If the pandemic has taught us anything, it’s the value of having goods delivered right to our front door. From fast food ordered on an app, to next day shipping from an online store, the distance from click to delivery is getting shorter all the time. What most of us don’t realize is that the hardest part of delivering an order is the last mile.
More Intelligent Tomorrow host Ben Taylor asked researcher Dr. Matthias Winckenbach to explain what he means by “the last mile” and why it’s so difficult to cover.
The term “the last mile” refers to the home stretch of the supply chain where goods are moved from the final distribution location to the customer. It can represent as much as 40% of the overall supply chain costs despite almost always being the shortest segment. This is why there’s a lot of effort going into optimizing it.
"The cost and complexity of last mile delivery very much comes from the fragmentation of shipments."
Customer expectations for faster, more flexible deliveries, with shorter lead times are running up the cost of covering the last mile. But financial costs aren’t the only concern. Our increase in demand might be having a negative impact on the environment as well. Increased demand means fewer opportunities for suppliers to consolidate, and that means more trips are needed to deliver the additional load of packages. Those extra trips use more fuel and put more traffic on the road.
The challenge is that you're adding complexity to the system. Now you're managing inventory across many more locations while trying to predict demand in any one location. With this move to hyper localized inventory, the traditional methods of optimization must be updated.
There’s a theoretical optimal solution to routing a package. It’s sometimes referred to as the Traveling Salesman Problem. But variables like traffic, customer availability, parking, and even the mood of the driver can impact the delivery in ways the optimal solution can’t predict. And all these things can change once the delivery is underway, meaning the optimal solution may no longer be valid. AI/ML can complement traditional planning methods to create better routes.
Ben wonders if aerial delivery drones are the answer. They can fly directly to a location and don't have to deal with things like traffic and parking.
Matthias thinks the answer seems to be no, at least not right away. There are regulatory, technological, and societal issues we need to overcome first. The density of deliveries in an urban area could lead to large swarms of drones buzzing overhead. The sky would fill up quickly.
From an economic point of view, it’s hard to beat an effectively designed ground delivery route. Drones are still expensive, and you’d need up to 300 of them to match the capacity of a single truck. But one day, there might be a hybrid solution where a drone and delivery truck could collaborate to deliver your package.
"A lot of people don't realize how difficult it is to get that package to the doorstep every morning."
Listen to this episode of More Intelligent Tomorrow to learn:
If the pandemic has taught us anything, it’s the value of having goods delivered right to our front door. From fast food ordered on an app, to next day shipping from an online store, the distance from click to delivery is getting shorter all the time. What most of us don’t realize is that the hardest part of delivering an order is the last mile.
More Intelligent Tomorrow host Ben Taylor asked researcher Dr. Matthias Winckenbach to explain what he means by “the last mile” and why it’s so difficult to cover.
The term “the last mile” refers to the home stretch of the supply chain where goods are moved from the final distribution location to the customer. It can represent as much as 40% of the overall supply chain costs despite almost always being the shortest segment. This is why there’s a lot of effort going into optimizing it.
"The cost and complexity of last mile delivery very much comes from the fragmentation of shipments."
Customer expectations for faster, more flexible deliveries, with shorter lead times are running up the cost of covering the last mile. But financial costs aren’t the only concern. Our increase in demand might be having a negative impact on the environment as well. Increased demand means fewer opportunities for suppliers to consolidate, and that means more trips are needed to deliver the additional load of packages. Those extra trips use more fuel and put more traffic on the road.
The challenge is that you're adding complexity to the system. Now you're managing inventory across many more locations while trying to predict demand in any one location. With this move to hyper localized inventory, the traditional methods of optimization must be updated.
There’s a theoretical optimal solution to routing a package. It’s sometimes referred to as the Traveling Salesman Problem. But variables like traffic, customer availability, parking, and even the mood of the driver can impact the delivery in ways the optimal solution can’t predict. And all these things can change once the delivery is underway, meaning the optimal solution may no longer be valid. AI/ML can complement traditional planning methods to create better routes.
Ben wonders if aerial delivery drones are the answer. They can fly directly to a location and don't have to deal with things like traffic and parking.
Matthias thinks the answer seems to be no, at least not right away. There are regulatory, technological, and societal issues we need to overcome first. The density of deliveries in an urban area could lead to large swarms of drones buzzing overhead. The sky would fill up quickly.
From an economic point of view, it’s hard to beat an effectively designed ground delivery route. Drones are still expensive, and you’d need up to 300 of them to match the capacity of a single truck. But one day, there might be a hybrid solution where a drone and delivery truck could collaborate to deliver your package.
"A lot of people don't realize how difficult it is to get that package to the doorstep every morning."
Listen to this episode of More Intelligent Tomorrow to learn:
Matthias Winkenbach (00:00:00):
What I would call the traditional operations research methods, like optimization, simulation, that kind of thing, while now there's hardly any research project that we do that it doesn't at least have a data science dimension to it, that doesn't even consider using what we now call machine learning methods to at least help solve the problem.
Speaker 2 (00:00:25):
Welcome to More Intelligent Tomorrow, a podcast about our emerging AI-driven world, critical conversations about tomorrow's technology today.
Speaker 2 (00:00:39):
On today's episode, host Ben Taylor sits down with Dr. Matthias Winkenbach, Director of the MIT Megacity Logistics Lab and a research associate at the MIT Center for Transportation and Logistics.
Ben Taylor (00:00:54):
Matthias, I'm excited to talk to you. And I would say but I guess it's where you are. What time is it there?
Matthias Winkenbach (00:01:01):
Right now it is , yeah. But yeah, thanks for having me.
Ben Taylor (00:01:04):
Yeah, thanks for joining us late. So to set the stage for the topic that we're going to dive into I'd love for you to describe to the audience what the last mile means to you.
Matthias Winkenbach (00:01:16):
Yeah. I mean, obviously you get different answers from different people, depending on who you ask, what they consider to be the last mile. We typically refer to it as really the last stretch from let's say, the last fulfillment facility or the last facility within the larger supply chain at which an item or a shipment is being touched before it actually gets sent out to the final customer, be that the business customer, be that the consumer, be the part of consumption like a restaurant. But that last leg of transportation basically from the major facility to who actually uses or receives the item, that's what we consider the last mile.
