GAINS On Podcast S2E14 – The Home Depot Expands Their Network Vision

Joe Davis (00:01):
Welcome back, supply chain innovators, planning pioneers and optimization enthusiasts. Joe Davis here, your Friendly Neighborhood podcast host with another episode of GAINS On, the podcast where we
explore the future of supply chain strategy, planning, and design, powered as always, by the brilliant minds at GAINS. Today we’re taking a look inside one of the most complex and wide ranging supply chains in retail. Home Depot’s network touches nearly every corner of the industry. Rapid deployment replenishment, direct to home fulfillment, extensive reverse logistics, and more. Joining me is Aaron Mosher. From the Home Depot’s network design and data science organization. Aaron and his team rely
on flexibility, experimentation, and long horizon modeling to support both DIY shoppers and their pro customers. Two groups with very different expectations, but equally high standards. We’ll learn why flexibility is essential in a competitive retail landscape, how Home Depot uses GAINS to test sensitivities and build alternative futures, and how the team’s scientific approach helps leaders make confident decisions even when the long-term outlook is uncertain. Ready to dive in? Aaron, welcome to the show.

Aaron Mosher (01:18):
Thank you very much.

Joe Davis (01:20):
You are the data scientist manager, is that right?

Aaron Mosher (01:23):
I lead the data science arm of the network strategy team at the Home Depot.

Joe Davis(01:27):
So what kind of things do you do there as the head of that department?

Aaron Mosher (01:30):
So the network strategy team, our mission is basically to help empower our leaders with kind of those big ticket investments. We need to keep our Home Depot supply chain truly cutting edge. So that can mean new buildings of all of the different platforms that we manage. It could also mean major investments inside those buildings. IE new robotics, new automation, conveyors. Yeah, that nature. It could also mean purely software of how do we take existing buildings, existing platforms, but allow them to connect together in a way that they haven’t connected before. Either to enable replenishment or fulfillment or just anything that we think could drive some savings.

Joe Davis (02:09):
Excellent. So it seems like you guys are keeping a constant eye on the supply chain and where you can improve that idea of continuous improvement. Is that

Aaron Mosher (02:18):
Hundred percent. I think and that’s really both a like to do and a must do. I think the retail space is incredibly competitive. Absolutely. And Home Depot wants to be a leader in that retail space and can’t be a leader by just keeping your head down and falling under the people,
Joe Davis (02:34):
Right? Yeah. Well I think Home Depot is a brand that a lot of people are familiar with, but what is something that you think would be surprising to you to learn about Home Depot’s, supply chain, something that’s unique to your business?
Aaron Mosher (02:45):
I think the first thing right off the bat is just how diverse it is. If you think of a lot of other supply chains, Home Depot probably has a piece of it, like grocery stores pioneered a lot of that rapid deployment, kind of get product to stores quickly and Home Depot a hundred percent has that.
Joe Davis (03:00):
Yeah,
Aaron Mosher (03:00):
A lot of people may think of Amazon Fulfillment shipping directly to your house and Home Depot has that. Home Depot also has a very robust reverse logistics center of getting it back and making sure we either dispose of it efficiently or occasionally get some returns back to the vendors. And another I would want the highlight would be our flatbed distribution center. So this is a little bit unique to Home Depot or at least the building materials space. How do you move product like lumber and concrete

Joe Davis (03:28):
Right
Aaron Mosher (03:28):
Things that you have to use a forklift to move and you have to use a flatbed trailer to deliver. And what’s really neat about that platform is it just smells wonderful. It all smells like lumber, like the lumber with
that as hundreds of thousands square feet. But also it’s really neat because it’s a dual purpose, full service building. Not only did it start life as a store replenishment platform, keeping that lumber aisle in stock, it’s also expanded to a really great tool for our pro customers because now not only can we ship to a store, we can load that flatbed with highly skilled associates and deliver it in job block quantities to your job to a work site. Yeah. And it has all of those kind of one-stop shop for many of those building materials type needs that you want. So you have these flatbed delivery trailers next to these 48 foot flatbed trailers replenishing a store and then it’s one big efficient platform all
Joe Davis (04:22):
Well, I mean it is two very distinct customers with two very distinct needs and you have to serve both of those customers well. And like you say, it’s not a one size fits all, model’s not going to work for that. So I imagine that the needs of the pro customer are much different than the needs of sort of weekend warrior DIY customer?

