Parallel Equations

By November 4, 2013June 8th, 2014Best Practices
Print Friendly, PDF & Email

The Army has a lot of stuff to move. Amazon moves a lot of stuff.


Nearly everyone knows who Amazon founder Jeff Bezos is. Less well known is the other Jeff—Jeffrey A. Wilke, senior vice president of the consumer business at Inc. Having joined Amazon in 1999, Wilke has been in a leadership position at the company from the time it was a small startup to the present day, when it is the top U.S. company in e-commerce and one of the world’s largest retailers, with more than $60 billion in sales per year. But Amazon also provides a host of online services, including Web hosting, cloud storage, and music and video streaming.

Wilke’s background in chemical engineering—with a bachelor’s in the subject from Princeton University and a master’s in chemical engineering as well as an MBA from the Massachusetts Institute of Technology—might not seem the most natural fit for consumer retail, but then is not your ordinary retailer. Indeed, Wilke started his career writing software at what is now Accenture PLC, but came to Amazon from AlliedSignal (now Honeywell International Inc.), where he was vice president and general manager for pharmaceutical fine chemicals. He had spent the previous five years in operations and management in the chemical, polymer and electronics industries.

The thing about Amazon that stands out the most, of course, is that it moves a lot of materials, from books, kitchen appliances, consumer electronics and (no kidding) a do-it-yourself casket kit, to the data it manages through Amazon Web Services (AWS). The U.S. Army is an AWS customer, along with some 300-plus government agencies. But that’s not why Army AL&T magazine contacted Wilke for an interview. We wanted to tap his logistical mind on the subject of retrograde.

More than 750,000 major end items, worth more than $36 billion, are currently in Afghanistan, according to DOD estimates. Some two-thirds of those items are in U.S. Army hands. Dealing with these items is expected to cost close to $5.7 billion and will require intricate planning, teamwork, creativity and innovation.

Army AL&T wanted to know how one of the greatest movers of stuff in the world does it, and Wilke did not disappoint. In an Aug. 7 interview with Army AL&T, he offered insights on planning, modeling and management of logistics that reflect both his chemical engineering and his business training. As it turns out, sometimes the solution to a problem in one area of expertise is entirely applicable to another.

Q. You’re said to be a logistics whiz, and Amazon has an enormous logistics operation spanning the globe. The military in general, and the Army in particular, have a massive retrograde underway. If suddenly that were your job, how would you approach it, both logistically and organizationally, to make the task more manageable?

A. This seems from the outside to be a gargantuan task. As a citizen, I’m so proud and amazed that our military will complete this mission. If presented this challenge, I suspect my approach would go something like this. Step 1: Be humbled. Whatever my teams have achieved in the past does not secure our future. Only great planning and execution of this mission matters.

I’d start by ensuring we have the right leaders in place: Do we have great field leaders who know how to make adjustments to a good plan in the moments when local judgment counts? Do we have the best analytical minds, utilizing the most current modeling techniques and machine intelligence to build a robust plan?

Do we have as my direct reports the very best leaders for each critical function? Once I’m confident we have the right leaders in place, I’d want our senior team to understand the situation as completely as possible. What is the definition of “mission complete”? What dates are immovable? Where do we have flexibility? What budget for talent and money is available?

There are many ways to build shared understanding of the situation among the senior team. I’ve found scenario analysis, including inspecting the inputs and outputs of a detailed model, to be among the most effective. My brain would first model the retrograde as a classic transportation problem, where we have source nodes (places where we have assets to move) and sink nodes (places where we want the assets to be, including destroyed.) Arcs connecting the nodes have characteristic flow times and capacities. Built the incomplete way, this model would assume that all values characterizing the system were deterministic or without randomness. But such plans are almost always too brittle. One might say, “Let’s build a plan that assumes every day, every moment will operate at the expected value of each of the inputs.” Unfortunately, things don’t operate at the expected value. They operate with a distribution of outcomes.

I’d like to see plans that assume we are operating a stochastic system—one with variation. The best way to think about this is that there are a number of scenarios that represent the possible actual outcomes of something that is as complicated as the retrograde that the military’s planning.

