In this episode of OnPoint with Korem, Christophe Charpentier, the director for transportation networks and delivery at Wayfair, the online retailer, discusses the challenges faced by organizations capturing immense amounts of data and in particular for retailers like Wayfair who are facing major obstacles with supply chain bottlenecks. Mr. Charpentier has also had an interesting career with stops at Esri and Amazon, so you won’t want to miss this episode of OnPoint.
Joe Francica: So, thanks very much again for doing this. You’ve had this rather interesting career, you’ve gone from Esri to Amazon, now Wayfair. You see the industry perspective kind of from both sides of the fence and maybe we’ll just start there, and if you’ve got any advice for those who are looking to go from the vendor side to the user side or vice versa. But again, curious as to how you started and maybe how you ended up in Wayfair.
Christophe Charpentier: Yeah, I think that, to me, the key is… one thing I’ve learned through the journey is learning to look at the problem through different perspectives. I think when you spend a lot of time in any industry and with any company, you tend to not see that you have biases of your own. And you think there is one way of solving a problem. What hit me the most was when I joined Amazon, how I joined an organization and besides me, there was only one single person who actually had some form of experience in the GIS space. And despite that, we’ve built a complete geospatial platform that now powers all of the routes of Amazon. From that lens, and certainly, I’ve uncovered that some people without any bias, without any reference points, can also solve problems in a very open-minded way without taking incarceration, any of the conventions, or any of the status quo that is operating in any given industry.
So, we know what was the very first learning. The second learning was: to be customer-obsessed. Always try to identify who your customer is, and your customer could be an internal customer. My customers today at Wayfair as I lead products for the Wayfair delivery network — so, the operator, the carrier that moves items, cartons that contained the products that you buy on the wayfair.com platform —, my customers, direct customers are operators, they are field associates, they are drivers, they are managers in warehouses.
They say you’ll gain to be very successful if you listen to your customers and work backward from the knees of these customers. And whether these are consumers, whether they’re internal customers, it doesn’t matter. Just work backward from these customers rather than having a very firm point of view of what is right and trying to force-feed them with that solution. You are seeing a lot more adoption, a lot more traction, and then a lot more… you created a flywheel where people are actually contributing to the improvement of the solution, the product, or the technology you’re building, and you’re going to apply this to geospatial or any other industry.
And to me, that’s really the second biggest point. And if I had to pick a third one, I would say, don’t do this for the science, do this to solve problems. Don’t do something because ”yeah, well, it could be cool if we could do this, this well, that way.” Well, is it really solving a problem? Is it really helping whatever the situation is? I think the key here is really to solve problems and be creative in that way.
And also related to this, I’ve seen many times over the temptation to do something extremely slick. Well, sometimes, something crappy is actually good enough. You’re going to spend 1/10th of the amount of resources on it, but you’re still going to solve 98% of the problem. So, you got two options: either I come to offer a solution that’ll solve 98% of a problem right away, or it’s going to take me a year to actually attempt solving a hundred percent, if even a hundred percent. Learning to be scrappy and accept the trade-offs between perfection and actually true problem resolution is also something that moves on the industry side, I’ve learned very vividly, both at Amazon and continued to experience that as well.
JF: So, let me pick up a gun on a couple of things you said. I’m a little surprised that Amazon hadn’t had any other geospatial, I don’t know, whether you want to say experts or at least experience, because I’ve seen in companies, I’m sure you have too, that there are some companies that are very, very expert at using geospatial technology. And maybe you’ll tell me that Amazon already was, but maybe didn’t have quite the focus that they needed to. How would you react to that?
CC: I would say that the point is we tend to… there were people who have worked with geospatial data and geo-related topics and problems before I joined. None of these people or very, very, very few of these people were GIS professionals. And so, I think there is not a direct correlation saying that it has to be a GIS professional, and you have to come from the GIS industry to solve geospatial problems.
But when I joined Amazon, I was coming from the GIS world and I got very surprised to realize that there was actually a team that had worked on generation one off something. And none of these people, to one exception, was actually somebody who had actually worked on GIS per se. Someone who had worked on either software, MapInfo, or equivalent. Only one person had had that experience. Everyone else was engineers, were product managers that were just listening to the problems that our internal customers had, but they were geospatial professionals, they were not GIS professionals.
JF: Very much to your point where there may be no need for the geospatial experts so much as the need to solve the problem and employ people that know how to come at a problem in many different ways. I’ve heard it expressed maybe slightly differently in that, now location-based data is pretty commonplace and it’s another data type and it’s not as specialists think. Maybe your former boss, Mr. Dangermond would think differently about what spatial is and how special it is, but maybe not so for other companies. And I wonder how the data is perceived in an Amazon or in a Wayfair?
