Jack Levis will be keynoting at Predictive Analytics World conference in San Francisco Tuesday, March 31, 2015 and in Chicago June 2015.
When you have an organization the size of UPS – with 99,000 vehicles and 424,000 employees – every single little bit of efficiency that can be squeezed out of daily operations translates into a big deal. UPS has been using analytics to do just that for a long time now, and keeps getting better and better at it. Network World Editor in Chief John Dix caught up with UPS Senior Director of Process Management Jack Levis for an update on their latest achievements.
How does UPS use analytics to optimize its operations?
Let me take you back 15 years ago and then work our way back to today, and then I’ll give you a glimpse into the future. Also, to frame the discussion, let’s think of analytics in three forms: descriptive analytics says, “Where am I today?”; predictive analytics says, “With my current trajectory, where will I be headed tomorrow?”; and then at the highest level you have prescriptive analytics, and that’s where you say, “Where should I be?”
The research says that as you move up this hierarchy your data needs grow, the skill set of your people increases, and the business impact grows significantly, and that’s been exactly our experience.
Gartner says that in the descriptive space only about 70% of organizations really understand where they are. For us that’s just old news. We’ve been doing that for more than 20 years with our drivers’ hand-held computers, called the Delivery Information Acquisition Device (DIAD). In the predictive space they say only about 16% of organizations are doing that, and we deployed some predictive models 10 years ago. Then the prescriptive space where optimizations occur, they say only about 3% are there yet, and that’s where our ORION system comes in — On-Road Integrated Optimization and Navigation — which we’re deploying now.
I bring all of this up because your vision as an organization can’t stop at descriptive analytics, because there’s lots of value beyond that.
Do you avoid the term big data by design?
Big data is a “how”; it’s not the “what”. The “what” is big insight and big impact, and if you do that through big data, great. But the key is the impact and the insight. It wasn’t called big data when we started describing deliveries of more than 16 million packages a day and building terabytes and terabytes worth of data. We’ve been doing that since the early ’90s. So it’s just data to us. I care about what we do with it, since that is how we derive value from it. Analytics is about making better decisions. That’s why I generally don’t use the term big data.
Back in the late ‘90s we had a lot of descriptive data that told us where we were, and it was very detailed. We measured things in hundredths of a thousandths of a second. Why? Because if we can reduce one mile per driver per day in the United States alone, that’s worth up to $50 million to the bottom line at the end of the year. So we measure things very granularly because little things matter to us. Our culture is, if you worry about the pennies, the dollars will take care of themselves.
So in the late ‘90s we had lots of information about what happened yesterday, but changing tomorrow was hard. We were a knowledge-, methods-, and procedures-driven organization and some of the data was in the heads of employees, some was in corporate repositories, some was in Excel spreadsheets and some was distributed. But in the late ‘90s we didn’t have a predictive data model that described how UPS operates. So we took on a project called Package Flow Technologies.
The idea was, if we knew where every package was at every moment of the day, and where it needs to go and why, then we could just flip a bit and change where a package is headed tomorrow and that would help make us more efficient.
We deployed that in 2003. And the deployment of these predictive models and planning tools meant that, rather than a driver starting the day with an empty DIAD and filling it up, drivers started the day with a DIAD full of things we wanted them to do. So instead of an acquisition device, it became an assistant. With that predictive nature of looking forward, we reduced 85 million miles driven a year. That’s 8.5 million gallons of fuel that we’re not buying and 85,000 metric tons of CO2 not going in the atmosphere.
We have what we call “all services on board.” One driver, one vehicle, one service area, and one facility. But we have different services — deferred service and premium service on the same vehicle – so that driver has some packages that have to be delivered at 10:30 am, some that have to be delivered at noon, and some that have to be delivered at 2:00 pm. So that meant the driver had to decide how he was going to service these multiple commodities at the same time.
And even though we saved 8.5 million gallons of fuel, we wanted to take it further by prescribing, using some very advanced mathematics that says to the driver, “Let me reorganize today’s route based on today’s customers, today’s needs, today’s packages, and put it in a very specific order.” It used to be the driver would figure out how to handle anomalies. Now it’s in a very specific order, optimized with data and analytics.