There will, of course, be a range of inputs, the things that happen every day that are part of the plan. There will be a range of possible performance each day in those inputs. And that leads to a range of outputs that are almost guaranteed to not be the average values that you would predict upfront. Of course, we should audit the model outcomes for reasonableness, which helps us learn together. Planning for variation usually produces a deeper understanding of how the team should react if the actual mission goes awry.

Stochastic models help uncover potential bottlenecks. (Bottlenecks are choke points that determine the overall throughput of the system.) I believe you have a good operational plan when you decide where you want the bottlenecks to be. Surprise bottlenecks indicate poor planning. As my team examines the model inputs and outputs, I would direct us to decide where we will accept constraints (or capacity limits).

As planning progresses, I would pay particular attention to the resources and leadership assigned to each bottleneck. Our team would ask: Where do we have the most flexibility in the overall plan? Where do we have the least flexibility and the fewest options for recovery? We’d perform what-if scenarios: What if we lose transportation capacity? What if a particular load area comes under bad weather? What if we have political interference in a particular country? And then what you do when you play those “what-ifs” is, you look at the outputs of the model and ask, do they seem reasonable? I have no idea what the percentage is, but suppose that 20 percent of our assets are in a particular part of Afghanistan, and in that area it will be very difficult to get the permission that we need to move things out. Well, a scenario that we would run would be, what if we can’t move 20 percent of the items for an additional month, two months? If the model predicts that this has no impact on the ultimate mission, I’m going to be very skeptical.

Q. Of course, in the case of Afghanistan, that’s hugely complex, because shipment through Pakistan is such an on-again, off-again situation.

A. Right. So you just described a political impediment to achieving the expected value—what time we expect it would take for an item to move from where it is to a location where we have a little bit more control over our ability to move it. This might be the item that ends up having the most impact on the variation in the plan, the place where we would need to have the most flexibility because we can’t be sure of our underlying ability to meet the plan.

I would expect to find variation all over the place. We’ll find it in the capacity associated with the natural “batches” that we use in moving items from one point to another. Those batches are usually constrained by the size of trucks, convoys, railcars or ships. Sometimes you lose capacity because of mechanical failure, or you have to substitute one mode for another mode. Goods can arrive early, exceeding the storage capacity at the loading point. And then you have humans involved all over the place. I would suspect that there’s a standard time to load and unload each of the items that could be moved. Sometimes our human team members will perform according to those standards, and sometimes, for whatever reason (humans get sick or weather interferes), we’ll see variation in the cycle time for the loading and unloading of items.

You put all of those sources of variation together, and you’re likely to end up with an outcome that isn’t the expected value of what you planned. And I think it’s very helpful upfront to consider some scenarios for what you would do when the plan is not met.

Q. All of which underscores that this is a hugely complex and not entirely predictable operation. What you’ve described, the what-if planning, is something that goes on in the military all the time.

A. Yes. The what-if planning has been going on for a long time, but we now have modeling and computing power that allows you to build more sophisticated models. The advantage of those is that you can sit with people and they don’t have to imagine it in the same way. You can do active simulations—imagine that we were in the heat of the moment and the following thing happens, what would we do? Well, you have the model in front of you, and you can, in near-real time, prepare answers with precision that in the past just wouldn’t have been possible because we didn’t have enough data in the models.

Q. What exactly did you mean by auditing the outcomes predicted for reasonableness?

A. I just mean that it’s an intuition test. When you have these kinds of computer models, the most valuable thing that they do, I think, is to help humans, especially in a one-off project like this. If you can run models regularly over a long period of time, eventually the model gets good enough that you don’t need that much human input.

An example is a control system in a complicated petrochemical plant. These plants run through significant transients, or periods of variation, without a lot of human input, because the control systems have been running for a long time. They’ve been tuned, and the computers know what to do. In this one-off project, any simulation that you build is not going to be such that the machine can run the project. It’s going to be such that the humans are better prepared to lead the project.