Aaron Mosher (04:40):
Yes and no. Okay. I would say that our pro customers also shop in the store, so having a positive store experience is as important for your delivering service to the pros as it is to DIY. And also I think a lot,
there are a lot of folks, DIY can mean a build it for me. It can also mean like, no, I’ve installed the deck, it took eight times longer than I need. We’ve all been there.
Joe Davis (05:05):
So Home Depot adopted GAINS pretty early on and one of the things that Home Depot identified is a need for flexibility and came to GAINS for that reason. Can you tell me a little bit about why you needed flexibility in your supply chain?
Aaron Mosher (05:18):
So there’s really two big reasons. The first you already broached, which we have a whole data science team supporting network design. And when you have that level of talent on the team, having some software that gets out of the way and lets them be creative and solve the problem they want to solve is great. And also that’s just what we have to be. The retail space is incredibly competitive and our recent earning calls, our leaders emphasize that Home Depot is a growth company, growth minded company. And one of the major areas we are expanding to is obviously growing that capability for pro customers. And as you mentioned, pro customers have a very high bar for service. So that’s pushing our buildings to be more creative, more efficient, and to offer more than they’ve ever done in the past, which means our models have to be able to do the same thing and concrete is slow and expensive. So the more you can test in that digital space, the more agile you can be in remain as a company.
Joe Davis (06:16):
So is that sort of what you’re doing at Home Depot now? You’re taking all this data, these data points and you’re kind of throwing them at the model and see where it breaks or see where it does well and try and replicate that or?
Aaron Mosher (06:27):
So I think we’d start to be a little bit more intentional. We always start with what are the key questions we’re trying to answer, what are we trying to do? And so we’ll do a little bit of iteration to say, well,
what factors matter? But we’ll try and design our models and our experiments to measure that. And I think it’s an important balance between, obviously you want to put a lot of things into a model because
then you can see some really interesting emergent effects. If you predefine and say, okay, I’m going to fix this in, I’m going to test this and fix this and then fix this. You miss when this thing impacts this or you
can find something kind of in the middle. So we love having that. But on the flip side, we’re planning five years out, we don’t know anything five years out. If we knew something for certain, as much as I love Home Depot, I don’t think I would be working at Home Depot. God, king of the Fed

Joe Davis (07:21):
Is that the official title God King of the Fed?

Aaron Mosher(07:25):
If you could predict five years out, I think they’d make, they’re

