Can a 1,000-Page Algorithm Named ORION Predict the Future?

September 20, 2015 • Blog Post • By Atlantic Re:think


  • UPS spent years developing ORION (On-Road Integrated Optimization and Navigation), the advanced algorithm that sends delivery drivers down the fastest routes possible
  • By the time ORION is done at the end of 2016, it will save 9 million gallons of fuel a year and have saved UPS up to $400 million

How UPS sees big data evolving from insights to clairvoyance

When Jack Levis tries to explain the enormous implications of the ORION system now being deployed at UPS, he starts small. First, he asks, "How many errands do you have to do this weekend?"

Let's say it's six: pick up eggs at the grocery store, drop off the leaf blower at the repair shop, take the kids to a soccer game, go to the bank's ATM, grab a case of beer at the liquor store and pick up a balloon bouquet from the party shop.

The typical human thought process, without us even realizing it, filters out and discards countless options. Is it possible that dealing first with the leaf blower could get you to the kids' soccer game early? Or that collecting the beer, on the left side of the street, before the balloons, on the right side of the street, might save you precious minutes? Our brains aren't wired for such minutiae, so options get filtered out automatically, like background noise.

In reality, Levis says there are 720 ways you can do six errands. Make it 10 errands, and you have 3 million options. Add two more errands, and there are 500 million possibilities.

If you're a UPS driver, you likely have 120 things to do in a given day, with each of them requiring you to figure out how to deliver so many packages to so many places along so many possible routes. Do the math on that for all UPS drivers, covering some 55,000 routes in the United States alone, and you get a wholly incomprehensible number: 6,689,502,913,449,135... plus 183 zeroes.

"Can you really think through every possible way to optimize those options? You can't", says Levis, who is senior director of process management at UPS. "ORION's thousand pages of code looks for ways to do the job that are counterintuitive. It will go down paths that you wouldnt think of on your own. It will find those needles in the haystack."

UPS has been developing ORION since 2003. The acronym stands for On-Road Integrated Optimization and Navigation. Basically, ORION is an algorithm equivalent to about a thousand pages of computer code. It looks at which packages need to go where, and it figures out things humans never could - like the fact that sending drivers on routes with fewer left-hand turns can save 9 million gallons of fuel that otherwise would have burned while trucks waited for traffic lights to change. By the time ORION is fully deployed, at the end of 2016, it will save UPS some $300 million to $400 million.

Work on ORION started long before terms like "Big Data" became part of the mainstream business conversation - or before anyone inside of UPS had ever heard the words. At the time, Levis says, UPS was deploying what it calls package-flow technologies - models that help predict where packages will show up within a few days - to help the company plan. Those models alone are saving UPS about 85 million miles per year of fuel and driving time.

"ORION, when it's fully deployed, will save an additional 100 million miles a year", Levis says.

Understanding how ORION fits into the world of everyday business requires a basic understanding of data analytics. As Levis explains it, analytics takes three forms. The first is descriptive analytics, which explains what has happened. The second is predictive analytics, which explains what will happen next if you continue down the same path. The third is prescriptive analytics, which uses data to answer the question: Where should I be headed?

"Only 70 percent of companies are using descriptive analytics", Levis says. "About 16 percent are using predictive and about 3 percent are using prescriptive. ORION is in the prescriptive realm. People are aiming to get to predictive, but few understand that there's optimization that comes after. If your vision aims at predicting, your vision is shortsighted." Alexandra Deschamps-Sonsino, founder of the London-based advisory company DesignSwarm, tries to help companies understand these types of differences and more in the realm of Big Data analytics and the Internet of Things. She says many CEOs are just now starting to assign core product dollars, albeit on a smaller scale, to projects like the ones UPS began more than a decade ago.

Looking out about 10 years, Deschamps-Sonsino sees a business landscape where successful leaders will be the ones who are investing today. Digitally native companies will lead the way. "They will bring their digital thinking to, probably, quite traditional infrastructure businesses that just don't know how to think about this", she says. "Right now, those traditional businesses are educating themselves. In five years, they'll be doing a pilot. In 10 years, maybe 20 percent of their profits will be coming from those investments."