Q. So they’re all on the same sheet of music, so to speak.

A. Exactly. The best audits are performed by the humans who, during the execution phase, will actually be managing and leading. In advance, you can have those humans sit around with the computer models that you built and test them for reasonableness. You start with human intuition about how robust the system will be, or how long things will take, or how effective we can be. And you want to look at the outputs of the model and use that great human intuition and ask, does it make sense? If a human looks at it and says, “I’m glad that analysts have predicted that this is what’s going to happen, but I can tell you, I’ve been in the field, and I know that this particular step is going to take longer than the model’s predicting,” then we can make the plan better. In these audits of the output of the simulation, we’re trying to catch things, assumptions, that are wrong in the models, applying human experience and intuition.

I’ve been talking about this from the perspective you asked me to think about, as if I were the leader of the whole thing. These models are incredibly useful at all levels of execution of such a mission. Since this precise military challenge has never been completed—just as no one had been through direct-to-consumer logistics challenges like we had in the early 2000s—I would not expect to have a computer model direct movement autonomously.

Thinking about this problem reminds me of our early holiday seasons at During our peak four or five weeks of ordering, which is between Thanksgiving and Christmas, we have an increase in our logistics activity of about three to four times the average rate for the rest of the year. So the challenge, of course, is to have a team ready to perform to a playbook that is very different during those four to five weeks than for all of the other weeks of the year. That is the primary leadership challenge at Amazon.

So, for example, in the Amazon world, these kinds of models aren’t just used by the corporate staff; they’re used by the staff running each of the individual warehouses in our network. They’re used by departments inside of those warehouses in the network, again, to do simulation and prepare in advance of the holiday peak, which is sort of a mission for us. The great thing about these models is that you can share them very easily, or parts of them very easily, with a whole range of leaders across the organization, which makes them all better prepared.

I would expect to have leaders on my team check in regularly with the field leaders to evaluate how closely we were delivering to plan, with our analytical team and automatic systems adjusting the models regularly for mid-course correction. During the early holidays, I held daily conference calls every morning where fulfillment, supply chain and transportation leaders provided key network status details. We would modify the allocation of resources in real time to balance the network, keeping our bottlenecks always front of mind.

It boils down to examining the actual performance versus the expected value for some of the elements that measure the networks. So, for example, are the queues that we have the size that we expect them to be? The consumer really wouldn’t see it, but a queue would be orders that we’ve taken from customers that we haven’t yet moved through a particular step in our logistics. It’s an accepted but unfilled order. They might be on a truck someplace. They might be in a particular state in a warehouse. The order might occasionally be stuck in a software queue.

Were the bottlenecks that we experienced yesterday where we expected them to be? Was productivity in each step of the process as we expected it to be, or not? When you have productivity that’s low, if it goes low enough in a particular step, that step can become the bottleneck for the day. Then, all of the steps behind the bottleneck push work faster than the bottleneck can process it, and you build up a queue. I would ask for an expected value versus actual performance. In each of these steps, you begin to build a fairly complete picture of how the network is performing.

If we were running an operation at the same rate all year long, it would be more like the petrochemical plant scenario, where you would just tune your systems and your people to a certain way of doing things and just do it forever. But we don’t have that luxury, because our consumers order a lot more from us during the holiday peak.

When the mission was complete, I’d expect to spend quite a bit of time saying “thank you” to all the folks who made this incredible performance possible.

Q. Speaking of the holiday shopping season, how do you manage to keep people motivated during this season and, in addition to all your other logistical tasks, make sure that people understand that their work is appreciated and that they’re doing a good job—or not doing a good job, as the case may be?

A. That’s such a great point. Throughout the holiday, our leaders—and, in fact, many of the folks who would normally be working directly on customer orders—for the holiday season actually end up serving as ambassadors, or leaders for associates who have recently joined us. So we have a lot of presence on the shop floor. We’re visible on the shop floor so that we can offer assistance and we can pat people on the back and thank them for a hard day’s work. We try to make it fun. Break rooms get decorated and some of the folks will occasionally get quite Christmas-y with their attire. It’s very motivating to see orders that you know are going to end up as wrapped presents under a tree. It’s emotional and powerful for people.