Joe Davis (07:27):
Probably right.
Aaron Mosher (07:30):
The Oracle of Delphi just buried in Knoxville, Tennessee or something. I dunno. So with that, I think the other balance is being able to run just sensitivity rather than trying to say, here is the answer, it’s a here
is a likely answer given this set of simulators, or here’s where two scenarios shift. And I think that is an important translation of rather than saying, here’s the business target, here’s the network. The network
shifts at this business target and this network’s cheaper, so maybe we should try and
Joe Davis (07:59):
I got you. Okay. So really it is, I mean there’s really just testing the limits of what things can handle, just kind of getting in there and pressing the different areas of the model with data to see where the shifts fall.
Aaron Mosher (08:16):
I think that’s actually one of the key benefits we’ve noticed from GAINS is that we can really have our data science team really lean into the science and do truly kind of open-ended research where we don’t know the answer and we’re just running experiments, testing hypotheses to say, well, what is the best way to run the network? What are the key factors driving this performance?
Joe Davis (08:36):
You’re using GAINS in some innovative ways and you described it as sort of creating alternative futures, which I think is awesome so that you could create a network that operates the best most of the time in an optimal way. Can you tell me about that?
Aaron Mosher (08:54):
For sure. So short of Being God emperor of Dune and truly knowing the future, some small portion of that power is to list out different hypotheticals and that’s how you understand where that shift is. Knowing exactly this one thread, where does that land is less important than knowing kind of again where that cliffs, where’s the valley between the two shifting sands of time. So I think that’s an important setup that we’ve done is to build that sensitivity into our model. I think this is a concept that we didn’t create. I think robust engineering is kind of a key topic in a lot of optimizations, research forums, and I think Home Depot definitely being on that cutting edge and being willing and able to take those risks to really drive the industry forward. I think that’s kind of where we’re sitting as far as the problem.
Joe Davis (09:45):
Yes. Well that’s great. Home Depots is not only able to give you that sort of is willing to give you that kind of leverage to say, Hey, we want this to be a success. We’re going to give you the tools to do that. Now go tell us how to be successful.
Aaron Mosher (09:58):
I mean there’s always a little bit of a carrot and stick here. Some of it is it’s the freedom to be able to do it. It’s also I think many retailers have seen the level of disruptions in the pandemic. I think many retailers felt that stick a little bit of it’s always going to grow. It’s always why do we have manage inventory? So I think it’s a little bit of both. It’s demand planning or risk is no longer a hypothetical.
Joe Davis (10:26):
Right. It’s going to happen. There’s going to be disruptions.
Aaron Mosher (10:28):
Well, everything’s fine. Tariffs have zero impact on retail strategy. Same. Yeah, everything’s flat, everything’s great. It’s fine guys it’s fine.
Joe Davis (10:34):
So beyond the tariffs and all the other factors that are always changing, I mean there’s a tremendous amount of seasonality to your business as well, not just with the inside and outside, but also if during storm season you could see hurricane season, you could see a lot of boosts and need for building materials and that type of thing. How do you cope with that seasonality?
Aaron Mosher (10:58):
So I think for us it’ll be interesting because when you’re making a five-year decision, you’re not making a seasonal decision.
Joe Davis (11:06):
No.
Aaron Mosher (11:07):
You’re making a decision for what will last over 15, 20 years. So I think it’s important to note, are you sizing to Easter Sunday? Are you sizing to that peak? Are you sizing the middle? One thing that’s really cool about GAINS is it offers a lot of flexibility to make a soft constraint where rather than just saying, my capacity is X, you can say my capacity is cheap up to X, a little bit more expensive up to Y and horribly expensive up to Z. Now there is a hard physical limit that at some point, but you start throwing money at
problems well before you reach that physical physical limit.
Joe Davis (11:45):
Right. Well I wonder in that, how do you find the balance of that? Do you sometimes say, I’m going to go for X even though it’s very expensive because there’s a net gain somewhere else, or it evens out
somewhere else?
Aaron Mosher (11:59):
I mean I think that’s kind of your mostly best most of the time. That’s where you can let the model decide what is your permanent long-term decision and what’s a short-term decision. Those short-term decisions probably have a higher cost on them and then by running multiple volumes, multiple scenarios and then constraining it to make one long-term decision across all those things, and then weighting the probabilities or weighting the cost, you can have essentially you’re doing an insurance premium at the end of the day. It’s just, well, what is the most likely cheapest? It’s not like, oh, this magic balance. You’re trying to predict it. It’s just like, well, if I think this is what I believe about the distribution of
problem, my actuarial tables and whatever else, what other data I have at hand. Right. At the end of the day it’s just, well, what is most likely to get success
Joe Davis (12:51):
Right. Yeah. When we talk about flexibility, it’s some of what I think we talk about what GAINS and that flexibility is the flexibility to have ranges as opposed to saying, this is our number and if our number is missed, we’re a failure. If our number is hit, we’re a great success. But having ranges like that you can say that you can plan for, because that’s that shifting flexibility I think, at least from my standpoint. Yeah. What changed when GAINS let your team shift from a will this solve to building a Home Depot
specific predictors for transport, labor, and that type of thing?
Aaron Mosher (13:26):
So I think, yeah, this is really allowing our data science team to function as a science team. And I think if you can think about it, the value of an optimization solver, the value of a network planning is really dependent on how good are your inputs and how much value can you extract from your
outputs. So time you spend on the solver isn’t really, it’s necessary. You have to have it solved, but it’s not actually directly driving that value. And so the less time you can spend babysitting the solver, the more
you can invest that’s directly driving that value. And so I think mechanically it’s just allowed us to revolutionize and really just consider a much broader scope of questions. It’s also allowed us to speak at
