The trucker and the professor

By: David Diamond

Date: Saturday, 02. September 2006

David Diamond is coauthor, with Linus Torvalds, of Just for Fun: The Story of an Accidental Revolutionary. This article originally appeared in Wired.com.

For a brief, shining moment of national obsession in the Ford and Carter administrations, America's truckers ascended to heroic cultural status - they were the cowboys or dotcommers of their day. Americans glorified the risk-taking and self-reliance and independence of these peripatetic heroes by adopting their colorful CB lingo (referring to motorcycles as "crotch rockets" and rest areas as "pickle parks") and praising them in song. Who could forget those lines from the 1976 hit "Convoy": "Yeah, them hogs is startin' to close up my sinuses. Mercy sakes, you better back off another 10."

The romance is gone. Today's trucker is likely to have a lot in common with Orlando Mitchell, the man to my left, a genial, contemplative guy in a plaid shirt who is hauling 42,000 pounds of batter from Cleveland, Ohio, to Gloucester, Massachusetts, as an employee of Marten Transport, a $260 million national carrier. From the second Mitchell starts his engine in the morning, a dispatcher tracks his exact location and sends him regular instructions, keeping him on a tight electronic leash that makes liberty and leisure a thing of the past.

In the freewheeling days of the CB craze, a trucker like Mitchell would deliver his load in Gloucester, amble to the nearest truck stop, reach for a pay phone to contact a dispatcher or broker, and ask for a new load. In some cases, he might search for shipment requests on a truck stop corkboard. The process, called "dialing for diesels,"" typically allowed the driver to lounge around and kill time until a load could be rounded up - it might have involved an overnight stay - and he would then "dead-head" (drive an empty truck) to his next pickup.

That sort of rambling life is unimaginable today. Like the routing of IP packets on a communications network, the routing of trucks on the nation's 42,794-mile interstate highway network is handled systematically by satellites and software. Mitchell rarely uses the CB inside his 17-ton Peterbilt Model 379; instead, his main connection with the outside world is what he refers to as his "Qualcomm," a laptop-sized device that serves as a satellite link to Marten's corporate headquarters in Mondovi, Wisconsin.

When a red light appears on the dashboard in front of us, Mitchell pulls over the first chance he gets and reaches back for the gadget - Qualcomm's Omni Tracks - to check his new LCD message. It tells him that after he drops the batter at 10 am, in a warehouse grafted onto the offices of seafood purveyor Gorton's of Gloucester, he is to adjust the refrigeration unit to minus 10 degrees Fahrenheit and wait for a load of fish sticks at 2:30 pm, which he will haul 1,086 miles to a warehouse in Lithia Springs, Georgia. It tells him where in Raphine, Virginia, to fuel up during that trip, how much fuel to get (63.8 gallons of diesel), and what it will cost. As he rumbles northward on the New Jersey Turnpike toward Massachusetts, he could at any moment get a message detailing where he should head after Lithia Springs.

Independence? The closest Mitchell ever gets to that is passing exit 12 on Interstate 70 - the turnoff for downtown Independence, Missouri.

Truckers like Mitchell are the pawns in the burgeoning logistics industry, a vast economic heartland that includes the shipping and warehousing of goods. A term once generally limited to the military, logistics is now viewed as a brass ring for business. When consultants talk about "squeezing efficiencies" from the supply chain that stretches from raw material to consumer, much of the opportunity for the squeezing lurks in logistics, a field so broad it accounts for an estimated $1.06 trillion of the US gross domestic product - more than defense. Despite the slowing national economy, total revenue of third-party logistics companies like Penske Logistics and FedEx Logistics rose 24 percent in 2000 to $56.4 billion, according to Cass Information Systems.