All Amazonians are deeply passionate about delivering a great customer experience. At Christmas, that means that the right product goes to the right address on time in a great condition. As you watch these orders go by, you know you’re basically serving as an elf in some way to help families all over the country. And that feels pretty good.

Q. In terms of product logistics, does the handling of books differ from what Amazon does with flat-screen TVs, or groceries or automotive supplies?

A. Yes, different product attributes require different logistics solutions. Some key attributes include: size and weight (can the item be handled by automated sorting equipment?), fragility (clothing items require different handling from most packaged automotive supplies, and some items, like food, are very temperature-sensitive), sales velocity (lower-velocity items may be located in fewer fulfillment centers, or FCs) and cost (we might not store $5,000 watches right next to $20 books.)

Q. A major area of study for the Army logistics community, as it prepares to draw down from Afghanistan, is core competencies. Does Amazon employ logisticians per se, besides you? If so, what skills is the company looking for?

A. I am certainly honored to be considered a “logistician,” though that is not my formal background. I have an undergrad in chemical engineering from Princeton and graduate degrees, and an M.S. Chem E and MBA, from MIT’s Leaders for Global Operations program. I started my career writing software at Accenture. I suspect I think about logistics systems as networks of nodes and arcs, or pipes, valves and tanks, because of my chemical engineering background. What matters is the ability to think analytically about such problems, using the math of optimization, process control and as much computer science and machine learning as possible.

Amazon does employ logisticians—lots of them. Some have backgrounds like mine, where they’ve made a switch from a different technical field to this one. Many employees in this area are computer scientists or software development engineers encoding our algorithms in software. We’ve found that individuals with a military background do incredibly well in roles all over our operations’ organization. They have the right bias for action and comfort with rapidly changing environments.

Q. What kinds of data does Amazon collect on its supply chain, e.g., safety stats, and how do these data reflect how Amazon likes to operate?

A. We start every operations meeting with a safety tip. Every operations metrics deck starts with our safety performance. In fact, it is safer to work in an Amazon fulfillment center than in a retail department store. Beyond safety, we measure everything you might expect us to worry about in a complicated logistics network: customer experience, cycle times (both mean and variance), defect rates, productivity, cost and capital investment. These metrics map to our leadership principles, which include customer obsession, ownership, frugality and delivering results.

Q. Amazon aims “to be Earth’s most customer-centric company where people can find and discover anything they want to buy online.” You personally are reportedly almost fanatically committed to customer satisfaction, especially during high-pressure times such as the holiday season. The Army acquisition community’s customers are the Soldiers who use the items procured. What could the Army learn from Amazon about managing “customer” relationships?

A. We have two leadership principles that help define Amazon’s approach to customer-centricity. First, we want leaders to display “customer obsession.” We ask leaders to start with the customer and work backward. They work vigorously to earn and keep customer trust. If you ask, “How obsessed is enough?” we answer with our leadership principle “insist on the highest standards.”

We think leaders have relentlessly high standards; many people may think these standards are unreasonably high. Leaders continually raise the bar and drive their teams to deliver high-quality products, services and processes. Leaders ensure that defects do not get sent down the line and that problems are fixed so they stay fixed. So you will find us correcting mistakes before customers get angry about them, sometimes before they even notice. We’ll drive to stores on Dec. 24 to buy the few out of hundreds of millions of items that we just can’t find in the FC. When we build new products, we write press releases to help visualize what we would like to announce to customers about the product at a successful launch. I love that the words “processes” and “fixing defects” are in our highest-level leadership vocabulary. Great customer experience starts with superb attention to execution.

I had the honor of visiting the Army War College at Carlisle, PA, some years ago, and I was struck by just how much consideration the Soldier-customer received by the military’s highest leaders. With respect to how we think about our employee customers, we borrowed some of our approach from the best practices of our military.