CC: It might be special because it is clear. There is a dimension that is not well understood besides the high-level concepts. But so far, my experience has been that people were actually simply smart, if someone is going to give them some level of understanding of the complexity of the geospatial data, from this point forward, engineers and analysts will understand the geospatial data. Now, granted in the retail space, we’re not solving problems like slopes and our running water and/or earthquake.
What I’m saying is likely not true with domains in which I require some domain expertise at a high level and is really very intimately geospatial and complex. What I’m saying is that in the retail world, you can actually achieve a lot of this with people that are going to invest a little time in getting educated but then we’ll apply some methods. When you deal with geo-coding, well, geocoding is complex, but it’s not rocket science either. And you need some understanding and you realize also that there is not just one version of what is said right from a geocoding standpoint. If you’re a retailer like I am today, my understanding of the right geocode for an address is very different from what is the right geocode for public safety, and these days I’m dealing more with large items, lost parcels for couches, vanity,…
When I was with Amazon, it was more of dealing with small parcels. So, when you drop off on the porch of your house, or if you drop off in the front of the garage, wherever the customer wants it, it’s where it is right. So, is the geocode the point that eases the location of the gate on the limits between the street and your property? Is it the door of your house? Is it the middle of your garage door? All these elements are important. Does that require very specialized expertise in GIS to resolve that? I don’t think so. It requires an understanding of mathematics because when you want to model, for example, where is the right location. Each time you do delivery, you collect a GPS point about 1, 2, 3 main street, and when Joe is going to order again, I’m going to collect a second GPS point and a third one, a fourth one, then they can learn exactly where is Joe’s address. Even if I don’t have intelligence that tells me 1, 2, 3, there is the house number, main is the street name, street is a street type, and so on and so forth.
JF: You’re bringing up an interesting problem. And today we’re sort of at the crux of the situation with the supply chain you’ve probably experienced it both at Amazon and now at Wayfair. Is it necessary to fall back on geospatial technology to solve some of the supply chain issues, whether that’s simple geocoding or whether it’s understanding the value of mid-mile, last mile, tour planning? Where do we sit? How important is geospatial in this industry?
CC: It’s fairly critical in the notion of… because everything is about location and movement. So, it’s critical in many different aspects. I think if you think about an e-commerce platform, there are basically two components: one is the storefront, and the other one is a supply chain with delivery to the customer. If you own your own supply chain, you want to optimize the movement of the goods from the moment it hits the land at the port of entry to the moment you deliver them to your house.
And we’ll go through the warehouse of the manufacturer or the seller, most likely through a fulfillment center of yours, then we’ll go into a pull point, a crosstalk, many of them maybe, and finally the delivery station and go through the house of the customer. When you do all of this, the first stage where geospatial is critical is having a good understanding of the graph that you’re managing, what are your options to move from point of origin to the final point of this nation? And what are your options in terms of the number of lanes that you have, how frequently you run them, and how accurate they are? So, it is geospatial in nature, but fundamentally it’s a pure graph. It’s math. That’s all it is because these are points and you basically look at the performance of your trucks moving from one location to the next.
Do you need the map for that? Not really. Do you need the intelligence of a graph that has interconnections and relationships? The answer is yes. When you come to the last mile is probably where mapping is probably the most pressing. Why? Because then you’re going to organize your route and your route is going to be the sequence of the different first stops in what is the most effective way of delivering all of these packages in one day for a driver.
And so, mapping plays a more pressing need because you’re going to build a route that is going to take into consideration actual physical distance, the real distance between two points, the speeds are either the real-time traffic or the model predicted traffic, and all the breaks and service time you need to do at each location. And the service time is not just simply to say for a couch, I need 20 minutes. No, actually, if I deliver in downtown New York, I first need to park, second, I’m going to have to find a way to get through the destination and go all the way through the 17th floor, deliver that couch, install it, and then move back.
All of these elements have detailed mapping solutions. You need to provide a turn-by-turn, but you need to have detailed and efficient route planning. You need geospatial technology to do that. You then need a turn-by-turn experience for the driver in order to get from each destination to the next. And then you will need also geospatial information to communicate with a customer. For example, at Wayfair, we do give you a call 30 minutes ahead of us arriving at your house. Knowing exactly where the driver is in the process, recommending the driver who actually plays the call now to Joe, and then calling Joe is based on knowing that I’m 30 minutes away from that delivery point. And this is very intimately geospatial in nature.