Ben Taylor (00:01:56):
And what is the cost ratio? That surprised me doing research for this session. How much does the last mile cost compared to the full supply chain pipeline? What percent?
Matthias Winkenbach (00:02:06):
I'm glad you asked that because some people underestimate the true cost of last mile logistics. It obviously depends a little bit on the industry, but in most industries that I know of, the last mile, even though it's the shortest final leg of the typically global supply chain, it is the largest contributor to overall supply chain cost. For instance, in the fashion supply chain, a good ballpark estimate would be around 40% of total supply chain cost would actually be related to last mile. So that's why even though it seems such a tiny part of the global supply chain, companies put a lot of effort in optimizing the heck out of it.
Ben Taylor (00:02:49):
Is that fraction increasing? Because I know during COVID, we now have grocery delivery, we have a lot of things that were not being delivered before, a lot of food and meals are being delivered now.
Matthias Winkenbach (00:03:02):
The question of the effect of COVID is probably not that easy to answer because I would figure that COVID raised supply chain costs on all sorts of levels. But in general, I would say that last mile logistics costs have been increasing relative to the remaining supply chain costs simply because at least for many products, for many types of shipments, customers are expecting faster delivery lead times, more flexible delivery, and all of that adds to the complexity of last mile logistics which at the end of the day results in cost.
Ben Taylor (00:03:33):
We talk a lot about intelligence on this podcast and how humans, we're very intelligent compared to other animals because we can actually intentionally delay gratification and we can plan for the future, but I feel like with some of this stuff, same day delivery, one day delivery, are we going backwards? What are your thoughts on whether or not we should be doing this? Should humans have same day delivery? Is there a negative cost from a global economy or from a environmental perspective when it comes to sustain-
Matthias Winkenbach (00:04:03):
Yes and no. So obviously things like same day delivery or even faster service peaks like pretty much on demand delivery are on the rise, and obviously that leads to additional costs that leads to less ability of companies to consolidate deliveries for instance and therefore, it typically increases the amount of transportation needed. So basically the number of miles driven per package delivered. The faster you have to move things, the less time you have to consolidate the more individualized your shipments become. So it seems like at least at first sight, the net effect of this on sustainability, be it emissions or congestion or whatever you want to measure sustainability by, is negative. But at the same time, what we observe in our research is that due to this increased speed and flexibility, companies also need to adjust their overarching network structure, so basically where to fulfill things from.
Matthias Winkenbach (00:05:07):
When we were in the times of next day, two-day, five-day delivery, fulfillment typically happened at least in the eCommerce space from some really big fulfillment center somewhere in the middle of nowhere, outside of a city, far outside of a city quite frequently. So the only way to actually get stuff into cities from there would be large typically combustion engine vehicles. Now that we are looking at things like same day delivery or instant delivery, stuff already needs to reside closer to the consumer before it even gets ordered. We are working with many of our research partners on more decentralized network structures with smaller satellite facilities in between the big inventory holding facility, like the big fulfillment center outside of the city, and the customer. These satellite facilities are already typically within the boundaries of the city, close to the customer, and the interesting thing about this is it enables a different choice of modes. So we can use more ecologically friendly vehicles for the actual final mile because these facilities are closer, you don't have to travel that far and so typically smaller and electrified vehicles for instance become more attractive.
Ben Taylor (00:06:22):
That makes sense. So you're saying with these satellite centers where we can begin to store material closer to the site then it opens up more possibilities, potentially a drone or other future capabilities, that are more... They're electric or they're more friendly than a combustion engine.
Matthias Winkenbach (00:06:38):
Yeah, exactly. So we've, for instance, worked with companies in the fashion retail industry or beverage industry who are facing the same challenges from their customers as let's say, the typical eCommerce platforms that we all know of are facing them, namely that customers want things faster, more ad hoc, more on demand and the only way to basically satisfy this growing need is to have everything already close enough to the consumer. The big challenge here obviously is you're adding complexity. It's much more complex to, for instance, manage inventory or manage transportation even across a network of let's say, 10 decentralized satellite facilities that let's say, serve the New York City market compared to serving that same market for one or at most two really big fulfillment centers somewhere outside of New York City. Even the complexity of the underlying planning and optimization problems grows exponentially basically because you're suddenly fragmenting inventory or fragmenting transportation and you have to make these planning decisions at a much higher level of granularity, basically.
Ben Taylor (00:07:56):
It's a very complicated problem. And I think you've talked to this, how it doesn't get boring. It's constantly evolving. How much has it changed in the last decade as you've looked into it and you've studied this?
Matthias Winkenbach (00:08:07):
When I was doing my PhD on this topic basically we were still living in a slightly different world. So I was actually working with a French postal operator back then on their first considerations of a decentralized network for a more efficient parcel delivery back then, and now pretty much every discussion that we have with companies all over the world across all industries are evolving around that same idea of becoming more local, moving towards hyper local fulfillment, changing their underlying network structures, changing the way we are routing vehicles based on this new network topology, changing the way we think about new vehicle technology, but also honestly changing the way we think about data and methods. Because let's say, 10 years ago in our discipline people were predominantly occupied with what I would call the traditional operations research methods like optimization, simulation, that kind of thing, while now there's hardly any research project that we do that it doesn't at least have a data science dimension to it that doesn't even consider using what we now call machine learning methods to at least help solve the problem. Not saying that operations research is completely being replaced by data science, but the two go hand in hand and actually compliment each other quite well.
Ben Taylor (00:09:34):
When you're talking about parcel efficiency I think the thing that most people think about in our space is they immediately think of the traveling salesman problem, and how do you think about parcel efficiency? Because I imagine there's a theoretical limit that we can never obtain because of changes in the state, like changes in traffic, there's things that are happening during the day before you set your route, so how do you think about parcel efficiency or route efficiency?
Matthias Winkenbach (00:10:03):
Technically, there's obviously a theoretical optimal solution to the traveling salesman problem or whatever the underlying routing problem is if you want to reach a certain set of customers on a given day, but reality always deviates from that for various reasons. As you said, there's uncertainty related to traffic, to availability of customers, availability of parking, and even the mood of the driver may have an influence on what actually happens on the ground.