But within logistics, nowhere is the potential for savings as great as it is among the 2.5 million Class 8 trucks that, at any one time, are driving America's highways. Transportation is the largest individual component of logistics costs, and trucking is the primary means of conveyance. Of each dollar spent on moving freight in the US, 80 cents is devoted to trucks. When companies outsource transportation procurement - as they are doing today in record numbers - they rely on software companies like Logistics.com, i2 Transportation, and Manugistics to squeeze efficiencies out of the hard-to-forecast areas of route planning and carrier bidding. In the game that is trucking logistics, scheduling, planning, and route mapping are increasingly determined without human intervention. To get a firsthand look at modern trucking, I've arranged to hitch a ride with Mitchell and his 50-pound sacks of batter. To get a firsthand look at modern logistics, I visit Yosef (Yossi) Sheffi, a balding 53-year-old Israeli, who has spent the past three decades bringing high-level mathematics to America's highways. Currently director of the MIT Center for Transportation and Logistics Studies, Sheffi has drawn from general mathematical insights into optimization to create decision-making systems that change the way trucking gets done. He has teased out algorithms and put them into practice both in companies he's owned and in those he has advised. He has founded five logistics-related companies since the late 1980s - so far he has sold four of them for a profit - and has consulted for some of the nation's largest shippers and carriers. David Closs, professor of logistics at Michigan State University, calls Sheffi "one of the few gurus" in the field, and Ted Farris, a professor of logistics at the University of North Texas, says simply, "In terms of linking the thinking to practice, Sheffi is the leader."

Sheffi's office is oversize for an academic's, containing no fewer than eight flip charts and a whiteboard. Perched on a bookshelf that runs for maybe 20 feet is his collection of model trucks bearing the names of various carriers. "Which truck goes where and how it is routed are very difficult questions," he explains, "because you have to simultaneously consider millions of possible movements. The number of decisions grows exponentially as you consider not only how the trucks go or how the freight is routed but all the possible combinations. And every time you change one variable, everything else has to change to achieve optimization."

The fundamental problem facing transport companies boils down to finding the best route between points A and B. Carriers come in two flavors: less-than-truckload and truckload. In the less-than-truckload sector, one rig carries goods for a variety of shippers just like FedEx carries packages. Less-than-truckload companies have a set of terminals and consolidate their shipments as they go along. As in the business of commercial air travel, locations and schedules are essentially fixed. The challenge for companies in this sector is to keep trucks full as consistently as possible. For example, when transporting goods from Philadelphia to Chicago, a carrier will determine which truck to use, at what time, and following which route - through Pittsburgh, Cleveland, or Detroit. (Or should the truck unload the goods it's hauling to be picked up by a different Chicago-bound conveyance and instead grab a new load bound for, say, St. Louis?) A single less-than-truckload carrier may have hundreds of terminals. "If you have four terminals you may have 10 decision variables," explains Sheffi. "If you have 400 terminals you might get 400 million." And like that butterfly in Beijing flapping its wings, each decision affects the entire network.

The so-called truckload world - the one in which Mitchell and his batter live - has its own set of daunting decisions to make. If a truckload company with a fleet of 10,000 vehicles is called upon to haul Pine-Sol from Boston to Detroit, should it take the job? And if so, what should it charge? Which driver should it use? If the rig goes to Detroit, what happens when it gets there in two days - will it have to wait three more days for another load? If the truck is already in Boston, is there a better arrangement that would mean hauling goods to Atlanta, instead of Detroit? Or should it just hang in Boston for a day to see if a better job comes along?

The cost of shipping one particular load of goods is not the only consideration. A carrier must also work around a truck's maintenance schedule, for example, or try to position trucks favorably for future trips, or deal with a driver who wants to end up at his kid's wedding in Santa Fe on Friday.
The best route between A and B is part of a complex logistical operation increasingly run without human intervention. This last point has become particularly vexing in the scheme of logistics. The era of deregulation has brought with it not only increased competition but increased employee dissatisfaction. Trucking is plagued by a 100 percent annual turnover rate - on average, an entire workforce quits each year. In response, carriers have resorted to a host of enticements. They have increased their reliance on "drop 'n' swap" routes, for instance, in order to accommodate drivers who don't want to venture too far from their homes. A driver leaving San Francisco, say, may be hauling freight that's bound for Seattle, and another driver leaving Seattle may be hauling freight bound for San Francisco. Instead of going the distance, the drivers meet in Portland, unhitch, swap trailers, and head back to their originating cities.

And there are other ways of improving driver loyalty. When assigning variables for optimizing routes, carriers will factor in such previously unheard-of considerations as whether a route takes a driver in the vicinity of his girlfriend's house or past the truck stop that has his favorite videogames. The challenge for Yossi Sheffi is to input every concern on a carrier's wish list - and output the most profitable solution.

