Brown Down: UPS Drivers Vs. The UPS Algorithm

UPS’s new algorithm can plot routes more efficiently than drivers. Just try convincing the drivers of that.

“A lot of times, I feel like an explorer,” says Jack Levis, UPS’s director of process management. “Often I’m telling the company: Just because we’ve done it this way for the past 50 years doesn’t mean it’s right.”

Levis, who manages a team of mathematicians who build the algorithms that help UPS shave millions of miles off delivery routes, is paid to tell the company things it may not want to hear. One of his major projects in the last decade has been rolling out a system called ORION (On-Road Integrated Optimization and Navigation), a kind of algorithmic overmind that knows better than any human how drivers ought to plan their routes.

ORION was first conceived in 2000, but wasn’t tested till 2008. Over the past four years, the system has rolled out to some 50 UPS buildings; it will take another half-decade or so to roll out the system throughout UPS. “It’s one driver at a time, one building at a time,” says Levis.

Developing a system of this magnitude–and making a 105-year-old company comfortable with it–was no easy feat. Fast Company caught up with Levis to glean a few lessons.

Of Math And Men

“Advanced analytics should be one of the top priorities for CIOs,” says Levis, who can talk of math in near-koans: “Beyond knowledge is wisdom, and beyond that is clairvoyance.” Math simply can solve problems that humans can’t. For instance, by running advanced analytics on reams of collected data from trucks, Levis’s team is now able to predict when a given part is about to fail: “preventative maintenance,” he calls it.

ORION is about 80 pages’ worth of math formulas–“like something Einstein would have on his board,” says Levis. So far, it has saved UPS something like 35 million miles a year, and Levis projects that it will save millions more.

Consider an average driver’s route. There are more ways to deliver such a route than there are nanoseconds that the Earth has existed. Take one particular problem: You’re a UPS driver, and you’re delivering a package. There’s also a package due next door–but not till later this afternoon. Do you deliver it quickly now? Intuition says yes. But then do you also deliver the package two doors down? How about the one across the street? And if you follow the rule of thumb of hitting all nearby houses in this neighborhood, should you necessarily follow that same rule of thumb in the next neighborhood? And the next?

“The combinations are astronomical,” says Levis. “What we do as people is oversimplify.” We decide only to early-deliver the priority packages. Or we drop off all the packages on this block but skip the others. “Rules of thumb don’t truly optimize,” says Levis. Math does.

But Trust Boots On The Ground, Too

Still, no one who’s been driving a route for a decade or more wants to suddenly be bossed around by some computer. Levis knew ORION was good for UPS. But UPS’s drivers needed some convincing.

When ORION first began to roll out, Levis admits he presented the system in a less-than-ideal way. “We’d go in in the morning and say, here’s your planned number of miles.” A driver who usually had a 155-mile route was suddenly being told a computer was saying he could do it in 140. It probably felt something like a put-down.

“So we changed it,” says Levis. His team put up a sheet that said, “Beat the computer.” It was a matter of framing: ORION was going to make a prediction about how fast you could do your route, and now your job was to do it one better (all while taking into account ORION’s suggestions).

Levis recalls one driver who normally did his route in 150 miles. ORION predicted he could do it in 140. One day, the driver came back from his route and said, “I told you, your system’s no good.” ORION’s prediction was wrong, he said–he had managed to do the route in 135, not 140. “To this day, he doesn’t really recognize that ORION is what caused this,” says Levis. “He just views it as another input to use together with his intuition.”

Tap Academia

The problem of route optimization that UPS works on is actually a well-studied math problem, the Traveling Salesman Problem. It makes sense, then, for UPS to partner with academia. Levis was elected to the board of directors of an organization called INFORMS (the Institute for Operations Research and Management Science). He’s the only board member without a PhD.

“It’s a funny marriage, because I’m always the dumbest person in the room,” he says. “But then again, I’m the person in the room who has actually implemented this advanced mathematics. It’s a wonderful way for UPS to get what the latest research is, but UPS gives back to INFORMS as much as we get.”

Common Sense Trumps

At Levis’s office, they’re continually asking if they’re smarter than a fifth grader. The reason is that a project manager on ORION recently visited his daughter’s school on career day. He explained all about traveling salesman problems, time windows, all the intricate math that UPS works on. He gave an example of someone who has to go to the barbershop, the grocery store, and a number of other places, but in an uncertain order; ORION’s solution had him going to the grocery store first.

A student raised his hand and said ORION didn’t work. “My mother would never do this,” he said. After all, you can’t have ice cream sitting in the trunk all afternoon, while you’re off getting a haircut.

“That’s the challenge we’re up against,” says Levis. “Moving from mathematics that happens to work mathematically, to mathematics that works that people actually do.”