JF: Yeah, there are a couple of things in there. Right at the intersection of now delivery and supply chain, there is this demand probably for more geospatial data. And we are kind of at the whim of a few data providers at least street network providers. Are you finding that you are being pressed into collecting data yourself from drivers directly? Or do we need more data from whatever supplier it wishes to provide you with the data that you need as a retailer to get from point A to point B?
CC: The challenge with commercial data — I’m trying to be sensitive on what I know is public knowledge or not public knowledge, sorry —. Here’s what I can tell you. Commercial data are very rich and, in some cases, really sufficient. If I compare and contrast large parcels versus small parcels, the number of delivery points per day is very different because you are either more of a white glove delivery. So, you have about 20 stops in a given day. When you deliver a small parcel, you’re more like in the range of 120 to 240 stops. So, each deviation, especially in a small parcel, to the plan is going to cause a problem.
The one thing that customers tend to do when they move into a new house is ordering a product. As soon as you have a new development and then to get them delivered is almost a day, they really place orders that two days before moving into their house and then they arrive on time when they are actually moving in. If you’re talking about new developments, those are, in most cases, not mapped by commercial vendors, or if they claim they’ve mapped them, yes, they have the data, but there is a chain of third-party providers that are actually doing the data integration, make it available to vendors that are, you know, whoever has a routing tool, and then those vendors actually publish an update themselves with our software. And usually, between the moment an item is actually injected into a commercial database and the moment it is actually used for commercial systems, you have at least six to nine months, if not twelve.
So, when you run an operation that is very sensitive to time and performance of your drivers, you always want to have the most current network. So, what is key for you is, at the very least, to be able to supplement the database with whatever you observe. Because if you run an operation and if you were to collect the GPS point every so often — every X number of seconds, one second, five seconds —, and if you have a density, then very quickly you realize that you are covering a good chunk of any geography area you’re servicing almost entirely over a week.
When you do this, you may not have the same level of accuracy, but back to my point about being scrappy, I’d rather have my driver being lost just once, but then I’ve collected the points, and yes, my route is going to be like a bit of a zigzag, but that’s okay. At least I’m going to get in closer to the destination and through the second route, I’m going to be able to improve my network, and through the third one, and so on, so forth. To me, it’s not a dichotomy ”either/or”, I think both are useful. I think commercial vendors have done tremendous investment in work, especially in terms of added value attributes, turn restrictions, speed, average speed, or specific roads on which actually larger trucks can go.
Wayfair would deliver with 25 cubic feet vehicles versus the vans that Amazon is using. We are facing some restrictions that the Amazon vans don’t have to be slowed down by. But we do have some turn restrictions that we have to comply with. When we build a route, we need to secure the fact that all of these attributes are all there. Both solutions are valuable, but I think there are situations where it’s worth collecting your own data because then you have very strong return insurance of the time not wasted on the road.
JF: Looking out a few years, if you can put what you just said, kind of in context, you need more data. You need more accurate data or as much accuracy as we’ll allow you. We know we have supply chain problems. What can you see in the future in terms of more accurate delivery? Are we going to autonomous vehicles? Are we going to drones? You know, again, a lot of your business is going to be predicated on the fact that you want to deliver on time, every time. So are we looking at different delivery mechanisms and is that part of your purview to it to say: ”Hey, let’s think of how we’re going to do this in a couple of years.”
CC: Vehicles will improve, and I think assisted driving will actually grow. Will it grow to the point that is going to be pervasive on every mile, that we’re going to be in fully autonomous driving? From what I’ve seen, from the data I’ve had access to, from the conversation that I had… I spent part of my time at Amazon with Alexa Auto, so bring Alexa into your car. I spent like two and a half years knee-deep in the automotive industry. I don’t believe that it is in the next five to ten years that we’ll see pervasive, autonomous driving and that the trucks are going to go away, or the vans are going to go away completely.
Well, there’ll be moments where autonomous driving may actually help with the supply chain, possibly. I can envision, for example, our long holes, when you do cross-country, eye movements, that you can have a ‘’train of trucks’’ that are autonomous because it’s a well-defined, well-controlled environment. I can see this coming in the next five years. I’m more skeptical in terms of the capacity of any of these vehicles, large or small, to be fully autonomous to the point it’s going to come in front of your house and move to the next customer without a human in the vehicle.
Will there be more assisted driving and more security features? Absolutely. But I think we’re still a long way away from fully autonomous vehicles. I know there have been lots of investments in sort of super-high-definition, mob data. I think it’s worth it in some situations. And I had the privilege to have a ride in a BMW X7 and go into fully autonomous mode on the highway in Germany. That was extremely impressive. And once you do kind of, like you said, the navigation to a destination and it’s the car on its own is going to move from the middle lane to the right lane and then take the exit on its own. It is very impressive and I’ve seen it live in action. It was not fake. It was real, the driver didn’t have to be holding the wheel.