Matthias Winkenbach (00:10:35):
Let's say in practice, we will probably never reach a point where we can 100% execute on the truly optimal solution to a routing problem because of all these uncertainties and because they unfold, as you said, throughout the day as the route's already ongoing. So that's actually one good example for where data science methods help us complement traditional planning methods, so optimization based methods, because at the end of the day, for many real world applications, it's actually not that important to find the theoretical optimal solution to the TSP, but rather to find a good solution that is close enough to the real complexity of the operational environment such that the driver can actually execute on the route plan as closely as possible.
Matthias Winkenbach (00:11:29):
So that's probably a little bit abstract, but to make this more tangible, think of a last mile delivery driver who probably delivers to the same neighborhood, the same route territory every single day. And over time, you collect a lot of historical data on the route sequence that he actually followed and the times that he actually spent serving different customers along the route, and they're most likely different from your assumptions that you used when you were theoretically solving the traveling salesman problem. Now, you can actually use data science methods to extract the tacit knowledge of the driver about traffic conditions, about customer specific requirements, because the driver typically knows at what time of the day a certain part of the city is particularly congested or a driver typically knows where he can find parking easily at what time of the day or when a certain customer might be available or not available. That kind of knowledge of the driver is extremely hard to encode in a pure optimization model that solves the TSP.
Matthias Winkenbach (00:12:31):
So instead, what you're trying to do is you're trying to observe the historical performance of the driver, learn from that, basically learn from where he or she actually deviated from what would be the theoretically optimal sequence to serve the customers. And because he or she probably does that for a reason, you can learn from that and recalibrate your planning model, so your traveling salesman problem optimization tool, to basically find a better solution next time. And that solution is not necessarily better in the sense of shorter or faster or more efficient in terms of let's say, miles per package, but it is more executable. It's closer to the reality faced by the driver on a day to day basis.
Ben Taylor (00:13:16):
That's really interesting because I imagine if I was a driver and you're trying to interview me some of these insights might be difficult because it's my intuition, I do this for a reason. But you have all the data, you have the GPS tracking, you know what I'm doing so you have my behavior during the day.
Matthias Winkenbach (00:13:33):
Yeah. And if you have a driver fleet of 10 people, individual interviews may be doable, but if you are dealing with a fleet of hundreds, it's no longer scalable. And also it's a dynamically evolving problem, traffic conditions may change as a function of weather or season or whatever, and that means you would have to repeat the interview process very frequently and that gets annoying and also, again, not scalable. So having a data driven method is a good compromise between asking the driver directly and having a scalable approach to solving the problem.
Ben Taylor (00:14:14):
That makes sense. And I think when people think about the future of extreme efficiency, everyone goes to drones, and I've heard you say on previous interviews that you don't believe we will go to a drone only delivery system because I imagine that would be perfect deficiency. Every package has a drone and it's going the way the crow flies, directly to the customer, so maybe explain why a drone only delivery service probably isn't practical in the longterm. What are your thoughts on that?
Matthias Winkenbach (00:14:42):
There's a variety of reasons, and I'm not actually an expert to speak on all of them. For instance, I know that there are many hurdles to still overcome just on the regulatory side to make individual customer deliveries by drone for commercial purposes a reality at a larger scale. I know there are pilots to do that already, but let's say serving a city like Boston entirely with drone based package delivery that's probably still a few years out just from the regulatory point of view. But more importantly, I don't think that this would be a reality that it would even be desirable for several reasons.
Matthias Winkenbach (00:15:21):
From a social perspective, I don't think it would be more desirable to have thousands of drones buzzing around our head replacing basically the relatively large fleet of delivery vehicles that we see on the roads today. Just think of the average UPS, FedEx, DHL truck out there, they carry three, 400 packages probably on one vehicle. That means you would have to replace that vehicle by three to 400 drone trips for instance, and that's just one vehicle. So you can imagine how congested the sky would get at some point, and that's not desirable for the people living there and it's not really easy to solve from a safety point of view. But also honestly from an economic point of view, it doesn't make sense. It's actually extremely hard to beat a well consolidated route of an efficiently operated ground delivery vehicle.
Matthias Winkenbach (00:16:17):
So our research actually suggests that the sweet spot might be twofold. One is you actually think of systems where the drone and the ground vehicle collaborate on the delivery process such that you basically outsource the really costly or the really difficult to reach customers to the drone, which may either operate separately from the vehicle or which may actually in the future take off from the vehicle and get back to the vehicle, but such that you basically still make use of the ability to consolidate of the big ground vehicle but you basically free up the most constraining stops from the ground vehicle tour, give that to the drone, such that the ground vehicle too actually becomes more efficient even though the individual deliveries by drone are still relatively costly taken by themselves.
Matthias Winkenbach (00:17:06):
Because you have to imagine that drone is also not necessarily cheap to operate even in a world where we are talking about full autonomy, the component cost and the lifespan of that drone make it relatively costly on a flight hour basis compared to a truck that runs for 15 years without a problem. And as long as you don't reach the full autonomic state, you actually need a highly educated person piloting this thing or at least safety piloting this thing, which also makes it quite costly per hour and per package. And that's at least ecologically speaking, I mean lifting stuff in the air, moving it around and making sure it doesn't fall out of the sky is one of the most energy intensive ways of moving things from A to B. So unless you already control where the energy comes from so you make sure that the drone is actually being charged with green electricity, if you wish, then the net benefit for, for instance, greenhouse gas emissions is also not that clear to me.
Ben Taylor (00:18:08):
That makes a lot of sense. I hadn't thought about the aesthetics where people complain about power lines being above the ground and how they're ugly. And you're right, if you had thousands of drone flocks in the sky I'm sure a lot of people would realize that wasn't a reality they enjoy, whether it's the noise or even just the disrupted view.
Matthias Winkenbach (00:18:26):
It's noise, it's view, it's privacy, at least in some parts of the world that concerns people. It's safety because if eventually one of these things falls out of the sky, that hurts.
Ben Taylor (00:18:43):
Thinking about other parts of the world, what are the countries that are pushing the boundary and what are the countries that still struggle with this? What is the global perspective on this problem?
Matthias Winkenbach (00:18:52):
The last mile challenge is obviously very different. If you look at, let's say, a Western mega city like let's say New York City or Paris compared to another mega city, let's say in India, like Bangalore is a city that we worked on quite a bit or large parts of Latin America are facing very different last mile challenges. So while I think North America, Europe, parts of China are seeing a highly technology driven approach to last mile logistics where we, as we just discussed, where we talk about potentially using drones in the future, or maybe delivery robots or some sort of autonomous vehicle technology at least to, again, tweak the last mile and squeeze out cents and seconds and miles from last mile delivery routes in.
