The interview transcript below has been edited for length and clarity.
Angie Brown: So we estimate across weather, geospatial, customer preferences, we've got about 1.5 million deliveries at risk on an annual basis. When we piloted Magic Apron in Chicago, and we're, continuing the rollout now, one of the headlines that really stuck out for me is we were having about a 5.3% failure rate on deliveries in Chicago before we implemented the very first version.
And once we had implemented the very first version, we had driven that down below three to 2.8, 2.9, and through there. So it was a pretty significant basis point drop. We're continuing to learn more along the way and continuing to get better. When it comes to customer sat, on-time delivery is a huge component of that, and we wanna continue to drive that number down as low as possible.
Sam Witteveen: Today we're talking with Angie Brown, CIO of Home Depot. Home Depot runs over 2,300 physical stores and also one of the largest e-commerce businesses in North America. I think it's around about $25 billion a year on online sales. They sell lumber, hammers, they can help you with your whole kitchen renovation, and that may seem a long way from the AI sort of use cases that we usually cover.
But it turns out that they are using AI in a variety of really unique ways, and the interesting story for us here is that they're using agents for things that happen in the physical world, figuring out how to deliver orders efficiently, answering phones in stores, helping contractors quote for projects.
Angie leads the team that is actually building all of this. Angie, welcome to Beyond the Pilot.
Angie Brown: It's great to be with you today, Sam and Matt. Thank you for having me.
Sam Witteveen: Angie, I mentioned the 2,300 stores. Can you give us a sense of the sort of technology business that runs behind this? Where does AI actually sit in here?
Angie Brown: So if I take a step back and just think about the company at large, just like what you mentioned, we've got over 2,300 stores and the sixth largest e-commerce business in the States. And so there's a huge, like, technical presence, but then how you run those stores, one of my favorite things to do is to take a new employee and go walk a store and point out all the different components of technology that make that store hum.
At the same time, like obviously with the digital presence that we're referring to a huge online presence a lot of times, it, people tend to forget the supply chain behind that powers it. So if you think about what really makes all this come to life, it's how we move products both out to our customers and to our associate- and to our stores.
And then you can just keep pulling the thread. If you think about marketing, if you think about pricing, if you think about merchandising, technology truly, when you look at our strategy for a company, it underpins everything that we're investing in and how we are bringing that experience to life for our associates and our customers.
Sam Witteveen: This is VentureBeat's Beyond the Pilot podcast. It's about enterprise AI in action. I'm Matt Marshall. Today's episode is presented by OutShift by Cisco's emerging tech incubation engine and driver of agentic AI, quantum, next gen infra, and beyond. So Angie we're talking about A- AI and agents.
Someone who's maybe putting in a new ceiling fan or renovating a bathroom. We've all done it. Walk us through, maybe I'm doing some phone research, I'm taking a photo, I'm loading it to your assistant called Magic Apron. How do you look at it as the person building AI? Walk us through each step.
Angie Brown: So one great place to start is just search online. If you come to the website and you start just looking at ceiling fans, maybe that's your first. There's a tremendous amount of logic that goes back into the search results that you're seeing. You can also engage with what we call Magic Apron on the site.
So perhaps you've narrowed it down to a couple of different ceiling fans that you're interested in, but you want to know the difference between those. You can engage with, a conversational agent that, and ask questions. You'll get prompted, "Is this going indoors? Is this going outdoors?" "What is the type of fan speed that you want?"
Or, "Do you want a light? Do you not want a light?" And help you continue to narrow down on that. There's little things along the way, like whenever you're looking for those ceiling fans, you're gonna see an estimated delivery date on each one of the product information pages. All of that. If you could think about all the logic that goes into that to compute behind the scenes how quickly we can get something to you.
We're showing you how many are in stock in your local store, if that item is actually in the local store. So maybe you've gone through an experience where, you've found a couple in the search results you're interested with. You've popped out to Magic Apron. You've better identified indoor, outdoor, et cetera.
And then the other things that you can start to engage with in Magic Apron are what are the tools that I need to get this done? Things as simple as a stepladder to, the right screwdrivers and things of that nature. Do you have light bulbs? Do you need to buy light bulbs? Our intention when you're interacting with that is to do a multitude of things.