I believe that there will be moments where these happen. So, from a supply chain standpoint, it’s just probably these elements or middle mile, stem time, potentially, when you want to leave a delivery station and then you go to an area where you need to deliver very locally, I can see where potentially some of this will happen. I don’t think this is going to be the biggest transformation that we’ll observe in the next five years. I think electrification of the fleet is going to be the biggest transformation that we’ll face in the supply chain in the next few years.
It’s going to be led by Europe. In Europe, you know, many CDs already dictated that personal vehicles after 2025 and diesel, will not be allowed to enter the streets. There are limitations on vans, and by 2030, most vans will have to be electric. This will be transformative for us and we’ll need to be accounted for as we think about the supply chain. But I don’t think that autonomous driving is yet a force that is going to transform our industry.
JF: Let me ask you one final question and then kind of put it all together, which is: there is, obviously, a need for more data for going somewhat autonomous. Technically, you know, geospatial data kind of is complex to compute. Can we foresee the time, or can you foresee the time where a lot more geospatial technology is moved to cloud platforms? You worked at Amazon, there’s a lot of movement in that direction, cloud-native in particular. Is it possible? Can we move the type of geoprocessing to the cloud that we need to do the things that you’re talking about with processing, doing routing, tour planning, etc.
CC: I think, it’s important though when we think about this, I think the GIS industry, because of its evolution, has had a tendency to think in terms of dichotomy: mainframe move to servers, move to desktop, moving to the cloud. And each time there is something like this, my perception, now that I’ve left is, that if it’s thought about in contrast: cloud versus desktop, cloud versus mobile, desktop versus mobile. No, I think we need to think about it in terms of… Ultimately, everything should run on each platform or some of them should maybe only be running in the cloud, and some others on local machines.
But that’s one aspect where I think we need to think hybrid and what is the most effective in terms of the use case I have at the moment. I should not have to choose between a platform, but rather what is the best implementation for me? What would work best? And even when you think cloud, AWS has already made a move of actually bringing back compute on the edge. I’m blanking on the name of their product, but they have an edge computer that is actually fully integrated with the cloud but actually delegates the compute on the edge on some device. And that is not really a desktop, it’s not really a server, it’s really the cloud on the edge.
And so, when you combine all of these forces, do we need in the supply chain industry to compute more data? The answer is definitely yes. Are we going to compute like 10 times, a hundred times more data in the future than we are today? The answer is also yes. My vision of what we are going to be able to do at Wayfair is that actually, I want to be able to collect a lot of data, as many central as possible, so that we can actually improve the quality, the accuracy of what we do.
I don’t want to rely on even a human using a mobile terminal as a phone and say: ”I’ve arrived.” I want to be able to detect this. And this is standard today, when you use a mobile phone, you can know that somebody has stopped their engine and is now becoming a pedestrian. We will need to compute all that data in order to improve the way we think about the routes that we build and ultimately be more accurate so that the promises we’re making to the customers are more accurate.
To do all of this, do I need to push every single point of data I have on my device or on my edge in the van to the cloud? No, maybe I can actually do some pre-processing and actually only push what is the relevant data. And to me, that goes back to what some OEMs automakers are doing. There are thousands of sensors on your car. Today, they don’t push every single bit to the cloud. They push the bits that are relevant. If you are doing a heartbreak, then you’re going to capture all the different signals they had in the 10 seconds prior to the heartbreak and push it through a cloud so they can understand the context.
We need to think along those lines as well, and actually have a vision where some of the data is correct, wherever it is on each edge, mobile device, laptop, tablet, or IoT. And then you have intelligence that actually figures out all the data that are relevant today to my processes, then collect those, push them to the cloud, get some extra process done in a cloud, because maybe I will likely have more computing power, and then potentially, either improve my reference data I have on my servers or push some of that data that has been post-processed back to my edge. I think this is how we need to think about it, not in terms of cloud versus PC, but really kind of leveraging what is best in any of these worlds.
JF: Christophe, thanks very much. I think we’ll leave it there, but great discussion on some of the forward-looking, thinking on where we go, in particular in the supply chain, and just into your process in general. Thanks very much for your time. Really great discussion.
CC: All right, thank you. Have a good one, Joe.
JF: Thanks again for joining us on another OnPoint with Korem. And if you like today’s podcast, please leave a comment in the comment box where this podcast is posted, which could be Apple Podcasts, Google Podcasts, Spotify, or YouTube. I hope you’ll join us next time for another On Point with Korem where we’ll get on point.