Matthias Winkenbach (00:19:48):
In parts of the emerging markets, we are still much more dealing with more fundamental problems like how do we actually overcome safety concerns? How do we overcome extreme levels of congestion during certain parts of the day? How do we respond to uncertainty related to weather? I'm not sure whether you've been to Sao Paulo, but for instance, we work quite a bit in Sao Paulo and during certain times of the year, the weather might change from one minute to the other and you might have a very heavy thunderstorm for half an hour that basically disrupts all your last mile operations completely.
Matthias Winkenbach (00:20:27):
So there's different challenges depending on also the state of development of the market that we are looking at. So I wouldn't say that there's one country that does it particularly well and one that does it particularly poorly. It very much depends on the types of challenges that you are facing. For instance, New York City doesn't have many of the challenges that I just mentioned for Bangalore or Sao Paulo, but is dealing with the problem of unavailable curbside space to legally park a delivery vehicle and things like that. So the challenges are very specific to individual cities rather than individual countries, I would say.
Ben Taylor (00:21:07):
You mentioned New York, one question I had, so we're thinking about deliveries, when I go to New York I'm amazed that I don't see lines of garbage trucks. So how does a mega city get trash out of the city without... That just confuses me. I think for that many people living in that small a region there should be nonstop garbage exiting.
Matthias Winkenbach (00:21:27):
That's true. I'm not an expert on garbage logistics, but my guess would be that the good thing about garbage is it's patient. It doesn't mind sitting there for a few more hours until you can actually pick it up during a time where there's not necessarily the rush hour happening at the same time. So in many cities that I know of, garbage collection for instance happens very early in the morning or during the night when there's not let's say, peak traffic for other purposes yet. So that is one big challenge that we are trying to overcome also for the delivery of goods, for instance. Can we actually decouple the inflow of goods into a city from other traffic conditions? So can we, for instance, deliver at night when everyone else is sleeping and certainly no one is commuting, typically? The challenge obviously is how do you do that because at night there's usually also nobody available to receive a shipment? So you have to come up with means of unattended delivery, which again is a technology question, but also a policy and regulation question that has to be solved. So there are a lot of initiatives thinking in that way, decoupling freight mobility demands from people mobility demand.
Ben Taylor (00:22:42):
And there's probably pressure from porch thieves as well to also have delivery that if I'm not available at my porch right now because I'm on a podcast, right now a porch thief can follow the delivery truck and go grab that delivery, and that happens pretty regularly. So I imagine there's other pressure to have delivery systems where people don't have to be present, they can just authenticate.
Matthias Winkenbach (00:23:05):
Yeah. And we all know, at least in the U.S. and Europe, other places we've seen quickly, we saw networks of, for instance, parcel lockers emerge from various vendors, which are a way of, in this case, decoupling customer availability from the delivery process and being able to deliver safely, not to the porch, but at least to a place nearby the customer so that he or she can just come and pick up the package whenever they're available. That's, again, a good example for a solution that is highly effective in a somewhat safe and well developed environment, like many parts of the U.S., Europe and other places, but which would probably not be very successful in other parts of the world that for instance have severe crime and safety issues. So for instance, putting up a smart locker in certain parts of Sao Paulo is a pretty bad idea because that locker would be gone very soon and so this is not a solution that works for them. For instance, unattended delivery might be more effective in collaboration with, for instance, convenience stores, which are staffed 24/7 so there's always someone there to make sure the safety isn't an issue. So different environments require different solutions.
Ben Taylor (00:24:23):
Yeah. I can't imagine porch delivery would work very well in downtown San Francisco.
Speaker 4 (00:24:28):
No.
Ben Taylor (00:24:36):
So there's a lot of excitement around self-driving trucks impacting supply chain. I think was it Anheuser-Busch or there was a beer company in Colorado that delivered one of the first warehouse to warehouse deliveries using a fully autonomous truck. Is this something that you think will offer big efficiencies to the supply chain problem? Where are we going to get the biggest efficiency gains before the last mile?
Matthias Winkenbach (00:25:03):
Yeah. So with regards to autonomous vehicles or trucks, again, on the last mile itself, I'm a little bit skeptical for several reasons because last mile operations, especially in cities, are probably the most complex environment to ever operate an autonomous vehicle in. So I would agree with you that at least the short to medium term potential is more on let's call it the middle mile. So for instance, between two warehouses or between two fulfillment centers, because obviously here distances are longer but still the operational environments, so the routes that these vehicles have to travel on, are at least to a large extent much less complex than an urban world network. And I think this is also one of the main areas of efficiency improvement when it comes to the stage before the last mile.
Matthias Winkenbach (00:25:56):
So think of an eCommerce platform. I don't want to name companies here, but think of a big eCommerce platform and you're ordering from them and it's likely that you're ordering multiple things and not everything is available in the same fulfillment center. Now, they have two choices. Either they send you two separate shipments from two separate fulfillment centers, which is costly and also environmentally questionable, or they find a way of basically transshipping inventory between one fulfillment center and the other so that they can actually send stuff to you in one consolidated delivery.
Matthias Winkenbach (00:26:32):
Now, to make that transhipment between facilities more efficient and also more flexible, because it needs to happen not just once a day in the morning but maybe more frequently throughout the day, doing that with traditional technology with a big truck and a human driver is costly and slow and not very flexible. So that's where technology like maybe autonomous vehicles that might be smaller in footprint but operate 24/7 at any time you need them or even larger cargo drones... We were talking about drones earlier, larger cargo drones making that connection between two facilities of the same eCommerce platform. Those types of technology could add the necessary flexibility to the supply chain or in this case, to the distribution network to allow the eCommerce platform to consolidate more orders to have less split orders and be able to serve their customers in a more cost efficient and also in a more reliable way and at the same time, safe on emissions, which is becoming increasingly important.
Matthias Winkenbach (00:27:39):
So long answer to a short question, I think that middle mile really is quite interesting for many of these, let's say, advanced technology solutions, be it autonomous vehicles or drones or the like to.