UPS, with its 12-year head start on ORION's research and development, appears to be ahead of that curve in a number of ways. Levis sees the UPS effort is akin to evolving the 108-year-old company into something that shares a surprising number of attributes with digitally native firms. "We spend about a billion dollars a year. That turns us into a data company", he says.

And of course, UPS is now thinking about what ORION can do next. The core technology, he says, is doing great things, but not in ways that are "Jetsons"-level cool. "I've got good news and bad news about ORION", he says. "The bad news is, ORION doesn't do a lot of the things that people think it does. It works so well and it's saving so much that people think it does things it doesn't". He adds: "The good news is, it doesn't do what you think. If we're going to save 100 million miles a year with technology that isn't perfect, think about what's going to happen when we make ORION smarter and smarter and smarter."

The key change coming to ORION in the next decade, Levis says, is that it will change from a static to a dynamic system. Today, ORION sifts through all the data points it has - solving 30,000 optimizations every minute - and then tells UPS drivers what packages to deliver in what order. After the driver hits the road with the ORION-created plan, though, the system is unable to adapt. It doesn't yet incorporate things like geographic mapping, dynamic weather data and real-time traffic reports. A version of ORION that incorporates GPS, for instance, is still in the prototype phase.

Those with a GPS system in their own car might be surprised to learn that last bit, but the GPS systems most of us know are far less sophisticated than the one UPS built for ORION to analyze. The logistics company created its own maps using more-detailed data points that describe some 250 million addresses. While a regular GPS might get you to a street address, the UPS data points that ORION is using know the latitude and longitude coordinates for your mailbox, your driveway, and your front door, which at some locations can be long distances apart.

"We created maps that are more accurate than any maps in existence, and we did it so ORION can work", Levis says. "If we're delivering to a Walmart, it knows we're delivering to the back of the store. It knows about driving through a parking lot. It knows where the alleys are."

Adding dynamic weather, traffic and other data to that type of existing data, and then letting ORION sift through it all, will someday let the system not just notice when a traffic jam has happened (descriptive analytics), but instead notice fluctuations in traffic patterns (predictive analytics), and suggest routes for drivers to take before the highway bottleneck occurs (prescriptive analytics). Having a dynamic ORION system will happen within a few years, Levis says, "but the true prediction and clairvoyance is probably five to 10 years out. Your data has to be precise, and ORION has to be smart enough to throw out the noise from the real issues."

Yes, Levis used the word clairvoyance, which is what he sees as the natural next step. "You talk about turning data to information to knowledge, and most companies are struggling to get to knowledge. I think we"re past knowledge with ORION. We're at wisdom", he says. "Beyond that is clairvoyance. These are tools that are so smart that they can predict problems and fix them before they happen."

UPS envisions that future not just in the delivery arm of its business, but in other departments too. The ORION algorithm can be laid atop other kinds of data. UPS is already using similar technology to predict which parts will fail next on its trucks. For example, UPS trucks have about 220 engine sensors report data that then gets analyzed to determine how one truck's alternator is behaving differently from another truck's alternator. With this data, UPS can tell the part is going to fail before it happens, reducing the failure rate on its trucks, as well as overall costs, by keeping the trucks on the road.

The direction UPS plans to take these systems, Levis says, is similar to what's going on with research into analytics-based medical devices.

"Imagine a world where we're all wearing our Fitbits and the doctor calls and says, 'I think you better get to my office because you're not reacting to heat the way other people in your demographic are reacting to heat", Levis says. "That's predicting. Clairvoyance will be where it automatically administers the medicine you need."

For UPS drivers, one of the next steps will be more efficient overall routing. The company plans to apply ORION not only to the package's final delivery route, but also to the entire UPS transportation network. ORION will sift through data that includes your front door coordinates as well as every data point between your welcome mat in New York and the office in Hong Kong where the package originated.

How many zeroes are at the end of the number that defines those possibilities? UPS is still figuring that out. One of the only things Levis knows for sure is that ORION will find efficiencies in places most people would never imagine.

"I think what we don't understand is the complexity we're under", Levis says. "That's the amazing part. You don't realize that there are so many ways to do your errands. You ignore the noise. The technology can analyze it. You might finish your errands thinking you did a great job, but ORION might come in and say, 'No, you could've done it an hour quicker."