UPS 20-year drive to optimization
Express carrier’s “turn data into insight” initiative is topped by ORION rollout.
Anyone who has received a parcel delivery from UPS knows its drivers carry small handheld computers that the company calls “delivery information acquisition devices,” or DIADs.
They are just the most visible evidence of a 20-year project at the company to “turn data into insight,” culminating this year in the rollout of the company’s “On-Road Integrated Optimization and Navigation,” or ORION project, on 10,000 routes.
In a recent conversation with American Shipper, Jack Levis, director of process management at UPS, described how the company’s use of data has evolved through three stages: from descriptive analytics to predictive analytics to optimization.
The company expects to deploy ORION on nearly all of its 55,000 routes in the United States by 2017.
“Twenty years ago we started gathering data with the DIAD and used it for internal purposes, but then we realized that data could give us insight into our business,” Levis said. “We started putting up data warehouses, and building what the industry calls descriptive analytics tools. Those tools of descriptive analytics tell you where you are today.
“Then 10 years ago we moved from descriptive analytics, which says ‘Where am I today?’ to looking forward to say ‘Where am I going to be tomorrow?’” he added. Levis referred to this as predictive analytics.
(While Levis spent most of his time talking about the company’s parcel delivery services, he noted “we have similar things in other areas.”)
“When Mike Eskew was our CEO (2002-2007), he said we need to treat every customer like they’re the only customer, every transaction like it’s the only one. So we built a model that said we know where every package is at every moment, we know where it’s going, we know where it needs to go and we know why it needs to go there,” he explained.
“The thought was, if we built a model that knew where things were going, then we could change the future by just flipping a bit,” or changing computer programs that modeled the movement of freight through the UPS systems, Levis said.
“We built tools to internally forecast when packages would be at a destination, forecast how many deliveries would show up a few days from now, keep the forecast updated.
“We built planning tools that analyzed the forecasts. So the tools would point out to our operators which routes tomorrow or the day after tomorrow or the day after that were going to be too heavy, which ones are going to be too light, which ones are going to be inefficient in their work,” he said.
And the shipment planner could make adjustments to routes before packages even show up. “We could move one package from one route to another if it made sense. We then upgraded that handheld computer for the driver, that was built as an acquisition device and we said ‘why don’t we give the driver the information to make better decisions?’
“Rather than a DIAD acquiring data, our model gave the DIAD information so the driver knows all the deliveries they need to make, they have a general order of delivery. So the model we put in 10 years ago said ‘here’s everything you’re going to deliver and here’s a general way to run through the area,’” he said.
But Levis noted final decisions were left up to the driver, so that if he or she wanted to deliver one package before another — or an air delivery package must be delivered first, the driver might modify the route.
(This, by the way, is one of the things that Levis said makes UPS unique. UPS drivers move a combination of commercial, residential, air, and international packages, while other competitors “might put four drivers on the road, all going to the same place, because they have separate networks. We have one network.”)
The driver may have on his or her UPS van (or “package car” as the company calls them) packages that are due for delivery at 10:30 a.m., noon, 2 p.m., 3 p.m., and have pickups at specific times. “The complexity moves to the driver to make the decisions on how to best service all of those customers, but we gave them information to do so, and then we keep our data model updated in real time. We make adjustments all the way through until that last package is loaded to make sure our routes are efficient for that day,” Levis said.
“When drivers are on the road, we communicate back to the ‘mothership.’ So we know in real time where every driver is, what work they have completed, what work they have left to complete. If a customer wants to meet a driver, we give it to the driver that’s near the customer when the customer wants to be served. We deployed all those things 10 years ago. That’s the predictive realm.
“When you do descriptive analytics of just knowing where you are, you’re going to catch low-hanging fruit,” Levis said. For example, if you have two drivers going to the exact same address, you can eliminate it.
But he said “there are limits to the gains you’re going to get from just knowing where you are. You’re always looking at yesterday and, unless tomorrow happens like yesterday, you’re only going to fix the easy things,” he said.
When the company moved to predictive analytics, the system might point out that on an upcoming day a particular route will have too much work for a single driver to accomplish. So an individual at UPS, guided by the system, would come up with an alternate plan.
“As we moved to predictive analytics, on top of everything we had already done, we ended up saving 85 million miles driven a year through our data models,” Levis said. “That’s 8.5 million gallons of fuel we’re not buying. That’s 85,000 metric tons of CO2 not going into the atmosphere. So the predictive modeling, just like the literature said, brought us more gains than the descriptive modeling. On top of an efficient operation we got better at predicting tomorrow, planning tomorrow and simplifying jobs.”
UPS estimates if it can reduce the amount of driving each one of its drivers does by one mile per day, it can gain $50 million annually.
The company also tapped the expertise of its drivers and put that information into its system, so if a driver was on vacation or being aided by a seasonal employee during the Christmas rush, they would know to leave a box on your back porch or with your neighbor if you weren’t home.
Levis said the company also “opened up our internal supply chain and logistics systems to our customers” by creating products such as “My Choice” that allows consignees to delay delivery, or alter where a package is delivered. UPS used to go to each business customer location every day whether it had packages or not; now, utilizing technology via its Smart Pickup service, customers have the flexibility to request pickups on demand and UPS only has to stop when they actually have packages ready to ship.