It's not just about getting you the one item you came to look for, but it's getting you the things you need to get the project done and the confidence and the know-how to actually go and execute that project. You've got AI all through that experience. Obviously, we've trained a lot of models with product information, but also the project know-how to help you through that experience.
Sam Witteveen: Hearing that answer, I think a lot of us will say, "Okay, there's this, an assistant that leads me through all those things that you've just talked about." But I think- If I'm understanding the story correctly, the assistant, this Magic Apron that I'm gonna be involved in, is just one part. There's actually more in the AI in the fulfillment and delivery apps, right?
Or in, even in the professional con- contractor business. So encapsulate what you're most proud of your AI journey so far.
Angie Brown: Keep going from there. You've built your cart now, and so you're ready to check out. There's a couple of different places now that AI starts to interact with that. Your estimated delivery date.
But what's happening behind the scenes to get you that estimated delivery date is a tremendous amount of fulfillment logic. And that fulfillment logic is looking at where you are, where the items are located, and the fastest path to get you that item. If it's in our store, the fastest path could actually be using, the store as a fulfillment engine.
We may be able to get that to you in two hours or less. If it's not an item that we carry in our stores, maybe it's coming out of our, one of our distribution centers. AI is all the way through that. If this was a bigger item that we needed to go a delivery truck to your house, AI kicks in there, too.
We're gonna be assessing what that route is, looking out for bad weather et cetera, making sure, to increase our on time and complete. And then once you get to the other side of purchase, if you have questions, returns, you need to interact with a call center agent, anything of that nature, AI powers a lot of those, too If during that journey, let's say you wanted to pop out and actually call a store and confirm, that they had it in there.
Maybe you're a little suspect of, whatever inventory shows online. You can interact with AI when you call that store as well. They can give you the information on the aisle and bay. They can even send you the link so you can shop directly there. So it's coming to life in so many different parts of the experience, and I think the biggest thing for us in grounding is the friction points of a customer when they are shopping with us.
We are constantly leaning into what are those friction points and how can we make them better? And increasingly, the answer is by incorporating AI into that journey.
Sam Witteveen: So Angie let's talk about the Order Intelligence. This one I found really interesting, right? So this is one of your sort of first agentic kind of things that you guys have rolled out, if I understand correctly.
It helps you to basically find people's houses or find where the actual delivery place is. Can you tell us a little bit about that and how, how is, how are you finding that working compared to what you had there before?
Angie Brown: Really, it looks out for three different types of risks, like a weather risk.
If there's been flooding, icing, anything of that nature, could that be impacting the route? The other thing that it looks out for is what we call geospatial risks. So do you have a narrow road? Do you have a narrow driveway? Do you have a neighborhood access challenge? Something of that nature. And we're comparing the t- for the truck type that we're getting ready to send the product on to those geospatials, and then we're looking to see if there's an anomaly there that would keep that truck from actually making it through the recommended route And then the third part of that is customer preferences.
So a lot of times a customer has a drop location that they're very specific about, or perhaps we've done this delivery route before and we know there's a gate code, those types of things. So you can think about it in those three chunks, like weather, geospatial, customer preferences, and all of those things go in together to give us one of three signals.
Do we need to contact the customer? Do we need to contact the driver? Or do we need to change the truck type? And so I will clarify, when you get to that stage and you're getting one of those three signals, it is still human in the loop to take action on those signals. Obviously, as we think about how we're continuing to mature this, we're pulling that logic further and further up, and then to where you could start to also automate those actions instead of having a customer call to say, "There's inclement weather.
The roads could be delayed. Do you want to reschedule?" We could be popping that through your proven or your chosen communication method asking you directly via text and letting you engage with the calendar there to choose a new delivery date if you wanted to reschedule.
Sam Witteveen: This series is brought to you by OutShift, Cisco's incubation engine.
By creating an open, interoperable infrastructure, OutShift is enabling agents and humans to share intent, context, and reasoning. The cognitive evolution for agents is here. Explore the Internet of Cognition at outshift.com. W- where I'd like to go next is how you actually built this, this, this agentic delivery on this side of the order intelligence.
Like what is the model? What's the orchestration? How does that agent work and hand, hand the work back to the human?
Angie Brown: One of the biggest things that we had to think about whenever we were starting on this journey was like, what is our taxonomy gonna be? What is our stack gonna look like? And things of that nature.