Ben Taylor (00:27:58):
So one of the things that I think is really interesting is I think for most people, if we were able to follow the full supply chain from birth to death, I think it'd be pretty depressing. I've got a nine-year-old son, he orders a Nerf gun, and most likely that Nerf gun was built in China, went on a cargo ship, and if you look at the full cycle and then once the gun breaks in four months, goes in the recycle bin and then that ends up probably on a cargo ship going back similar to where it came from. I'm curious, do you think we'll see a reality where there is no Nerf gun that has to travel around the world, where I can just 3D print that at home? Wouldn't that be the ideal thing for certain types of items and toys where 3D printing becomes much more common where we don't need same day delivery for most of these items? Or is that too futuristic?
Matthias Winkenbach (00:28:57):
I mean, it is relatively futuristic, but it's an interesting thought. I'm not sure whether 3D printing would necessarily help us avoid shipping stuff around the world, per se, because even if you have 3D printing that toy in the U.S., probably the raw material for the 3D printer would still be coming from China and the broken toy would probably still go back to where it came from. But I think the 3D printing and, let's just call it 3D printing or additive manufacturing, or however you want to call it, is probably a good way of actually reducing waste.
Matthias Winkenbach (00:29:34):
Because if you think about the fashion industry, for instance, a lot of the waste that we see coming, like product waste, is coming from the fact that they have to early on customize products. If you get the latest running shoe, that size of shoe, that color of shoe that you really want was already produced in that way in China, and then shipped over the ocean and maybe they're lucky and you actually want to buy it, but quite frequently, nobody wants to buy it. And then at the end of the season, it gets thrown away without ever having had any purposeful use. That's where I think... Well, I know that a couple of sports companies, fashion companies are actually experimenting with additive manufacturing as a way to late customize the products and basically to localize at least the final assembly of the product such that they only or more closely match demand with supply. So they only produce what the customer really already said he or she would buy and that would hopefully help us reduce quite a lot of ways, and fashion is just one example. I think there's many other examples that work in a similar way.
Matthias Winkenbach (00:30:48):
Another big potential of additive manufacturing might indeed be the recycling of raw materials. I know that a few years ago, Adidas, for instance, had a first concept presented where they've built the first running shoe that was entirely made out of additive manufacturing materials. So they basically "3D printed" the shoe, and that material was also designed in such a way that you could recycle a large part of it and basically 3D print the next shoe out of 45, 50, 60%, I don't know, of the shoe you were wearing before, simply speaking. So I think that's a great idea to have that new way of localizing final manufacturing of certain products. And at the same time engineering the materials of certain products such that they can be recycled and can be reused in additive manufacturing settings more than just once or twice or three times, but several cycles.
Ben Taylor (00:31:52):
That's amazing to imagine that you could recycle a shoe. I really like the point you're making before that fashion is very complicated because it's not just about the size, it's about the look as well, and so if you can customize some of the look and the color later in the stage and you're not going to overproduce... Because it's impossible to anticipate all demand.
Speaker 4 (00:32:11):
Now.
Ben Taylor (00:32:19):
So Matthias, I have a darker sense of humor sometimes so prepare yourself. So I was laughing thinking about, "Well maybe in the future we won't need any supply chain or it'll be very minimal because we'll just be in the metaverse or VR and you'll just need your delivered. So you don't need to have a puppy, you don't need to have this Nerf gun because you'll just have it in the metaverse." Do you think we'll ever get there or do you think we'll always need to see the sun and get out-
Matthias Winkenbach (00:32:46):
I sincerely hope not. Maybe I'm already too old for this, but I still hope that most people get joy and happiness out of living in the real world and consuming real products, however harmful that might be to the planet, but not sure whether the metaverse is any less harmful.
Ben Taylor (00:33:08):
Maybe the line in the sand to be crossed is in the future, if I reach out to you and say, "Hey, instead of going to Iceland and doing a motorcycle trip with my friends, I just did it in the metaverse." And as soon as people start doing it instead then that'll be going down that slippery slope where I think you and I would agree that no, you should go to Iceland and you should go on the motorcycle tour or you should go to Hawaii and you should go scuba diving. You should not discount that just because you can. So it's an interesting thing to think.
Matthias Winkenbach (00:33:38):
Yeah. I agree with you obviously you would think, "Well, not going to Iceland is probably good because you don't have to fly there and whatnot." I think a more likely scenario of something like, I don't know the metaverse or let's just call it augmented or virtual reality in general, a more likely scenario of that actually being helpful in streamlining supply chains and also reducing its environmental footprint, for instance, is changing the way we experience product and where we experience products.
Matthias Winkenbach (00:34:11):
So think of buying furniture these days, I guess we all know the big blue furniture store with the yellow letters and if you go there today, that basically means you have to take your car, drive for several miles outside of the city to go to the big box furniture store and experience the live product right there and then you probably buy it and you have to somehow organize how to transport it back to your apartment and all that kind of stuff. So there's a lot of effort, a lot of inconvenience to you as a consumer, but there's also a lot of transportation of individual items required of you going there, taking stuff back. So the environmental footprint of this entire process is not negligible.
Matthias Winkenbach (00:34:53):
While in the future, you may actually save that trip. You may be able to go to a local little furniture showroom where you're still able to experience certain things physically like the garment of your sofa or whatever, but you actually use augmented reality or related technologies to understand in which different ways colors, shapes and forms does this sofa actually come? How does that fit into my living room? How can I combine this with other products? For which you currently need to go to that big centralized facility, and now you can actually decentralize and provide the same experience to the customer in a little store or a little showroom two blocks down the road.
Matthias Winkenbach (00:35:36):
So you save a lot of that travel of the customer, you make it more convenient to the customer and you're able to consolidate and ideally make the delivery of the product that the customer chooses to buy more efficient because it's not the individual customer taking a software home in a Sprinter van anymore, but you can actually then basically have scheduled delivery routes of consolidated trips where you serve multiple customers with the same vehicle and it obviously reduces the miles driven and the emissions created per so far in this case. So I think there are ways of using that type of virtualization of our world to some extent to reduce the carbon footprint of supply chains and the way we consume.
Ben Taylor (00:36:22):
I really like this topic, the topic of changing buyer behavior. The other thought I had was you could also bring that into the home. So Amazon already has, I think on their app you can do augmented reality, some objects you can visualize it in your home. So you can imagine a scenario where you have augmented reality and you're walking through your home and furniture's being recommended and projected in the setting almost like you have a digital interior designer that is constantly trying to redo your home from a furniture perspective or design perspective, but then you can visualize it right there. You don't even have to go to the brick and mortar shop.
