Watch Now


Accounting for variability

Stochastic transportation modeling more like ‘human mind’ as it embraces variability, not averages.

   Mike Mulqueen has an interesting description for the average transportation management system.
   “I see TMS as an idiot savant,” he said. “It does some amazing mathematic things, but if you move the window two hours, it throws things off.”
   Interesting words considering Mulqueen is a senior director at Manhattan Associates, one of the biggest providers of transportation management software.
   Dig deeper, and what Mulqueen is saying is that the traditional form of TMS has relied on what in the mathematical world is called deterministic modeling. That essentially means a model with no variability, or randomness, built into it.
   In TMS terms, that means a system is built around a certain number of static input variables, and then the system produces a plan according to those variables. That’s a nice way of doing things for shipments that are repetitive and uncomplicated.
   But in the real world, transportation is rarely repetitive and uncomplicated. In fact, it’s the very definition of dynamic, and deterministic systems are largely unable to account for the variability inherent in getting cargo from point A to point B.
   Hence Mulqueen’s “idiot savant” comment.
   What Mulqueen and Manhattan have been building is the company’s so-called “stochastic modeling” capabilities. You can visualize stochastic as the counterpoint to deterministic, a type of model that attempts to incorporate variability.
   “That’s not the way the human mind works,” Mulqueen said about deterministic models. “You don’t want to say a supplier typically gives me three pallets per week, because that’s not accounting for variability. They don’t always give me three pallets. Sometimes they give me one and sometimes they give me five. There is a mathematically true average, but the average is meaningless if you don’t understand the variability.”
   Mulqueen has been advocating for more widespread adoption of stochastic modeling as far back as 2012, when he wrote a piece for Logistics Viewpoints opining on the subject.
   “When developing the data that make up the transportation policy, most analysts still rely heavily on statistical averages,” he wrote. “This approach completely ignores the inherent variability that underlies nearly every aspect of transportation. From order sizes, to lane volumes, to travel times to fuel prices, using a single, discrete value to represent each of these variables during strategic planning will lead to a poorly designed transportation policy that will ultimately be reflected as inefficiencies in daily planning.”
   Mulqueen chalked the reliance on deterministic transportation models up to ease of comprehension. It’s simply easier to understand averages than the way natural variability bends all the inputs that go into a transportation model.
   “Stochastic modeling tries to get at robust optimization,” said Chris Caplice, executive director at the Center for Transportation and Logistics at the Massachusetts Institute of Technology. “You’re trying to figure out what the variances are.”
   Along with that more robust modeling comes uncertainty. Simply put, the more inputs, the more variances you have to account for, the more uncertainty is created.
   “There’s demand uncertainty, or maybe supply uncertainty, and possibly price uncertainty,” Caplice said. “You’re not going to solve the optimization with a new random set of inputs. They give you a different answer each time. And companies have a hard time dealing with that. Stochastic modeling will cost more, because you’re building insurances. This idea that everyone knows I come up with a plan and it won’t be used.”
   Of course, that’s the idea of modeling a transportation network, or even an end-to-end supply chain with modern tools. You may model it a number of different ways, but you’ll only use one.
   Manhattan is not the only company offering shippers, logistics services providers and carriers the ability to stochastically model. JDA, another heavyweight in the TMS space, as far back as 2011 said it could plug stochastic inputs into its supply chain network design tool.
   Indeed, this is clearly the way forward for any transportation-oriented software provider that looks beyond the day-to-day execution tasks that most TMSs can already tackle with ease.
   Consultants that spoke with American Shipper on the matter said stochastic modeling is not yet all that prevalent in automated transportation procurement. Mulqueen, however, said Manhattan has already attracted engagements.
   “One of the big variables we’re modeling is just quantities,” he explained. “An organization will have all their historical data, but when we first start a modeling engagement, we spend a week or so doing all sorts of statistics analysis. Get all a customer’s orders over the past year.
   “The first thing I do is mathematically quantify their orders. I don’t care about last year’s numbers, because then I’m just optimizing last year’s orders. I take their history and apply some kind of forecast,” he added.
   Mulqueen said the initial statistical analysis is a bit of an art.
   “The setup of a modeling engagement is the hardest part—you’re setting a baseline,” he said. “Does this make sense to you? Does this look like a network? Once you have the baseline established, running the optimization is relatively easy. I know the constraints are right, I know the demand is right.”
    Mulqueen makes clear two things when he talks about the use of stochastic modeling’s application for transportation. First, transportation is only part of the puzzle.
   “I’m providing very accurate pieces of the puzzle,” he said. “I’m providing a piece of the calculus but not the totality of the calculus. What we’re doing is transportation. And there’s not one solution that handles all of these. There is supply chain modeling, but that’s not solving transportation modeling to the granularity that the VP of transportation would be happy with. Alternatively, if you just look at transportation, you’re missing things like taxes.”
   Second, as discussed in American Shipper’s November 2014 cover story (“Supply chain by design,” pages 8-12), modeling is something that needs to be undertaken over and above the daily grind of optimization.
   “Daily planning is about optimization,” he said. “There’s only so much you can do in daily planning. This [stochastic modeling] tool is not designed for incremental change, that’s the job of a TMS.”
   Mulqueen said the most advanced users of stochastic modeling currently are carriers.
   “Being a carrier is all about probabilities,” he said. “You take a load from Cleveland to Atlanta because there’s a high probability I’ll get a load from Atlanta to Chicago.”
   Shippers are gravitating toward the concept more methodically.
   “Our clients are going to be more progressive than the typical shipper,” he said. “They’re mostly $5 billion-plus in revenue. The leading edge guys, they get it.”
   But Mulqueen emphasized modeling isn’t just for multibillion-dollar shippers.
   “You can make a modeling solution as big or as little as possible,” he said. “You can do it for budgeting transportation spend, but more interesting things are figuring out where [distribution centers] should be located. Those are the big, impactful decisions.”

This article was published in the January 2015 issue of American Shipper.