It would be a stretch to use the words "truck stop" to describe the location in Newark's edgy Downneck neighborhood where I arrange to meet Orlando Mitchell's 18-wheeler. The place consists of a fuel pump and a spot to park a few trucks set amid barbed-wire fences and car junkyards and the fumes of a nearby refinery. An hour after our agreed-upon time, Mitchell's truck glides by, a pure gem in this trash heap. It is stately in appearance - from its blue, square-nosed hood to the door on its white trailer, the freshly washed vehicle stretches regally for 53 feet and stands 13 feet, 6 inches tall.

Together we traverse the Jersey Meadowlands, the George Washington Bridge, the Cross-Bronx Expressway. He talks in a high voice, explaining that, if he were alone, he would be listening to jazz or blues or maybe inspirational leadership tapes from a college basketball coach or a lecture on postmodernism. And shortly after New York gives way to Connecticut, we pull off the road for dinner, parking behind other trucks whose engines are still groaning. We are in McDonald's at a state-of-the-art truck stop. Mitchell ingests two cheeseburgers and a banana milk shake and explains the rules of the road. Marten Transport pays him 35 cents a mile. Truckers are entitled to one day off each week. Or they can opt to continue driving for three or four weeks in order to take a single, three- or four-day break at home. Mitchell himself hasn't been home in three months.

But Mitchell doesn't actually have a home. A Milwaukee native who has held a succession of factory and janitorial jobs, he moved to Sacramento, California, in 1993 on a whim. He learned to drive a truck and worked for a host of companies before settling with Marten. He drove long routes, returning every couple of months to his apartment in Sacramento. Eventually, to avoid the risk of break-ins while he was gone, he gave up his lease and put his belongings in storage. Now, the rare times he goes "home" to Sacramento, he deposits his trailer at a Marten facility in nearby Stockton, and then drives without the trailer - that's called "bobtailing" - to a truck stop near Sacramento, where he lives out of his cavelike cab.

Sheffi is at the whiteboard, professor-style. I have asked him to explain the algorithms that are involved in routing trucks like Mitchell's. Each day, more than 60,000 trucks rely on the optimization software provided by Burlington, Massachusetts-based Logistics.com, where Sheffi stepped down as CEO this year, selling his majority stake to Internet Capital Group in September. In addition, hundreds of thousands of trucks benefit from Logistics.com services like OptiBid, an online auction for shippers and carriers that is based on Sheffi's core technology.

"Structured optimization is basically an elimination process," Sheffi begins. "It's a combination of stepping though a solution space and discarding as I go the large part that I don't have to deal with. Throughout the process, I'm focusing on where possible good solutions lie, rather than actually checking each one." As an algorithm continually tests a small number of solutions, it uses the properties of those solutions, and some characteristics of the problem, to discard whole groups of other possible solutions without explicitly evaluating them. The system then repeats, testing some solutions and discarding many others until, from among all the possibilities, the best have been adopted. Says Sheffi, "The fastest computers would take years to solve just one of these problems if they had to check every solution."

But what is the space that this algorithm metaphorically creates? This notional area - also known as the feasible region - is best understood as a polysided, multidimensional box. Optimization problems involving two variables (such as two trucks picking up some portion of a shipment from separate warehouses) can be represented geometrically on a standard x-y axis, and the feasible region - the area in which the optimal solution can be found - is easily identified as a plane bounded by lines representing such constraints as driver availability and the speed at which the trucks can travel. In problems with more than two variables, the feasible region is represented not as a mere plane but as a multidimensional contour. Sam Savage, a consulting professor in Stanford's Department of Management Science and Engineering, uses the metaphor of a ball bearing being dropped into a tilted cardboard box to describe this mathematical process. The ball, going downhill, represents the algorithm, solving systems of equations, adding and removing solutions as it rolls along the planes of the enclosure. And the corner - where the ball always stops - is a point that represents the simultaneous solutions of the remaining equations. It could be the point, for example, indicating where the fewest number of trucks would be needed to make the required deliveries.

The underlying mathematics for creating much of logistics software is called the simplex method of optimization, a theory that was invented by former Stanford professor George Dantzig in 1947, when, as a Pentagon statistician, he was experimenting with mathematics to improve flight scheduling for the US Air Force. The simplex method has become the basis for generations of advances used for solving linear programming problems - that is, any problem where there is an objective to be achieved and constraints to deal with. Simplex is widely applied to everything from optimizing financial portfolios and cash flow management to determining the proper mix of octane in gasoline. Dantzig's big innovation was determining that optimization solutions always lie on the boundary of the geometrical feasible region, so there's no point in even exploring the interior.

