Saia, for example, the average lag time for customers to file cargo claims is between 30 and 60 days after the incident occurs. By law, customers have up to nine months to file a claim unless otherwise provided by contract.
“If you are waiting for a customer to have filed a claim before you look at issues, it is too late,” says Rick O’Dell, chief executive officer of the Duluth, Ga.-based company. To predict the risk of future cargo claims, Saia uses a measure called “exception-free delivery.” The measure shows the number of deliveries made free of notations for damaged freight.
A customer may sign a delivery receipt with a notation of damaged freight, but choose not to file a cargo claim. Even so, the measure is used as an indicator for the likelihood of future cargo claims, as well as a measure of quality, O’Dell says.
By analyzing trends in exception-free delivery, Saia is able to reduce its risk of future cargo claims and improve quality by taking preventive steps, such as pinpointing a terminal or group of dockworkers that need additional training for loading freight properly.
Saia has found something else that’s interesting and not necessarily as intuitive. “If the frequency of damages goes up and exception-free delivery deteriorates, the severity also goes up,” O’Dell says. “If you are damaging something often, the cost per claim increases as well.”
At its core, predictive analytics is a way to identify such relationships between past occurrences. Once identified, these relationships are used to predict outcomes and even automate decisions for reducing risk.
The discovery process begins by choosing a predictor – like exception-free freight delivery – or a combination of predictors.
Assessing the ability of a single predictor to predict an outcome – exception-free delivery for cargo claims, for example – is relatively simple. Correlating two or more variables to an outcome is a bit more complex. Suppose you wanted to determine the relationship between accident frequency and two or more variables, such as driver age and moving violations. You could use a statistical technique called regression to identify the link and use that information in building predictive models, explains Brett Vitrano, operations research analyst at Accelerated Freight Group, a 70-truck carrier based in Theodore, Ala.
Microsoft Excel has the basic statistical tools needed to build predictive models. (For an article describing this function, visit https://www.microsoft.com/microsoft-365?legRedir=default&CorrelationId=04bb8843-e3b1-45d2-b18c-6ad86c41f7c2 and search “Perform a regression analysis.”) Vitrano uses a more advanced application from SPSS Software. As a rule of thumb, he prefers a minimum of five variables for a regression analysis.
“I’ll take any variable as long as it relates,” Vitrano says. “The more data and more variables you use, the better you can predict what will happen.”
Some variables, like the price of fuel, are too volatile to include in a predictive model, Vitrano says. But AFG has been able to predict fuel consumption weeks and months in advance through a regression analysis of multiple variables such as idle times, weather conditions, freight volumes, etc.
“We have been pretty accurate,” Vitrano says.
Currently, AFG is compiling an accident database and is using regression analysis to determine what groups of drivers are more prone to accidents. The groups can be defined by a number of possibilities, such as male drivers between 35 and 43 years old, female drivers, younger drivers, older drivers, etc. Vitrano also has used regression analysis to evaluate the financial impact of possible changes to operations, such as increasing the fleet size.
“We have been able to make changes based on that information,” Vitrano says. “We have seen some very good successes.”
In many cases, simple spreadsheets and database programs such as Microsoft Access have all the features necessary for building predictive models to isolate risk. A driver scorecard can be used as such a model to predict future driver behavior.
In 2005, Sycamore, Ala.-based Floyd & Beasley Transfer Co. developed a driver scorecard as the basis for rolling out a new performance-based pay plan for drivers (see “Paying for value,” November 2006, page 65). Top management selected four categories or indices to use in the scorecard: average weekly revenue, fuel network compliance percentage, accident and incident ratio, and violations ratio. Each index was given a unique weight so as to prioritize the company’s goals for each category.
The scorecard and pay system has been an effective driver incentive, says Barry McGrady, vice president of information technology and human resources for the 200-truck carrier. When the program began in late 2005, 16 percent of drivers had scores that qualified for Level 1 and Level 2 pay; 84 percent were in the lower two pay levels. Today, 60 percent are in the upper group.
The system also has proven to be an effective predictive tool. McGrady found that drivers in the Level 3 pay tier accounted for 45 percent of the company’s overall turnover. In general, drivers in the lower pay groups also are more likely to have problems in areas not measured by the scorecard, such as refusing loads and being late for pickups and deliveries, he says.
Of the four categories, McGrady has found that adherence to Floyd & Beasley’s fuel network has the strongest correlation with performance in the other areas – good or bad. “By and large, the guys trying hard to be fuel-compliant are the ones that care,” McGrady says. “They are trying to do something that helps the company and the other drivers. They are probably having fewer accidents, too.”
Last fall, Wannemacher Enterprises developed its own driver scorecard that it uses to qualify drivers for a $300 bonus every 15,000 miles. To receive the bonus, drivers must meet all of the company’s goals for idle time, miles per gallon, safety, average miles per week and on-time percentage.
“We took all these things and figured out how it would be fair,” says Jeff Sacher, operations manager for the 50-truck fleet based in Lima, Ohio. To develop the driver scorecard, Sacher initially used Excel, but he since has moved the scorecard to Microsoft Access because the database functionality proved more useful for reporting and analysis.
“By working hand-in-hand with safety, I saw how much little things had been sliding past me (before),” says Sacher, who also recently began looking at the number of hard-braking events and found that drivers with poor scores in safety generally had more hard-braking incidents than average – an indicator of tailgating. Sacher now is considering adding hard braking as a category in the scorecard and driver bonus program.
For a handful of predictors, simple tools like Excel are sufficient. But you really need more complex tools to build predictive models that use dozens of data sources and variables. Sophisticated projects require more sophisticated tools, which generally are called business intelligence or data mining applications.
