Driver dispatching is fundamentally a manual process, with humans sitting at arrays of screens making decisions.
A tantalizing goal is to automate this process as much as possible. There are parallels in the automotive industry, where traditional manufacturers have evolved from the early days when the automotive assembly was entirely manual, to modern plants where there is a growing presence of robots that were initially used for painting and welding.
While the older companies have steadily introduced robots when they could demonstrate that they were more effective than people, companies like Tesla started from the assumption that the assembly line would be entirely automated, and where humans are used only where a robot cannot do the work (handling wiring harnesses is one example).
It is tempting to assume that truckload motor carriers are closer to the traditional automotive manufacturers, but don’t forget about Uber (and Uber Freight), which both use completely automated dispatch systems. Both are solving simpler problems which makes this easier, but it is important not to fall into the trap that “we cannot do this.”
It helps to recognize all the benefits that can be traced purely to automation, as opposed to how well automation performs compared to people, These include:
- Robustness to turnover of driver managers. A high-quality, experienced driver manager can do a very good job, but if they leave, those skills are simply lost and their replacements can initially be quite poor. It takes time for them to learn the job, and as with any skill position, some are better than others.
- Improved productivity from driver managers. As a carrier grows, having to grow the dispatch floor adds to payroll costs, facility overhead to hold the staff, and increased management costs. With automation, we might triple or quadruple the number of drivers a single manager can handle.
- Consistency. A room of 20 driver managers means 20 styles for dispatching drivers. Each will have their own skills and biases toward serving the driver, serving the customer, and serving the company. Differences in styles will reflect a mixture of the personality of the dispatcher along with the drivers and customers they are serving. A computerized dispatch system will use consistent rules, adapted to the requirements of each driver and load, along with the characteristics of each region.
- Simulate first, then implement. It is impossible to test different policies for dispatching drivers when people are implementing the policies. By contrast, we can simulate changes in dispatch policies using the strategic platform in CORE.ai, making it possible to tune policies before implementing them in the field.
- Adaptability to changing conditions. Companies will face changing market conditions: driver shortages require a greater emphasis on driver retention, while market downturns may require spreading available work around a fleet so everyone is making enough money to get by.
- Self-dispatch - An automated system can produce a ranked list of loads that could be displayed to a driver (say, on a smartphone), further reducing the need for dispatchers.
- When a dispatcher makes mistakes, corrective actions may be taken to improve future performance. However, this training benefits just one person and is lost when they leave. If a computer model makes a mistake, the modification immediately applies to the entire company and becomes a permanent improvement to the process.
- In addition to the steady improvement as you work with the dispatch system, you will benefit from the improvements accumulating from the combined experiences of all the other carriers using the system.
- Facilitating growth. A growing carrier has to hire and train new driver managers.
- Performance metrics. It is typical to evaluate the performance of a dispatcher using various metrics such as empty miles. The problem is that the required number of empty miles depends on the region, making it impossible to directly compare the performance of different dispatchers. A computer model adapts to regional differences, optimizing based on the opportunities available in a region. [A side benefit is that the model can produce performance metrics for each region, making it possible to compare the performance of each driver manager to the performance of the model in the same region.]
The challenges of automation
While recognizing the appeal of automation, it is important to understand the challenges that have to be overcome. These include:
- Missing information - A computerized dispatch system can only respond to information in the computer. Missing information in truckload trucking arises along multiple dimensions:
- Drivers - Easily the most complex resource a carrier manages is the driver. ELD systems for tracking driver hours have dramatically improved our knowledge of the status of a driver and their tractor, but drivers are people, with individual preferences (I don’t like driving into the northeast in winter) and transient requirements (I need to get to Florida so I can sell my boat). This information may be communicated by phone calls, emails and texts which are invisible to an automated dispatch system.
- Shippers - There may be special requirements for a load: handling requirements, meeting at a warehouse, picking up a special trailer, refrigeration (whether for food or drugs) are all issues that may be communicated via phone call or email, or in the notes field of a dispatch system.
- Loading and unloading - Each warehouse has its own characteristics in terms of loading and unloading delays.
- Information representation - A shipper might say “can you pick up the load in the morning, we get busy in the afternoon” is not the same as “you have to pick up the load between 8am and noon” which is how the previous statement might get translated to a computer.
- Sensitivity to errors in data - People are better at saying “that can’t be right.” Properly designed computer systems will use rules to flag many data errors, but this remains an area where people excel.
Fortunately, the flow of data in truckload trucking has been steadily improving over the years. Shippers have been steadily evolving toward computerized load tending, simply because the alternative is so clumsy (especially for the larger shippers). TMS systems have improved dramatically over the years, and continue to improve. Pattern recognition technologies have learned how to process emails and paper documents.
Moving forward
While the Ubers of the world can jump to pure automation, existing carriers have to follow the path of the legacy automotive manufacturers who have to evolve an existing process. This will require a hybrid system, but our belief is that an implementable process works as follows:
- Driver preferences are easily the hardest to quantify and change quickly over time. This is easily handled using ranked-order dispatch, where a driver is given a ranked list of loads and allowed to choose. The ranked list is constructed by optimizing across the quantifiable metrics for all drivers.
- We believe that the ranked-order dispatch should satisfy a very high percentage of driver assignments, but exceptions will arise. We need a process for flagging problems. Some of these may be recognizable using pre-programmed rules (e.g. identifying stranded drivers, drivers not meeting a TAH commitment, …, or loads that are not being picked up on time). However, phone calls, texts and emails may still be used to highlight problems.
- Dispatchers should be responding to problems (also known as “edge cases”) rather than routine decisions. This means that the dispatch automation process has to have a carefully programmed set of flags to indicate potential problems that can be fixed before they become real problems.
This vision is of a hybrid system where driver managers continue to play the critical role of solving complex problems while allowing the automated dispatch system to handle routine decisions.