Here's Dantzig's own view of the beauty of simplex: "You have a bunch of people and a bunch of jobs and you want to assign 100 people to 100 jobs. There's a value attached to assigning the Kth guy to the Lth job, and you want to maximize that value, the sum of the values attached to assigning those 100 people to 100 jobs," he explains. "The number of possible combinations turns out to be 100 factorial [100 x 99 x 98 x 97...], which happens to be more than the number of atoms in the universe. If you tried to go through every combination and put a value to the combination of the sum, you could never do it in a million lifetimes. Yet the simplex method is so powerful it solves this problem in a blink of an eye."

When combined with advances in CPU speed and satellite communications, simplex set the stage for Sheffi and others to build algorithms informed by their growing expertise in commercial transport. Rick Murphy, for example, now president of Integrated Decision Support, which sells software to carriers, was the first to use network models for truckload (as opposed to less-than-truckload) transport. Later, Warren Powell, who teaches operations research and financial engineering at Princeton and founded Transport Dynamics, developed algorithms for assigning one driver to multiple loads.

"The trucks running throughout America," says Sheffi, "are doing so based on the work of lots of academics who contributed to the theory and practice of logistics."

And their use of the simplex method isn't limited to solving for a single best scenario. While profitability is typically the first criterion considered in truck routing, when others are added - say, the need to get the truck in the Santa Fe vicinity for the trucker's daughter's wedding - relative profitability can change. Sheffi explains the process of setting up a problem with a variety of goals: "If I have many factors, I can create a balancing function between the factors," he says. "I may give 40 percent of the weight to profitability and 60 percent to driver consideration. So I look at every solution, and in some sense combine them into one number."

Contrast this high-level mathematics with how your typical trucking company tackled the scheduling problem 20 years ago. Relying not on simplex but on gut feel, dispatchers eyeballed a job, looked at a few options for each truck, and made a decision. "They didn't know if it was the right decision or the wrong decision, they just made a decision," says Sheffi. "And they made it in isolation, without considering its impact on the network." A soft rain is falling on I-95 in dark eastern Connecticut. Mitchell sips the remains of the morning's Mountain Dew. He talks about how his savings have grown since he started living in the truck, how he's beginning to pay down his credit card debt, and how he wants to buy a high-definition television or a Dell PC, not that he has a house in which to put either. He explains that he owns neither a cell phone nor a computer. How do outsiders reach him? "If my mother wants me to call her, she can contact the company, and they can send a message to me on my Qualcomm," he says.

We talk prostitutes. Turn on the CB radio, Mitchell warns me, and you're just as likely to hear a trucker asking about the prospects for prostitutes in a particular city as you are to hear about road conditions. "It's always, 'Where are the girls?'" he reports. In the lingo of truckers young and old, prostitutes are referred to as Lot Lizards, and they typically market themselves by simply knocking on doors of parked trucks. "If you see a truck with a 'no Lot Lizards' sticker in its window - a lizard with a red line through it," Mitchell says, "that means they don't want any bothering them."

Yossi Sheffi is a serial entrepreneur trapped in the body of an academic. Since joining the MIT faculty as a professor in 1978, he has lived in two worlds: one in which he has taught thousands of students, and one in which he starts companies, builds them into flourishing businesses, then cashes out.
A company with 400 drop-offs might have 400 million decisions that affect the entire network. Sheffi's code crunches them all. As a young man and a captain in the Israeli air force, Sheffi served in both the Six-Day War and the Yom Kippur War, shuttling cargo on a Hercules C-130. His Austrian-born engineer father fought with the British in North Africa and later joined the Hagganah. His Kiev-born mathematician mother worked for Israel's Mossad as a "librarian." Yossi raises his eyebrows when he uses the word. "She never talked about her work, so maybe she was a librarian, but with her mathematical background she probably had something to do with deciphering codes," he says half-jokingly. He's animated and confident.