So just in general, we had to put a lot of time and energy in, into what our spine was gonna look like. What is our platform gonna look like, and things of that nature. And really, we've layered this out. You can think about it at the top as the experiences that you interact with. You're gonna interact with things in a variety of different points of entry.
Underneath that, we've actually put a really thin layer tied to that client, and we call it a client planner that helps determine inference, pick up the right next high-level agent, et cetera, to pass it off to. And underneath that, then you get into the meat of our high-level agents, and these are typically driven by domain.
So for example, like we will have a domain agent around selling. We will have a domain agent around post-purchase. We will have a domain agent around delivery. Underneath that, we even have a finer grain level of sub-agents. One of the really good examples of that is if you're selling, you may have a sub-agent specific to paint.
You may have one specific to power tools. You may have one specific to configurables, et cetera. And then underneath that, a lot of this takes advantage of tools and common APIs and, our data stores, et cetera. And so when you think about this, specific with a supply chain use case, a lot of it was getting the data in underneath to where the agents and the sub-agents could interact with it.
Yes, leveraging tools like Google Maps. One of our biggest kind of grounding principles on this is that we expect reuse at the experience level. And we really had to work hard to figure out we're not gonna have one agent that rules them all how to create a sub-agent architecture that let us take advantage of the client-specific context of what it was trying to achieve, get it to the right place as quickly as possible but still allow reuse across different experiences.
Sam Witteveen: Yeah that that's fascinating that you've decided to go with that planner, which fe- feels fairly light. So it's planning f- but then you quickly divert to these sub-agents through specialists. So one thing that's missing from that is, is some sort of meta agent that's saying, "Aha, this person's just buying, a single hammer.
How do we make sure that we're ha- helping them with the entire project and lead them into that?" If you have just a planner and then just immediately designating it to a specialist how do you have that meta-agent that's, that, that has that bigger logic?
Angie Brown: I think one of the things, and we're testing our way into this, but one of the things that we are finding is you can inference a lot of context from how the customer is coming in to engage with us, if they are coming in online versus they, if they are coming in-store.
So again, adjacent from the delivery con example, but let's talk about Magic Apron as an example here. If I am using Magic Apron online versus if I am using Magic Apron in the aisle, and the, both of those things are accessible. However, your interaction model is actually very different in a store.
We feel like we can jump to a more meaningful, contextualized conversation based on how you're coming to us. That application layer of where you're starting can be a good indicator of that as well.
Sam Witteveen: So Angie, so let's stick with Magic Apron. Okay, so my guess is that the sort of the first early versions of that e- existed long before agents, right?
That, Yeah ... your sort of your online store and stuff like that. How do you then de- decide, okay, we're gonna put agents into Magic Apron? What sort of timeframe does that take to actually do that? And then also, how do you evaluate that?
Angie Brown: It's anything from, the agent development kit, which, we're a pretty hefty Google shop, so we've got a lot of Google ADK running around there.
Also within there is like, the agent registry discovery and onboarding, the tools registry, MCP enablement the control plane, agent-to-agent enablement. When you go underneath there, we have LLM evaluation and observability, around like obviously observing everything that's going on, and you keep on going down the stack into prompts and services, and we have d- a prompt management hub underneath that with memory management.
We get into models it, that have been fine-tuned with different pipelines. And of course, underneath all of that is a knowledge foundation with both unstructured vector databases, hybrid search and a data mesh underneath there as, as well, all wrapped with a whole ton of security guardrails and safety evaluations and agent identity and all of that fun stuff around it
Sam Witteveen: So I'm curious, like you mentioned 80K.
How do you make the decision to go with 80K as opposed to, LangChain or, OpenAI's offering or something? You mentioned that you're mostly a Google shop. Is that purely like the decision on 80K is just f- for that? Or what sort of timeframe do you give, and how do you evaluate these things before you decide, yes, this is the one we're going to production?
Angie Brown: The timeframe varies honestly by the, the decision you're trying to make and like we have a number of partners. I think our Google journey started back when we moved our, website to Google Cloud over a decade ago, and a lot of our data sits within Google BigQuery and, the vast majority of our data sits within Google BigQuery.
So there's some synergies that you get from that We're always testing, different things as well. Like for example, we've got Azure OpenAI. We like the tension between using, Google on one side and then OpenAI to judge the outcomes of, what we're getting from Google. We do something similar when it comes to writing code too.
