Many of the current risk management strategies employed in the commercial transportation industry are either directly derived from the insurance industry, or are hybrids developed through industry consultation. The type of risk management strategies employed or required of commercial fleets varies according to insurance market cycles that are either hard or soft and tend to extend over several years.
As such, commercial transport operators often fall into two categories:
* Those that rely heavily on their insurance carrier for risk management strategies and in the process succumb to the vagaries of the cyclical nature of the insurance industry.
* Those that have established their own in-house risk management and safety program and control the management of risk and the cost of insurance simultaneously.
To demonstrate this point, let's take a look at the U.S. commercial motor-vehicle insurance portfolio over the period from 1995 to 2006.
From 1995 to 2002, the commercial motor-vehicle insurance industry lost money at a staggering rate during what was known as a soft insurance market. According to A.M. Best & Co. Inc., the insurance industry in 1999 paid out $1.18 for every dollar in insurance premium they collected.
After Sept. 11, 2001, the insurance market took a dramatic turn, which was characterized by a significant reduction in insurance capacity and a sharp increase in reinsurance rates. This helped insurance companies return to profitability and began the hard insurance market cycle, which took full effect in 2002 and has lasted well into 2006.
REAL ACCIDENT CAUSATION
While risk managers and safety specialists in transportation believe they are in the business of managing risk, they might merely be managing claims and premiums. Most insurance companies don't have access to all available driver, vehicle and operational data regarding accidents and, therefore, have to mostly rely on subjective claims and loss-control engineer reports.
For example, the same truck accident description could be coded in an insurance company database in numerous ways, such as "hit guard rail," "run off road," "rollover" or "upset"; yet, they could all be caused by the same thing--the driver fell asleep at the wheel.
The transport company and insurance company database often records the description of what happened. It does not necessarily record what caused the accident, and, therefore, the insurance company underwriter, loss-control engineer, transport risk manager and safety manager might overlook the most beneficial corrective action.
Predictive modeling risk management in the transportation industry can change this paradigm by implementing a driver performance program that objectively concentrates on historical and real-time data to predict future events. This would allow transport operators to better understand risk causation, improve their level of self-insured retention, and in the process improve the costs of insurance and claims.
But the substantial investment that many companies have made in fleet management technology and the efforts of equipment manufacturers and service providers to deliver "information" has created unprecedented amounts of valuable electronic data. This data traditionally resides in multiple disparate data storage systems. The challenge for many transportation companies and their vendors is that this data is collected, updated, stored and analyzed in total isolation.
The many forms of data available include driver and other personnel demographics, sophisticated psychological profiling, human physiology and alertness measurement, location, speed, direction, weather, workload and diagnostics engine management, lane departure, brake management, electronic engines, smart and adaptive cruise controls, head-up displays and automatic gearboxes.
This electronic data can be moved from disparate locations across a company's network and stored in a relational database that is capable of identifying the very distinct patterns within all of the noise that massive amounts of data produce.
Once organized in a relational database, managers can identify distinct, often counterintuitive, patterns about how their fleet works while adding a layer of value to systems already in place. Individual factors in a predictive model could include a monthly fatigue score from electronic log analysis, driver alertness, driver's psychological profile, age of trailer, number of deliveries per load and dispatcher skill level.
Taken together, these elements might reveal the risk signatures of all prior accidents over a three-year period. Once understood, the data could help rank drivers (using current data) on the likelihood of having a future accident. That intelligence can then be used to protect the fleet from the risk of financial loss through predicting and proactively managing any number of intuitive and/or unexpected risks.
Once all of the electronic data a fleet has to offer has been gathered into a common database, it is possible to mine that data. Numerous levels of data analytics can be used to identify true accident and risk causation.
For example, the frequency of accidents could be reclassified by time of day, day of week, event description and causal factors. Additional classification of claims is also required to identify the level of driver control of the vehicle at the time of the accident. Analysis that classifies driver control under "loss of control" and "poor judgment" adds a new field that will identify what level of control the driver had over the vehicle at the time of the accident.
Once implemented the business intelligence platform will produce information that includes a variety of new and often counterintuitive contributing factors that characterize defined events within a fleet. The manager's response to the data will generally fall into one of two categories:
* The data element will be a direct reflection of the driver's performance, and as such the intervention with the driver will relate specifically to that data element. An example would be where the predictive model picks up that a certain driver has a habit of speeding more often than is permitted. The solution in this area may be to simply speak to the driver about their speeding tendencies, identify the cause and suggest a strategy to reduce idling time.
* The data element will be an indirect reflection of the driver's performance, and as such the intervention will relate to a different set of factors. An example would be where the predictive model picks up that a certain driver has a habit of varying his shift start time each day, possibly due to problems at home. The solution would be to speak to the driver, identify the cause and to find out why they are either early or late.
DATA, RISK AND DUPRE
Deciding on a course of action for improving driver performance will most likely encompass programs already in existence at the company. For example, they could include physiological programs that optimize driver alertness, behavior modification programs that improve performance and professional development programs that improve a driver's skills.
In April 2006, Dupre Transport, a provider of transportation and logistics services based in Louisiana, implemented a predictive analytics model utilizing the system described in this article.
At the end of each month, the predictive model ranks current drivers against a statistical and demographic profile of Dupre Transport's theoretical best and safest driver. Each tier represents one-third of the current driver population. Tier-1 drivers represent the "best and safest," while Tier-3 drivers represent those needing the most help in improving their performance.
At the end of each month, actual accident results are compared to the model's predictions. Tier-1 drivers have been responsible for only 10 percent of all preventable accidents in the aggregate, while Tier-3 drivers have been responsible for 70 percent of the overall accidents. This predictive segmentation allows Dupre to focus its training and risk mitigation efforts on a small subset of their drivers, rather than the entire driver team. Based on the initial results of the production system, Dupre's management team designed a suite of targeted risk management and performance-monitoring strategies which include:
* Focused driver training programs
* New safety field representative training
* Terminal and management scorecards
* Driver scorecards and enhanced incentive programs
* Alerts from in-vehicle technology
* Dispatch training and incentive programs
* A modified recruiting and retention program
The early production results at Dupre are very exciting with the frequency of preventable accidents now down 44 percent since the driver performance model went into production. Of those drivers identified by the model as needing the most help to improve their performance, 51 percent have improved their model score and driving performance, indicating a positive response to management help.
So far the most dramatic intervention has been to simply talk to drivers about their lives and how it effects their driving performance. The key with the FleetRisk Project is being able to help Dupre identify not what to do with all drivers, but with whom, when, where and why in a targeted and timely application of existing resources.
is chief product architect for FleetRisk Advisors.
April 1, 2007
Copyright 2007© LRP Publications