In the late 1970s, as a student in MIT's graduate civil engineering program, Sheffi took an interest in network systems - representing them on computers, modeling and optimizing flows. That's when he got a taste of the beauty of transportation logistics. A company called IU International, which owned various truck lines, gave MIT several hundred thousand dollars to study how to improve the flow of trucks over its network - to lower costs and better serve customers. The company operated an array of fixed routes that it changed on an annual basis. Using the simplex method, Sheffi and his colleagues spent two years developing the algorithms for making IU's network far more dynamic, setting up a framework that could be changed on a weekly basis in response to the demands of customers. "Actually," Sheffi says, "our system meant they could change their networks on a daily or an hourly basis; they were constrained not by the calculations involved or the computing capabilities but by the ability of workers on the docks to absorb change."

Then in 1988, with Warren Powell, of Princeton's Castle Laboratory, and David Cape, who was Powell's student, Sheffi cofounded Princeton Transportation Consulting Group (PTCG) to sell decision support systems to the transportation industry. In the same year, Sheffi cofounded LogiCorp, one of the first pure logistics companies in the US - with no trucks and no products - which he sold six years later to Ryder Systems, where it now is known as Ryder Integrated Logistics and generates $1.5 billion in annual revenue. In 1997, Sheffi cofounded e-Chemicals, an electronic commerce company for the chemicals industry, which he sold three years later to AspenTech. And in 1999, he cofounded Syncra, which develops online collaborative planning, forecasting, and replenishment software.

Perhaps Sheffi's biggest coup so far occurred in the early weeks of 2000, when, based on some sketches on a whiteboard, he was able to obtain $30 million from Internet Capital Group to buy back PTCG, which he had sold in 1996 to American Airlines Sabre Group. This operation, renamed Logistics.com, is now a 121-employee company and, despite recent layoffs, remains one of a handful of investments made by ICG - an incubator famous for squandering money in the B2B sector - to survive and show promise. Today, Logistics.com counts Wal-Mart and Procter & Gamble among the customers for its network bidding services, which are provided over the Internet in ASP mode. Sheffi estimates the company has saved shippers more than $400 million in logistics costs. An illustration: Because carriers can bid in advance on shipments that will fill up their unused capacity (the dead head from Dallas), they can afford to charge less; one shipper, The Limited, says it saved $1.24 million in a single year using Logistics.com's automated service for negotiating annual contracts.

And it's this company, offering software based on the simplex method, that organizes the lives of truckers like Mitchell. Companies such as Marten Transport access the software on the Web and type in the factors - volume, destination, and so on - they're considering in scheduling a route. The input is crunched and sent back as a series of instructions. These are the directions that ultimately get spit out on Mitchell's Qualcomm, as he rumbles along America's interstates.

It is 7 am in Providence, Rhode Island, and Mitchell's 18-wheeler grinds through the rush hour on I-95. Soon it is grinding through the rush hour on Route 128 in Boston's western suburbs. An on-again, off-again rain keeps the roads slick and slows traffic to an annoying crawl. We pass the big highway signs for Walden Pond State Reservation and Minute Man National Park. Mitchell spots a Honda Civic cutting off another vehicle. "Look at that four-wheeler," he yelps, using the trucker's term for a car. "That guy shouldn't be driving." Mitchell is a careful driver, but the truck helps him: It can't exceed 65 mph. He repeatedly pulls out a map to calculate how far we are from the Route 128/I-95 split. The rain intensifies.

Mitchell tells me that trucking gets in your blood, that it's hard to take a sedentary job once you've become hooked on the highway. He has his favorite spots (I-5 just north of the California-Oregon border) and his favorite time-killers (spotting eagles in Washington State). But mostly, he is motivated by the same thing that motivated those other cultural heroes, the dotcommers: money.

Trouble is, a typical trucker's current salary of $41,000 doesn't go nearly as far today as the average $35,000 did during the CB madness of the Ford and Carter years. To some degree, truckers were romanticized then because many were making good money for folks without a college education. Many were able to own a house and a car and a cabin on the lake, and they had time to enjoy it.

Not Mitchell, the modern trucker. "I miss my TV," he says.

Finally he veers toward the highway that will be our last leg. "This is Gloucester," he says, as we cross the town line, and soon he is negotiating two rotaries and an old harbor road that leads us to the Americold Logistics storage facility, next to the Gorton's of Gloucester headquarters on the waterfront, right where his Qualcomm sent him. Birds circle the harbor and rain pounds white clapboard houses as Mitchell backs the trailer into a loading dock. It is 9:58 am. He's two minutes early.