We obviously have a lot with Claude, but then we keep OpenAI models as well to help us on code reviews and things of that nature. So it's not that we are a Google only shop, and I should definitely say that when you get into things more like computer vision or traditional machine learning, things of that nature, you're gonna find a lot more open way, like open source solutions out there.
We're heftier with Google on the generative side. I think one of the things is, the context training that has to be done anytime, there's a switch, so there's a stickiness, level there that comes in. We've also grown up our own like homegrown platform as well that we call Magic, Magic Forge, and depending on the type of compute that we're talking about, some of our more, most complex things are also using, Magic Forge and then with others on ADK.
There's a constant evaluation there. It's really hard not to go and chase every, potential new model, that comes out. But I think we try to take a pragmatic view into is the juice worth the squeeze? One of the biggest things that we're trying to do is drive for cost efficiency, and so testing different models from that perspective on the cost side too.
Sam Witteveen: I th- I think you're describing what, what seems to be kind of an industry best practice, right? Which is you have that control plane. You're building your own layer there, and then you're being able to then use whatever solution you need for that particular application.
We were going through Magic Apron. You rolled it out and there was an interesting reaction.
Angie Brown: Story went, we were super excited about Magic Apron, put it on the website. Customers liked it. We were early days. We were evolving there. There was still some things that we even learned on the consumer side, but the pros, like you said, just were not engaging and so we have been investing in changing that.
We have relaunched it on the B2B site. A lot of times you wanna talk about that project coming to life, that is typically what a pro is trying to interact with. How to complete a job. And so we have really leaned into a few different things like List Builder and List Builder AI, trying to help them, tell me a little bit about your job you're doing.
It could be, putting in a fence, give me the size, that you're working on, and we can generate a materials list for you based on that. So you're not having to do the math on the number of post counts or, things of that nature. How much concrete you need. Do you need overage?
And so those types of question prompts are, just very much more helpful for the pros. Even outside of Magic Apron, we've found that some of the ways that we benefit the most is for things like blueprint takeoffs. So I don't know if you've ever built a home from scratch or worked with an architect or had someone that did, but, it starts with a blueprint, and you have to be able to take that and understand, a lot of the core materials that you need to get started in construction way before you ever get to faucets and lighting and all the pretty stuff.
So from that perspective, your time to take that blueprint and get a quote back in those pros' hands is critical to how much, how interested they are in working with you. So we've got an entire workflow on blueprint takeoff that turns around a quote in, 24 to 48 hours instead of the week plus that it used to be.
And so from that pro perspective those are some of the things that we found, like speed really matters. How do we get them to a material list as quickly as possible? We built in the ability to take all sorts of input show me your napkin. We can translate that into
help you translate that into a materials list. We're helping them do something they need to do and do so even faster, helping them think through the prompts to create that list. It's not that a consumer may not want those things as well. That's always analysis we're looking at as to how those, I don't wanna say come together.
This is a great example how the built on the same core infrastructure, but we are contextualizing the experience slightly different on B2B. But, are there other signals to where we could start to expose more of that on the consumer side as well?
Sam Witteveen: That's a great example of how it's really not the, the technology to some extent, right?
It's thinking through what their use case is, what they actually need. And so y- they need speed, they need that blueprint, they need that list, they need the quote within a certain amount of time. When you roll that back out, is there anything that you can tell us around the technology side? Has, has your framework you talked about the framework being similar or different depending on the use case.
Has that framework evolved, right? What I mean by that is the agent to age there, there's the client layer and then the specific move to the d- to delegate to these specific agents. Has that, that all remained the same throughout or is that evolving your logic there?
Angie Brown: Can I say both? I think the same, high level intention has stayed pretty consistent, but obviously the evolution that we're doing underneath there is continuing to evolve.
A couple of examples I would give you of that, like so much of this work and bringing it to life is really less about the LLMs and more about the engineering around them. When we first started on this journey, we historically had to provide a lot more summarization in order to get that context, and now newer, better models, are better at that context.
The, the models are more intelligent, for sure, and we see that, coming to life. But I think some of the biggest things for us was, the depth of the questions that we get. These are not one-word or one, one-line responses, and so you have to truly build up that share context across and, go from it's not just about product knowledge, it's about project knowledge too.
And you can see it going even further into you could get into building codes and regions and things of that nature, and you just need to keep adding on additional context around this conversation to make it as, meaningful as possible.