Ben Taylor (00:37:02):
Another thought I had is I feel like with Amazon going to same day delivery or some of these other logistic suppliers, is that going to be the death nail in brick and mortar? Because right now for me, if I have to decide, am I going to go to a particular brick and mortar shop, it's normally based on time, that I need it today, I need it soon. This year, I had my first same day delivery where I ordered it at 3:00 PM and it was delivered that evening. So do you think brick and mortar stores will still be around with some of these efficiency gains?
Matthias Winkenbach (00:37:37):
Yeah. The interesting thing is let's say for several years, people basically predicted that brick and mortar retail was going to die and actually it's specifically services like same day and instant delivery that make me believe that brick and mortar has a strong future, but a different one than people thought it would.
Matthias Winkenbach (00:38:02):
So to give you an example, we were talking about those satellite facilities before that companies would use to actually fulfill same day or instant delivery orders from that they probably received online. I believe we're not going to be able to reverse the trend that more and more things are going to be bought online rather than physically, but the retail store, the brick and mortar retail store itself is actually a strategic asset to enable faster, more flexible, more reliable delivery services for the online channel. So we will probably see that the nature of the retail store changes.
Matthias Winkenbach (00:38:44):
The footprint of the retail store may no longer be primarily dedicated to walk-in customers actually buying things on the spot, but it will probably be turned into more of a showroom kind of an experience type of setting where people may go in, see a product, they may take it with them, but they may also just place an order for it to be delivered to their door a little later in the day. But then the remaining part of the retail store would actually serve as a local fulfillment or micro fulfillment hub where there's a lot of inventory in the back that can be used to serve urgent orders that came in online or that customers placed that were actually using the showroom part of the store.
Matthias Winkenbach (00:39:30):
We see that already for some retailers that it's happening, and I think that trend will continue because to enable these delivery speeds, eCommerce platforms, retailers who want to offer them need to identify suitable real estate and prime urban locations and typically that's where the retail stores are today. They were built close to where people lived or where people worked. So they are in the locations that we need now more for the logistics behind eCommerce delivery, but still basically can be used by the companies who own and operate them.
Ben Taylor (00:40:12):
I'm very interested in extremes and in limits. When you think of mega cities, is there an upper limit? Is there a limit where logistically it's too complicated, even the best systems break down or do you feel like people could continue to push the limit on mega cities? What is the largest mega city? I don't know, off the top of my head, but how big can we go? Could we have a hundred million people in a city?
Matthias Winkenbach (00:40:41):
We can. I mean, technically there is no limit to it. And especially if we look at mega cities as they pop up left, right and center in large parts of Asia, for instance, these are cities that didn't just historically grow this large but these in part at least are cities that were planned from the beginning to be able to become this large. Because let's say many traditional cities, think of Mexico city and others, they have grown humongously large, but nobody, when Mexico city was founded, thought of it that way. So that's where many of the struggles come from because for instance, the transportation infrastructure, the road network was never meant to accommodate that many people. And that's why we typically say that the bigger the city becomes, the more dense it becomes, but the road network and other transportation infrastructure doesn't grow proportionally and therefore it becomes congested and hard to operate in.
Matthias Winkenbach (00:41:41):
And for these historically growing cities, I would say there is some sort of natural limit to their growth simply because they can't fit more people in the same space and at some point the transportation infrastructure would entirely collapse at least theoretically. But if you're thinking about these more planned cities in China and other places, I don't really see a reason why a hundred million or a 250-people city wouldn't be at least thinkable, as long as the plan was initially to have it grow this big and you have the necessary transportation infrastructure in place to support this growth because that's where the reality currently limits most of the existing mega cities from efficiently growing further.
Ben Taylor (00:42:26):
How do we plan these mega cities better compared to something that just grew out of nothing like Mexico City?
Matthias Winkenbach (00:42:33):
Yeah. Most of it obviously relies to balancing the needs for both people and goods mobility. In most traditional cities that I know of, both of these needs are being met by pretty much the same infrastructure, namely roads. While in a well-planned mega city, I would say, why would you even plan for freight to go on roads? Because freight is plannable, you know pretty much exactly where freight needs to go. You don't know how much and when and whatnot but you know pretty much exactly the possible destinations of where freight might want to go. So why wouldn't you, for instance, already establish an underground derailed system for freight that, I don't know, has the ability to feed into every single apartment building or whatever to actually deliver to individual households while being at least predominantly underground and both invisible to the people and therefore making it nicer to live in the city, but also less congested or less disruptive to the remaining traffic flow of people that may happen above the ground.
Ben Taylor (00:43:40):
That's fascinating because normally when we think of underground transport we think of the subway, but you're suggesting that you could even have some type of underground conveyor system where you have micro fulfillment centers at the skyscrapers that then distribute where it then the highways are less of a concern going into a mega city. That's really interesting.
Matthias Winkenbach (00:44:01):
Yeah. Obviously it depends on geological conditions and whatnot, but generally speaking, if you have the money and if you have the ability to plan it from scratch, why wouldn't you plan with an underground system? Because also keep in mind, we were talking about autonomous vehicle technology before, and I was telling you that autonomous vehicles in a real world network with human drivers around is going to be difficult to accomplish. But if you have autonomous freight vehicles in an isolated system, like a tunnel system where there's no random interference, that is very much doable. So I think if, one, we're planning to build an entirely new city, one should leverage that fact.
Ben Taylor (00:44:48):
That's really interesting. When you look online and you see the videos from the Amazon fulfillment centers, it's fascinating to see how quickly the packages flow through there, so you can imagine there's smart cities that are being built right now. I know in Saudi Arabia they have Neom that's being built where they are planning for the future. So you could imagine maybe a big supplier like Amazon could potentially build something like that sooner to demonstrate the feasibility or the possibility of servicing packages for 50 million.
Matthias Winkenbach (00:45:20):
Yeah. The challenge there is obviously if someone like Amazon or some other single vendor was establishing this, it would most likely be an infrastructure that would be closed, useful for Amazon, but not for other vendors for instance. That's a challenge because as soon as you have a private infrastructure, that doesn't solve the overarching problem. So you would ideally have to have this infrastructure open to different vendors using it, which also requires for instance, standardized containers and that kind of stuff. So it's not easy, but if you were doing something like this from scratch you would have to think of it on a systems level, not on an individual vendor level.