a much more truly strategic level. If you have to take a problem and kind of compartmentalize, chunk it down, what runs on a laptop essentially makes sense. It means that you have to make expert decisions or you have to make alignment decisions. You have to say, my business target is X, where that business target is getting all your leaders across those groups to align. So that just adds alignment discussions and it really slows the process down so that you really can only deliver one answer rather than by not having to babysit the solver. By being able to leverage cloud compute power, you can throw a lot more options, a lot more sensitivity into that problem and then turn it around and say, Hey, again, the network you can get shifts at this business target, So let’s make that the target. Cuz that’s what truly changes. If you’re really bad, then you’re going to buy into so much space anyway. At a certain point you’re just like, well, I have all this space, I’ll figure it out. Right, right. I think Home Depot, when I first joined, we started with the sink program, which is rolling out that rapid distribution center. There’s this concept of the river and the rocks, which is talking about inventory. Inventory levels are high. You can hide a lot of rocks. There’s a lot of problems you just don’t need to worry about because you got the inverter to write it out. As soon as you start bleeding that inventory down, you’re like, wait, why is this the time varying by three weeks? Why is this in spiking over here and all of a sudden you have a lot more problems to deal with when that inventory is
Joe Davis (15:41):
It is a warm coat, as my boss likes to say. So you’ve said that you were able to use the data that you get to speak to leadership at Home Depot about the cost of a strategy whole in for shorter and clear recommendations. So what does that look like in practice?
Aaron Mosher (15:59):
So as I mentioned, what it used to look like is just a lot of alignment discussions, whereas now it’s a lot more of a kind of an agile product conversation. We start with the, well, what are you wanting to do? What are the concerns you have about this network? What are the key time deliverables that we have to answer? So we start there and then we’re
able to come back and again, show that sensitivity, show rather than saying, align on this, here’s the buildings. We can say, here’s the cost before and after this target. And we can even highlight, hey, these
are the targets that are driving you. Right? You may be worried about inventory space, but if you have a speed to customer delay, maybe you’re not worried about inventory space. You have to build a new building just to get to that customer at that point to hit the number to have some extra space there.
Joe Davis (16:51):
Yeah. So have you found that having that data and being able to speak in those ways with your leadership has improved decision speed?
Aaron Mosher (16:58):
I think so. I think it also just allows us to work better and I think is actually closer, kind of, our role. We are very much an internal consulting team and a consultant, a good consultant generally doesn’t tell you the answer and say, here’s what it is. They say, here’s what matters. Here’s the information to empower you to make a good decision so that as things change, as you learn more, as you take a national model and handing over to a real estate team that says, Hey, I actually have to go build this thing now.
Joe Davis (17:27):
Right, right.
Aaron Mosher (17:27):
There are things that always change, and so I think having that sensitivity in that direction allows us to be more flexible. So does it make the first answer that much faster? Probably not. Okay. Does it make all the following answers and that agile pivot to keep it valid? Yeah, a hundred percent.
Joe Davis (17:46):
So it really kind of helps you dial in and then small corrections from there.
Aaron Mosher (17:50):
And it also gives us more confidence that we’ve explored more of the space rather than before it was a single shot in the dark. Is that the best shot in the dark? Maybe
Joe Davis (18:00):
Can you talk about a time where you found a tempting local savings or so you thought, oh, we’re getting a really good price on this, or this works out to be really cheap. But then when you played that out in the scenario or in your model when you played that all out, you ended up that these sort of cheap fix led to a long-term cost. Have you ever run into that?
Aaron Mosher (18:22):
So I think a interesting interplay on both directions where things can be easy to undervalue and overvalue is if you can think about your garage or your workshop and there’s a bit of a trade off between upfront money or capital labor and space, if you stack everything in boxes, you can free up a lot of space. But of course that tool you need is always on the ball list. So you’re going to spend some time moving boxes around. You can also spread things all out everywhere And be able to find everything you want in that it’s totally organized chaos out of site where you walk is
someone else’s problem, right? Just careful dance. Or you can spend a bunch of money having really fancy shelving, which tries to cut a balance where, hey, now you have more vertical space. You can kind of access things quickly, but also have it put away in a neat, accurately tidy organizational structure. There’s a cost and there’s that upfront cost to that storage. That same problem I think plays out in the distribution center where that custom shelving more and more could be automated storage and. Right? Right. Yeah. Capital cost is now in the millions rather than a few thousand for the Husky kit. Home Depot totally sells them. If you’re looking for some garage storage, they’re great. I have one. Great.
Joe Davis (19:43):
So in business, you’ve got a lot of experts and some of those experts use data like yourself, and some of those experts use intuition or lead with their gut. Let’s say there is value in both, I think. But for you, I mean you’ve told me when we spoke previously that 80% of your decisions align with internal experts. It looks like a good idea. They come to you, it is a good idea, but for that 20% that the data doesn’t align with the instinct, how do you handle that?
Aaron Mosher (20:22):
Yes, I think that’s great. I think in general, you mistrust an expert wholesale without reason than about your own peril. If they’ve been in a space for twenty one years, you know there’s data there. It could
have just been built up or constantly time of really getting to know their space. Where we think we find the model can deliver, and this is actually something we’ve noticed from GAINS, is being able to put all of those problems in together where they can really interplay. Humans are not always the best at seeing, they can see one or two step interactions that three or fourth degree interaction maybe not. Where like hey, me doing this causes a problem in someone else’s building. That’s when we mostly find that delta. So I think having the flexibility to show that emergent, that’s kind of an output. And that’s the value of being able to run a holistic network where you’re having all the variables play off each other,
Joe Davis (21:15):
Yeah