Q. You mentioned that you found that military people have a penchant for action and are comfortable with rapidly changing environments. Can you give me a little bit of background on how much Amazon has worked with military people?

A. For 13 years, we’ve been actively recruiting everyone from former enlisted folks and junior military officers to more senior officers because of these traits. And it’s proven to be a highly successful hiring channel for us. There are hundreds and hundreds of folks who are veterans who are at Amazon. And we expect to hire 1,200 this year.

I mentioned my own experience with the Army War College. That experience was just prior to my joining Amazon in 1999, and so it was very fresh in my mind. The new leader for human resources for worldwide operations in 1999, Dave Niekerk, was a West Point alum.

Q. You talked about leaders having relentlessly high standards. That’s exactly the kind of thing that DOD is trying to get at these days with better buying power. Central to that is raising the standards of leadership and independent thinking in acquisition professionals. DOD is working very hard to have top-level acquisition professionals recognized and documented as the elite that they are. What you said about high standards gets right to that point. Does Amazon ever do anything like better buying power?

A. Not per se. We have 14 leadership principles (online at, including the leadership principle “insist on the highest standards.” This principle of independent thinking is close to one that we have called “disagree and commit.” What we want our bright people to do, when they’re pretty sure they’re right, is to disagree when they think that somebody else’s approach is not correct or they think they can improve on an already good approach.

And it’s interesting, because people asked for a long time, “Well, when do you stop disagreeing?” My answer is, when you’re not sure you’re right. So I tell people—I don’t know how this would work in the military—but in our world, if you’re sure you’re right and your boss tells you, “No, you’re wrong,” you have the obligation, after telling your boss that you are going to do this, to go to his boss or her boss. And if that boss says, “You know, I don’t think you’re right” and gives their reasons and you’re sure you’re right, you have the obligation to keep going up the chain until you get to our CEO. And if he tells you, “I hear you, but we’re going to do this other thing, then we want you to disagree and commit to the choice that the company’s made and move forward.”

So I’ll tell people, when you’re not sure and your boss asks you to do something for us to move forward, the right thing to do is to say, I don’t agree with that, but I’m not exactly sure, so I will commit to the plan of record, and I won’t complain about it. We’ll just move forward.

Q. Obviously the chain of command in the military means so much in terms of order and discipline and just getting things done. Once a course of action’s been decided on, you get less of that open dissent. But the principle you’re talking about is definitely a value to the military. What about the other leadership principles you mentioned?

A. They start with customer obsession. Everything we do, we start with a customer, and we work backward from there. The second principle is ownership. We want leaders to behave like owners, and mostly that means that they think long-term, that they don’t sacrifice long-term results or long-term value for short-term results. We want people who never say, “I’m not doing that ’cause that’s not my job.” We want people to do whatever the mission requires. The next one is “invent and simplify.” We require our leaders to be innovative even when others don’t understand what they’re doing, and sometimes they won’t understand for a long time. The next one is “are right a lot,” and that basically means we want to hire smart people. We think leaders need to have a level of intelligence that makes them capable to do their work of leadership well.

We want to hire and develop the best first. That’s completely consistent with the military. We talked about “insist on high standards.” We want our leaders to think big. Very simply, we think that thinking small is a self-fulfilling prophecy. So we ask people to envision bolder directions because they’ll lead to bold results, and it also inspires the team to think differently.

Frugality is the next one, and that’s about not spending money on things that don’t matter to customers. We want our leaders to be vocally self-critical, and actually this is my favorite line in all of our leadership text. We say leaders do not believe that their or their team’s body odor smells of perfume. I love that one because we’re all human, and if we can’t say, “I was wrong, I don’t know” and are just full of ourselves, our teams especially will see right through it. And I’m sure that it’s that way in the military. When an officer has screwed up and doesn’t admit it, I guarantee you that they lose some respect from the members of their team.

“Earn trust of others”—we want leaders to be able to show respect and to gain trust. We expect leaders to dive deep, really deep, because we think no task is beneath leaders, although they can’t, of course, do everything all the time.