Ben Taylor (00:46:03):
That makes sense. Already I think some people would argue that, again, not to name names, but the power of some of these suppliers become significant because they begin to control airfare, trucks. There's a lot of innovation that comes that makes it difficult to compete. I guess the other thought is you have Elon Musk's, the boring company concepts where maybe we'll just have all of these tunnels under cities in existing cities that are used for package delivery rather than just traffic.
Speaker 4 (00:46:45):
No.
Ben Taylor (00:46:46):
From a technology perspective, what is the most challenging to you now? What are the technology aspects of this that push you the most when you think about the hurdles or the breakthroughs that we're running?
Matthias Winkenbach (00:47:01):
It depends a bit on what you define as being technology, but from a pure research lens, I look at the work that we currently do. Let's say boundaries are not so much being pushed anymore on let's say finding more efficient algorithms to optimize a route, but boundaries are being pushed on finding new ways to make use of data, like I discussed before, like how do we actually learn a route rather than optimize the route? How do we actually learn from the living system on what solution, what route plan, what distribution strategy works in certain parts of a city and doesn't work in certain parts of the city? In theory, or conceptually, it's easy to talk about this like we did before, but actually methodologically it's far from trivial for various reasons. A, because at least in our field, many of let's say the applications of many machine learning methods haven't really been explored in depth or not deeply enough at least.
Matthias Winkenbach (00:48:14):
Even in the academic literature, there's still somewhat of a lack of experience, but even on the data side, we think that today we are living in a world of abundant data availability and big data, or however you want to call it. But actually, if you think about the complexity of the problems that we're trying to solve, most of the times, we still don't have enough data.
Matthias Winkenbach (00:48:36):
So think of a problem that we discussed before, an eCommerce platform wanting to offer an on-demand delivery service. So you'll buy something now and you get it delivered within 60 minutes or something. What you actually need to be able to do to pull this off reliably and somewhat cost efficiently is predict demand, which sounds boring because we've been predicting demand for decades, but we've always been predicting demand let's say on a high level of aggregation, predicting demand for an entire region or an entire city or an entire day or an entire month. But what we now need to do is we need to predict demand let's say within the next 60 minutes at the spatial resolution of a block, of a city, or a zip code. And then your seemingly big data becomes relatively small if you think about how many observations you actually have to learn from, and it's not that many, even for big companies who have years worth of data, it's not that many, or at least not enough to train the types of methods that are very powerful, but typically also very data hungry to become powerful. That's in a nutshell some of the methodological barriers that we are pushing, so research technology, if you wish.
Matthias Winkenbach (00:49:52):
And then thinking of technology as hardware, obviously the usual culprits are, are we ever going to be able to make advanced vehicle technologies like autonomous vehicles, drones, delivery robots viable from a technology point of view, acceptable from a social and regulatory point of view. But most importantly, are we going to be able to make them economically viable? Because it's actually, as I said before, not that easy to match the cost of delivering a package with a big brown van by a human driver with an autonomous system. It's not that easy to beat that cost.
Ben Taylor (00:50:32):
You make me think of a new category that I'd never thought of before, and that is this concept of, for the last mile, you're anticipating additional demand so you're actually putting packages on the last mile truck that are not tied to a customer yet and now you're dealing with a dynamic pricing problem. So if something's in high demand, I think I'm going to sell it but then in real time, I'm adjusting the price to ensure I sell it during the last mile, which is really interesting. It sounds very complicated, but it would be fascinating to have last mile trucks that are anticipating demand. And now when I order something it's to my house in an hour, which would confuse me because I know the fulfillment center is an hour away and that doesn't make any sense to me, why I would get a package in an hour, but we would know it was anticipated demand in your area.
Matthias Winkenbach (00:51:25):
It's actually an interesting idea. We had a research project a couple of years ago where we thought, well, what if we actually didn't only have physical stores for, let's say, instant delivery services to hold the inventory, but what if we also inserted two or three big trucks into the city that would just keep moving and try to be close to where we expect the amount to happen within the next hour or so and then basically use them as a popup fulfillment center pretty much to serve those customers more instantaneously? It turned out to be more complicated than it sounded and also harder to make work in a cost efficient manner, as you can imagine, but it's probably a concept that we might be seeing in the future.
Ben Taylor (00:52:07):
Maybe you just need to combine that with a gaming company. So I'll be playing Candy Crush or something, and it'll say you won some free eggs and then eggs will be delivered to my house, but you'll be paid with advertising dollars.
Ben Taylor (00:52:23):
So I know Amazon has the famous patent on the blimp with the drones coming out, so you have a flying fulfillment center. Is this something that you think could be feasible or, again, does it depend on the location where maybe in a rural setting that can make a lot of sense but in an urban setting... Or maybe I have the concept flipped where it'd make more sense in an urban setting.
Matthias Winkenbach (00:52:49):
Let's say I'm not sure whether that absolutely has to be a blimp because that in itself is relatively advanced technology that comes with a lot of challenges and costs, but I guess the underlying idea isn't that stupid. So having a fulfillment facility that is not huge, so that can't store everything but that may be able to store the most important or the most urgently needed items and basically fulfill these orders or fulfill these items on urgent order more flexibly because it's closer to where the amount might happen. So that might obviously be the blimp flying over Manhattan, but it could also just be a barge on the Hudson river, which is probably much easier to pull off from a technological point of view, but almost would serve the same purpose because the important part is that it's mobile, that you can move it around to where you assume the amount is going to happen within the next hour or two.
Matthias Winkenbach (00:53:51):
And the reason why you just wouldn't put it in a physical facility in downtown Manhattan is probably cost and availability of such facilities. It's not that easy to find suitable real estate for such an operation. And if you do find it, it's probably extremely costly. So that's why you would probably be willing to put stuff in the air or on the sea or in some other shape or form. So the blimp concept itself, I have my doubts also because it's very sensitive to weather and whatnot so I think there's a lot of question marks around that, but the ability to have a mobile fulfillment facility is very valid in whichever form it might come.