Ben Becker, chief information officer of Arcadia, Ind.-based Tradewinds, currently heads up a business intelligence project to build predictive models for the 140-truck truckload carrier.
“Right now, we have some very weak reporting structures, so our ability to apply mining algorithms and get creative with our analysis has been strictly hampered,” Becker says. “This will all hopefully be changing in the very near future.”
While Tradewinds is developing its own predictive analytics from scratch, there are some products on the market specifically designed for transportation. In 2005, Lafayette, La.-based Dupre Transport began working with FleetRisk Advisors, a firm that provides technology-based risk management services. Over the last two years, the two companies have worked closely to build a risk-based predictive analytics platform that uses data from across Dupre Transport’s enterprise.
“We are measuring many different factors and things that drivers touch,” says Tom Voelkel, president and chief operating officer of Dupre Transport. Some of the many variables included in Dupre Transport’s predictive model are drivers’ fatigue level, engine idle time and work schedule.
After months of building the model, FleetRisk Advisors in January began providing a monthly report that attempts to identify, with a “60 percent surety,” the group of drivers that will cause 60 percent of the fleet’s accidents the following month. With 650 drivers, having this report at the beginning of each month helps narrow the field considerably for deciding what drivers most need training. “This allows us to focus our resources and be most effective,” Voelkel says.
The report from FleetRisk Advisors also helps Dupre Transport select the best countermeasures for each driver on the list, based on what items are pushing up a driver’s risk factor. The local manager uses this information to work with drivers individually.
“The best results come from taking the driver outside of his or her environment to discuss the countermeasure,” Voelkel says.
Using the predictive analytics platform from FleetRisk Advisors has led to a tremendous improvement in the company’s “Big 4” frequency (measured in incidents per million miles), Voelkel says. “Big 4” accidents include losses from lane changes, rollovers, rear ends and intersection incidents. This year, the “Big 4” frequency has gone down by 20 percent from last year – which had been the best year in company history.
Solving business problems
Safety lapses represent the biggest financial risk for most carriers, but SCG’s Laughman has found predictive analytics useful in other areas. For more complex projects, Laughman uses a business simulation and predictive modeling tool called Performa to look at the impact of making certain operational changes. The Performa tool is especially useful in situations where cost is a component of the decision process.
“If you don’t do it, and you make decisions based on assumptions, it can be quite costly,” Laughman says. For example, SCG runs an expedited less-than-truckload operation with 1,800 dispatches a week. Suppose one terminal in Dallas is moving partial loads to Denver, and another terminal in Chicago also is moving partial loads to Denver. What would happen if the company redirected the freight from both of these lanes into a terminal in Kansas to create full loads into Denver?
“Now I can use predictive modeling to say, ‘Am I going to have a full load?’ That assumption is correct,” Laughman says. Performa also can be used to simulate the impact of such a decision on other processes and procedures, such as the cost of handling more freight, or on customer service.
“That’s where activity-based costing comes into play,” Laughman says. “You may in fact save on your line haul, but increase the cost in terminals. When somebody says they want to make a change, I can simulate that.” (For more on SCG’s use of Performa, see “Visualizing Growth,” November 2006, page 40).
Through predictive analytics, an ocean of data can become a steady stream of valuable, virtual reconnaissance that can help eliminate errors before they happen.
Information systems moving toward predictability Providers of fleet information systems continue to develop new features that help carriers quickly sift through mountains of data to find information that can be used as predictive tools.
Last year, Xata added a third-party business intelligence tool called Business Objects to both its Windows and Web-based Xatanet fleet management system. The tool allows fleets to access a data warehouse model for easy extraction and reporting of data. Fleets can compare metrics against timeframes, such as the number of miles driven in a particular speed range by driver, or the number of brake applications at different speeds, says Tom Flies, product manager for Xata.
PeopleNet recently introduced its Speed Gauge product that enables fleets to predict speeding patterns by comparing drivers’ actual speeds to the posted speeds along routes. PeopleNet also offers a feature to detect significant acceleration or deceleration events – indicative of tailgating and speeding.
In the near future, PeopleNet is looking to create a “master scorecard” for driver safety, performance and compliance by compiling a broad range of data such as hard braking, on-time percentage, hours-of-service violations and over-speed events. “Predictive is the way to go,” says Brian McLaughlin, PeopleNet’s executive vice president of marketing and product planning.
PeopleNet also is working with DriveCam for an even more in-depth look at risky driver behaviors. DriveCam uses a palm-sized recording device to capture audio and video snippets of risky events such as hard braking and swerving. By providing an immediate feedback loop to drivers of their driving, the frequency of incidents goes down by 30 to 90 percent, says Doron Lurie, chief operating officer of DriveCam.
Qualcomm Wireless Business Solutions (QWBS) and FleetRisk Advisors, a firm that provides technology-based risk management services, have entered into a strategic partnership to develop an analytical presentation module.
“Our customers collect a lot of data, but they do not have a good way to analyze and use it in a valuable and real-time environment,” says Bjorn Svinterud, QWBS director of business development. FleetRisk Advisors would provide the analytic “engine” for the content or answers, and rely on Qualcomm for the presentation and integration with fleet information systems, says Sam Wilkes, chairman of FleetRisk Advisors.
Developers of dispatch and enterprise software systems also offer many features for predictive analysis. Builders Transport, a 450-truck flatbed carrier based in Memphis, Tenn., uses its McLeod LoadMaster system to project its lane balance several days in advance so as to prevent certain areas from being overloaded with trucks.
Builders Transport also uses an advanced profitability and yield management tool called Netwise from Integrated Decision Support Corp. (IDSC) to project lane balance and profitability when evaluating whether to add new shipments to its network.
“We just try to be as accurate as possible to model what the future is going to look like,” says Dwight Bassett, chief financial officer for Builders Transport.