There are two more. The second-to-last is “have backbone.” That’s the idea of disagree and commit. And we want leaders to respectfully challenge decisions when they disagree. Finally, we expect leaders to “deliver results.”

Q. Given that a lot of materiel that will be brought home from Afghanistan will be obsolete before long, what would your priorities be, in broad strokes, to position the Army for the future when the retrograde effort is over?

A. First of all, sunk costs are sunk. If we’ve spent money on something that is now worthless, its value is $0, not whatever we paid for it. I would make sure we have programs to maintain the materiel that has ongoing value first. Some items provide “option value,” in case we do need to spin up again. I would plan to hold on to additional materiel beyond peacetime minimum, especially where the lead times for replacement are very long. Finally, I would dispose of what we cannot expect to use, hopefully salvaging value by selling to the private sector or using the assets in other branches of the government.

Post-retrograde, I would make sure the Army’s processes and logistics information systems are ready for the future. I’d use our recent experiences to build sophisticated simulations to help keep our people fresh while they wait for the next crunch.

Q. How would you characterize the corporate culture at Amazon? What are the top three defining characteristics that you want Amazon employees to appreciate fully, and why?

A. Our culture is customer-obsessed, fast-paced, and truth-seeking. We try to hire people who are smart, possess high standards and know how to get things done. Importantly, we find that we have the best match with people who would say they feel “fortunate.” Such people are most likely to say “yes” instead of “no” and to foster an innovative, optimistic environment. We’ve found that individuals with a military background do incredibly well at Amazon in a variety of roles. Our company’s leadership principles match up closely with our nation’s military. As a result, for years, we’ve actively recruited members of the military into roles at Amazon, whether they are retired from the active military or reservists.

Q. Tell us about CRAP—or “can’t realize any profit”—which seems almost as much about streamlining customer experience as it is about unprofitable items. What can the Army learn from the philosophy behind CRAP?

CRAP came from observations Jeff Bezos and I made while working on the shop floor in Kentucky one Christmas. We were spending too much time (to the great pleasure of the associates watching us!) building custom boxes to hold folding chairs offered as that day’s big deal. There was no way, even with well-trained, productive employees (versus us), that we were making money on these items. We committed ourselves right there to find other similar items and either make them profitable or stop selling them. Most of the time, our vendors have been able to work with us to make such items profitable, including by changing the packaging.

Q. You and founder Jeff Bezos used to go out and spend a week each year visiting Amazon distribution facilities, even fulfilling orders. What benefit, in your view, would it be to Army brass—both military and civilian—to spend time working at the ground level a week each year the way you did?

A. I think a significant portion of leadership is showing up, in person, to listen, say thanks, and “walk the talk.” As our jobs get bigger, it can seem harder to find the time to visit. Certainly at many times in my career, I’ve fallen prey to this mistake. But I know that face time matters. Associates notice. I suspect our Soldiers and civilian employees notice when leadership takes the time to visit.

Recently, a longtime “Amazonian” software developer decided to leave our company. On his last day, he sent me a very personal note. He thanked me for some things, and then he offered a piece of incredibly valuable, sincere advice. He noted that in the early days, we were small enough that I communicated regularly and informally with the entire corporate staff. As we grew, they saw less and less of me, to the point where now some new engineers joked that I might not actually exist.

He had a simple suggestion: Pick some lunch areas randomly and occasionally, and just show up with little announcement. Engage in Q&A with whoever happens to come. Though I probably haven’t done it frequently enough, I implemented his suggestion and have been pleasantly surprised with the attendance, questions and, most importantly, the opportunity to preserve our special culture by tying my answers to our leadership principles backed up by a few stories from our past.

Q. Jeff Bezos is fairly well-known for his vision and ability, and part of that vision was hiring someone like you to rationalize and make that vision work. What’s the best part of your job in supporting that vision?

A. I believe that the field of operations matters. Too often companies, governments and other entities build great programs and products, only to have them fall short of their potential impact because the underlying operations just don’t scale. At Amazon, we understand that process, technology and especially leadership make a huge difference in our success or failure. I love working at a company that’s proud of