Ben Taylor (00:54:30):
With algorithms and with human behavior, especially on these social media apps, TikTok, Instagram, things like that, but also with Amazon and double click, we get a very good sense of people's personas online. When it comes to product recommendation, this is getting better and better. Do you think we will see a demand... I have seen this already, there are services out there that would deliver you a monthly surprise package and it has goodies. It has things inside that they think you would like and they've got some process for you to return stuff. Do you think we could get to a future where part of logistics and delivery is anticipated demand where I know you need a toothbrush on this cadence, I know you need these things, where there's a lot more planning so I can discount the pricing but you and I now get deliveries every Friday and it's for things we did not ask for but the algorithms are better and better at anticipating what we want?
Matthias Winkenbach (00:55:29):
Yeah. I think that's certainly part of what we're going to see in the near term in a way, we are already seeing it. You can already set up a recurring delivery schedule for certain pantry items and whatnot. I believe that for instance, when it comes to grocery deliveries, something that also more and more people actually do online, order groceries online, it's relatively easy I would say for someone offering that type of service to learn from your consumption patterns to anticipate when you might be running out of milk and therefore basically already triggering you to place an order early such that, A, they don't get the order when you actually urgently need it but they have a little bit more of a lead time to plan for it and to consolidate and to make this both more cost efficient for them, but also less disruptive in terms of the ecological footprint of this delivery operation.
Matthias Winkenbach (00:56:25):
So I personally do see a future where you may automatically get your grocery replenishment delivery, I don't know, every Friday afternoon after you return from home without even pushing a button and maybe the only thing that you can still want to do is add items that you typically don't order but want to be included in this particular shipment so that you at least make the bulk of those deliveries more plannable, more anticipated that's by the vendor and therefore streamlining their logistics processes, not just on the last mile, but also inbound. So if they know what they have to ship out of the supermarket or wherever they ship it from on a Friday, they already know what needs to come in on a Thursday night basically from somewhere out of the city. So I think there's a huge potential for streamlining, especially the logistics of perishables, again, reducing cost, reducing emissions, but also reducing waste by using the data that we generate voluntarily or involuntarily by shopping online.
Speaker 4 (00:57:31):
No.
Ben Taylor (00:57:39):
One of the other questions I had for you, I mention there's overlap in your research with other domains, so ambulance, placement optimization, first responder, stuff like that, where it's a similar problem, they're not delivering packages, they're delivering services but they're having to deal with logistics, what's the time of day, what's the potential demand, which is not demand that you and I like to think about, but it's risk, whether there's different events and things. Do you see overlap in your research where you see publications or researchers that do try to focus on these areas or are they borrowing from your space when it comes to ambulance placement?
Matthias Winkenbach (00:58:20):
The domains are obviously very different, ambulance services and last mile delivery of eCommerce packages have very little in common from a domain point of view, but from a methodology point of view they're extremely close. I even used this example in my lecture at MIT because different approaches of thinking about routing or network design depend on what your objective is. Let's say, if you distribute eCommerce shipments, then your objective is typically to minimize your average cost for delivery. While if you are figuring out where to put your ambulance, your emergency response station in a city, your objective might be to place them in such a way that you minimize the risk of taking more than 15 minutes to get to any random location in the city with your ambulance. Obviously you're still trying to minimize costs, but that's the hard constraint, you want to be everywhere or anywhere you need to go within 15 minutes. But methodologically, these two problems, eCommerce distribution, network design and routing are extremely close to these other areas. And there is a lot of exchange between researchers that may work on different real world applications but use very closely related methods to solve them.
Ben Taylor (00:59:37):
One of the last questions I have for you is, are we to the point now where we are able to do a midday route re-optimizations or do we just do route optimization at the beginning of the day when it comes to last mile delivery? Is anyone doing midday adjustments based on traffic or other things or real time?
Matthias Winkenbach (00:59:56):
I'm sure people are doing this. The challenge obviously is, how do you get reliable, real time data that you could actually use to make these re-optimizations? Another challenge is that actually depending on how complex your route is re-optimizing it might be quite computationally costly, so you can't do it within a second. And I would say it's still relatively rarely done in practice simply because it's only really necessary to do this if you are predominantly carrying highly time sensitive deliveries. So if it really matters whether you lose the five minutes in traffic or not. So think of delivery services that exclusively deal with extremely tight delivery time windows like an express courier service or something. For them, that is really important to keep the intro-
Matthias Winkenbach (01:00:46):
Day dynamics of traffic and weather and whatnot in check. For, let's say, the standard parts and delivery service of today, it's probably not that relevant yet. But as we were speaking before, since more and more people want things faster, more flexibly, more tied to their own availability I think this becomes more relevant.
Matthias Winkenbach (01:01:07):
Actually, I have an interesting connection here to a topic that we discussed before, namely drones. So one area of research that we are quite interested in right now is using drones to replenish delivery vehicles intra-day. So imagine you have a delivery vehicle moving around the city, which usually starts leaving the depot at, I don't know, 8:00 AM or so, basically what that means is that any package that hasn't reached the depot before 8:00 AM in the morning won't get delivered that day.
Matthias Winkenbach (01:01:39):
Now, you may, for various reasons have late comers or just parcels that are super urgent but haven't been placed until 8:00 AM yet, so they arrive at the depot at 10 or at noon or whenever, and you still want to get them out there, you still want to get them delivered, but you don't want to send an individual courier vehicle for a single package. So you might want to use a drone to pick up that package, fly to the delivery vehicle that's already on route since the morning, and basically replenish that vehicle with additional packages to deliver. And at that point, you obviously need the ability to dynamically re-optimize because you're basically inserting new delivery stops into the route while the route is already being executed. And that's also futuristic, but not unlikely to happen.
Ben Taylor (01:02:30):
That's really interesting. I've really enjoyed talking to you. This is fascinating. It's interesting because this is a problem that impacts everyone, literally everyone, all of my neighbors. This impacts everyone.
Matthias Winkenbach (01:02:40):
But a lot of people don't realize how difficult it is to get that package to their door step.
Ben Taylor (01:02:45):
They don't appreciate it, but they do get mad when it's not there the next day when they said it was going to be.
Ben Taylor (01:02:53):
Well, this has been a pleasure. I really, really appreciate having you on and appreciate you giving us your time and your expertise.
Matthias Winkenbach (01:02:59):
Sure. Thanks for the opportunity.
Speaker 2 (01:03:04):
Thank you for joining us on this More Intelligent Tomorrow journey. Discover more and join the conversation at moreintelligent.ai. The future is closer than we think.