Chuck Holland, vice president of industrial engineering at UPS, noted services, such as My Choice, have become increasingly important as the e-commerce and home deliveries have increased and more people are moving into cities.
He also said the share of deliveries to homes is growing faster than commercial deliveries, and drivers are making deliveries to more stops. “It would have been very difficult for us to respond if we hadn’t laid the groundwork 20 years ago,” Holland said.
Optimization. This year UPS began rolling out ORION, which Levis said moves from giving general advice to a driver about the route he or she should take to very specific manifest giving instructions on what order to visit addresses.
ORION takes into account the customer’s needs, UPS’s needs, union rules and gives the driver an optimized route.
“That requires some huge advanced mathematics to truly optimize a driver’s day,” he said. It also required building into the company’s database a lot of additional information — the amount of space in the driver’s truck, how well they can walk around, and how late businesses can receive commercial deliveries. And better maps.
David Abney, chief operating officer of UPS, told a conference of the American Highway Users Alliance that when he started out as a 21-year-old package driver out of college in Pascagoula, Miss., he had to make his way around a town he’d never been to before using hand-drawn maps provided by another driver.
Things have improved much since then, but Levis noted there’s still “no off-the-shelf map or GPS system accurate enough to solve a driver’s day.”
For example, ask your GPS unit or Google maps how to get to Walmart and it may put you on the street in front of the store, when the driver needs to get around back where the loading dock is. A customer’s driveway may be a half mile long, or two stores may be on different streets, but be connected by an alley that the driver knows about.
To truly optimize a route, the computer system needs to know about all those nuances, and be upgraded with the intelligence of the driver and their knowledge of the area, Levis said.
“We have 250 million address points. So if you’ve ever used a GPS system that says you’ve arrived somewhere but you look and it’s really down the block a little. Ours doesn’t do that. Ours knows exactly where each delivery is by address,” he said.
ORION is a big project, with more than 500 people working on it — from Ph.D. mathematicians and programmers to specialists who review satellite images and update maps, or drive around in rental cars and interview drivers to improve maps and train them on how to use the system.
“We want the driver’s intelligence and we say, ‘does this route work?’ And the driver may say, ‘No, you can’t do that because you’re getting to these businesses… or I’ve got traffic in this area that you’re not taking into account.’”
ORION then uses advanced math to find the best route for a driver to take. It’s a surprisingly complex problem that can’t be solved merely by checking every possible permutation.
Even with just 25 stops, Levis said there are 15.5 trillion trillion different ways to run the route. And each day UPS drivers may make 125 to 175 stops. (The variation in UPS routes is enormous, of course. Some drivers can be kept busy all day serving one or two buildings in Manhattan, while others may drive hundreds of miles in rural Wyoming.) UPS said the number of route combinations that a UPS driver can make in a day is far greater than the number of nanoseconds the earth has existed.
UPS said its proprietary ORION algorithm equates to nearly 1,000 pages of code and evaluates more than 200,000 alternative ways to run a single route.
The PBS science show NOVA thought the math behind ORION was interesting enough to feature it in an episode this fall.
ORION can solve the problem on how best to serve a 140 stop route in about six to eight seconds, Levis said. When a driver uses ORION, “you are definitely altering what they would do. ORION will find nonintutive ways to run a route.
“As a driver you hate to drive by a delivery stop,” he added. “I’m driving by, I see three deliveries I have to make, and I’m on the street, why don’t I deliver it right now? ORION may say nope, that’s not a good decision. Because what it’s going to take into account, is how does an overall day look.”
Levis said the program will make thousands and thousands of adjustments trying different things until it finds the cheapest cost, and “often the cheapest cost is to bypass something early to get savings that will turn up later on in the day.
“It’s finding all these little nonintuitive things to save a quarter mile here, a half mile there, etc. and ends up with some significant savings,” he said.
“We’re not publishing what the savings are,” Levis said. “But I can tell you that we monitor every single week where we stand and we are absolutely on track to our initial gains. In fact, we are over our initial estimate, which is why we’re so excited about ORION.” He did say, however, the company is saving 1.5 million gallons of fuel with the 10,000 drivers using the program so far.
This isn’t to say ORION can’t be further improved. The system gives the driver a static manifest, but does not adjust for changing conditions for time of year (driving past a shopping mall in July versus a week before Christmas) or traffic — a major accident on the highway.
At some point in the future, Levis predicted ORION will be able to produce dynamic manifests, which will take into account if a driver gets to a stop late, or faces traffic congestion, which will update in real time.
“That is the No. 1 request from our driver is to update this in real time every time they make a decision. It doesn’t do that yet, but it will. And all that will mean by the way is we will get even more gains. It’s just a matter of building the infrastructure to do so.”
Still Levis said UPS is on the leading edge of using operations research to run its business. He said the Gartner Group found about 70 percent of companies use descriptive analytics and 16 percent use predictive analytics, while only 3 percent use proscriptive analytics or optimization.
(Eric Johnson and Eric Kulisch contributed to this article.)
Correction: Jack Levis is director of process management at UPS. His surname was initially misspelled in the story.