Putting it in perspective, the advanced math around determining an order of delivery is incredible. If you had a 120-stop route and you plotted out how many different ways there are to deliver that 120-stop route, it would be a 199-digit number. It’s so large mathematicians call it a finite number that is unimaginably large. It’s in essence infinite. So our mathematicians had to come up with a method of how to come up with an order of delivery that takes into account UPS business rules, maps, what time we need to be at certain places and customer preferences. It had to be an order of delivery that a driver could actually follow to not only meet all the business needs, but with fewer miles than they’re driving today. And this is on top of the 85 million miles we’ve already reduced. This is the ORION system that takes it to the next level of prescription and that puts us in that three percent category of companies using data for prescriptive analytics.
Where do you stand with the ORION deployment?
We started initial deployments in 2012 using a few hundred deployment people, so it was a relatively small deployment. But the results were so remarkable we sped deployment. By 2013 we had 500 people deploying and now we have 700 people full time. We’re truly able to both simultaneously service our customers as they want to be serviced and reduce our miles at the same time.
How do you classify success?
We’ll be announcing our expected results on full implementation soon. But in 2013 alone we saved 1.5 million gallons of fuel, and that was with only 10,000 of our 50,000 drivers.
When do you finish the rollout?
We’ll complete the rollout of the current version of ORION by the end of 2016. There are some things it doesn’t do in its original version. For example, when a driver leaves in the morning, the route they have in their hand-held doesn’t change. It doesn’t update if something goes wrong, which is the No.1 request from drivers. They ask, “Can you update this when there’s a discrepancy?” So it doesn’t do that yet. It doesn’t take traffic into account yet. And, it doesn’t take weather into account yet. In fact, our drivers don’t even have the navigation system. They just have the order of delivery.
So that’s the bad news. The good news is that all those things don’t exist yet because we’re getting all of these gains without them, and that’s why this is a roadmap for the next 10 years. We’ll add new features. We are developing the ability to update in real time. We’ll use ORION’s algorithm all over the business.
Often when people talk about data, they say they want to go from data to information to knowledge. We did that when we did the predictive models. We’re making tomorrow’s decision from the prediction, so that was knowledge. After knowledge is wisdom, and that’s where we are with ORION, because a newer driver with his new order of delivery and a navigation system will be wise just like the driver who’s been on the route a long time.
“Imagine a data architecture and an analytics system of the future that predicts a problem is going to exist and solves it before you even know something is wrong. We’ll look like Sherlock Holmes.”
But after wisdom is really the Holy Grail, and that’s clairvoyance. Imagine a data architecture and an analytics system of the future that predicts a problem is going to exist and solves it before you even know something is wrong. We’ll look like Sherlock Holmes. And that’s where I think we’ll be one day. It will be a transaction-by-transaction optimization. What do I do with this particular transaction, at this particular time, with a brain like ORION pulling the strings? That’s where we’ll get to.
Let me end with a stupid question. “MythBusters” once did a piece exploring the myth that a delivery truck that only made right-hand turns would be more efficient. Have you found that to be true?
Let me tell you what is true. A left-hand turn is more expensive in multiple ways. Your vehicle is idling longer, it takes longer to make that left turn and it’s less safe. So we try to avoid left-hand turns. The “MythBusters” episode showed in San Francisco that only making right-hand turns was quicker for a truck than turning right and left. So our routes are set up in such a manner that we need fewer left-hand turns. That’s our method.
And by the way, that came from Parade magazine. They wanted to know what an average everyday person can do to reduce their fuel footprint and that was one of a number of things we supplied. As we told Parade, it’s important to use the right vehicle for the job (e.g. don’t take a minivan if you can get by with your compact car), consolidate trips (e.g. don’t go out twice if you only need to go out once, or park once and walk between the stores instead of driving between stores), and try not to make left-hand turns. And everybody picks up on the left-hand turns. That’s how it came about.
It’s the No.1 question I get every time I speak.
 |
 |
 |