Sam Witteveen: Do you look at that as the end goal is to build this sort of oracle that knows everything about construction, about, all, all those sorts of things?
I and the, and the thing I'm curious about too is you've talked about that really what has worked the best perhaps is just making what you currently have more efficient as opposed to, totally new products. How do you think about those sorts of things as well?
Angie Brown: A- and maybe this sounds a little goofy, but one of the things that I would say is, Home Depot, first and foremost, has always been about customer service.
We want to be the place where you can come and get the know-how you need and the confidence you need to get your job done. These tools give us an amazing way to, one, assist associates in having that confidence in order to, help asso- customers through a project, but then also help, assoc- customers directly through our digital platforms.
And so do we have this like, ah, be the oracle, goal? I think, I would restate our goal to be like, customers come to us when they have problems and they need to get projects done, and we absolutely want to give them the projects, the products, the know-how, and, the list to get those done as easily as possible, so they've got confidence to take on even more projects at their household.
If we can make the small one easy, you can move on to your next one and your next one and your next one, and we're there to help you along the way.
Sam Witteveen: Okay. So one of the, one of the other things that was really interesting that came up was the whole sort of use of IVR or an AI version of IVR i- in the stores.
Can you tell us a little bit about that? And then how are you finding... I'm also really curious to find that like you said, you've always been about customer service. You've got this, this huge team of people that actually work in physical stores. How do they feel about these new tools that you're bringing in?
Angie Brown: So when you start with the IVR, when we rolled out NGA with Google, we've got it in about 1,000 of our 2,300 stores right now. So when you call one of those stores, you're gonna interact with a conversational IVR instead of your traditional tree-based IVR. That IVR can help you with a few different things.
One, it is multilingual. Not all of our associates are multilingual that's a good thing. It can check your a- order status, it can help you, of course, with hours and things of that nature. Then it can go deep onto item information. So this morning, funny enough, even before this call, I was playing around with it.
I need to buy every driver in my family one of the jump kits for their battery if their battery goes down. They're not that expensive. They're wonderful to have on hand, and you can easily get your car started again. As you can tell, I don't even know what to call it, right? I call it a jumps kit.
There's probably a, a better term for it than that. So I call one of the stores that has it, and I was like, "I'm interested in buying, something I can jump my car off with. Can you tell me what you have in store?" And she asked me a question I didn't know the answer to, and I said, "I d- I don't know the difference.
What are my options?" She comes back, she tells me the two that she has in store, and then she asks me if she want... if I want her to text me the product. And I said, "Absolutely." She confirms it's the phone number I'm calling from, and then so she texts me a link to the actual product on the website to where I could pick it up and buy it from there if I wanted to or, obviously, put it in a buy online, pick up in store order, and I can read more about it.
So that concept of helping you understand what's physically in the store and a little bit even helping you get the product knowledge to get going, we estimate things like that will drive Probably close to a three times increase in the number of questions that we can answer on the IVR versus having to pass you to a person.
Now, the person's help is always there, and that's one of our guiding principles. We want you to be able to get out of this IVR experience as quickly as you want to get out of it if you want to talk to a human. But think about the other side of that. So let's say I am trying to page to that person in the aisle to help you and answer that question.
You get into the complexity of in what department, the product knowledge of that associate, are they available? Are they assisting another customer? Are you having to wait on the line while they finish that assistant? So none of those things are intended to, replace the the employee knowledge.
What we see happening more than anything is those employees can stay belly to belly with the customer that's in the aisle and make sure they're giving them great experience. And then, of course they can pop out to the phone if you want to. All of us in corporate, we spend at least a day a quarter working in each one of the stores.
And I was talking to one of the act- one of our deepest technical resources, you know, on this, and he was telling me about in one of his store days, a guy walked up with a picture, and the picture was of a plumbing something. And there were valves and switches and et cetera, et cetera, all over that, you know, pipe.
And the guy shows it and says, "Can you help me find this?" And Hari is looking at this, and there are, like, eight different products on this, and he's not comfortable in the plumbing aisle. A lot of associates aren't. And so he loads the picture up into, Magic Apron in the store in store mode, and they start to talk through what, component is broken.
And, in no time at all, Hari's able to, get that down to it's a ball joint that has to be replaced. They find the aisle in the bay. They walk over there. W- they get it. They can talk through any other products that they need. So that confidence, especially when you're talking about aisles like plumbing or electrical and things like that, for an associate to take, a customer through We are getting amazing comments from the confidence it gives our associates, because we're not looking to, do anything other than augment the knowledge of the associates and give them that exact same confidence.
Sam Witteveen: You have customers all over the States, so you have a much sort of broader sample than San Francisco, for example, right? Roughly, what is the percentage of people that are prepared to talk to like a, a more intelligent AI IVR nowadays compared to, like in the past, I think everyone was just like, "Give me the person, give me the person, give me the person."
As it's starting to actually work, and it's fascinating that you're saying that this is helping to answer three times, a- as many questions and things like that, how many people actually stay on are prepared to talk to the AI?
Angie Brown: I don't know. I don't have those numbers up in front of me, I will tell you that for sure.
I love listening to the calls, though, that start out, like you can almost hear the eye roll, right? As soon as you hear the IVR kick in, I think a lot of human nature is like, "Oh my gosh, I've gotta talk to, Can I just talk to a human?" And then the number of times like we, we listen to those calls and they divulge into then like the, "Okay, hold on, I'll give this a try.
Okay, hold on, that was helpful. Okay, hold on No, you did answer my question. Thank you. And like the times that you will hear a customer come back and say thank you. We all always are watching that. I don't think we're perfect yet and, by any means, and that has actually been some of the learnings.
Like I said, we're in 1,000 stores. We're, we're working on correcting a few things before we get it out to 2,000. And we want you to be able to get to an associate when you want to get an associate, and we don't want that to be frustrating. We have found that, the conversational nature of generative AI, it really likes to help, and sometimes the human's asking for a different kind of help, and AI keeps trying.
And so it's been an interesting mix where, like we're trying to now... Like we're gonna go back and layer in like deterministic controls on escalation. So it's great you start out conversational, et cetera, a few things, business rules when you ask for help, policy boundaries, et cetera, those are much more deterministic, and so we're bringing those things together.
Sam Witteveen: If you're in Silicon Valley, you're hearing, "Hey, this is a revolution happening." And I'm h- seeing it with my colleagues, friends, and acquaintances, where you're starting in an LLM, you're starting in Claude, you're starting in an OpenAI. You don't go to Magic Apron, right? You may eventually if you're forcing me to, but at what point does it flip?
Are you seeing that happen? Are you ready for it? And I've got a, I've got a question related, which is then how d- how do you appro- maybe we can get to this in a second, but how do you approach that, right? This idea of the oracle, the idea of that context and keeping that context for somebody alive, and how do you do that?
Angie Brown: I was recently at a customer solutions conference, and it was primarily retailers that were talking in there, and it was really interesting. The gentleman, from the company couldn't fathom walking in a phys- physical store Period. The end. There is a little bit of a difference in opinion.
And, whether or not that is where the masses will, move to, I think is a little bit too early to tell. Right now we see very nascent traffic coming from, LLMs when it comes to actually purchasing. So a lot of times we see a lot of product knowledge or question and answer that happens there.
And then when it's time to purchase, sometimes you're seeing that pivot move over directly to the retailers. But we're building to be ready for anything. We're building to be ready for, wanting to complete that transaction directly within the LLM. Of course, we think about how, that could come to life.
We think, knowing the customer in those instances is going to be, really important, being able to provide kind of a contextualized, personalized experience around that. So if you think about how loyalty may start to play into these even if you're on, an OpenAI, ChatGPT, et cetera, of that nature, and you're starting your experience there when it comes time to buy, the more you can even bring that customer knowledge and the knowledge of that relationship with a retailer up funnel.
But we're not waiting for it to take off, I think is the headline. We're definitely approaching the interaction model to be ready for that. Is it growing? Yes. And when you look at the growth that's happening, can you say things like it's doubling, it's tripling? Yes, but it was a really small number to begin with.
You can't help but think back to the conversations when, even with digital commerce, we thought physical stores were gonna go away, and they didn't. They became great complements. In a lot of ways, we think agentic commerce will feel similar. It'll be a great complement to other models that continue.
Sam Witteveen: There's this notion that Lowe's kind of won the race with the assistant, getting it out there early. But that Home Depot kind of has won on the plumbing, right? Every-- all that context around it. It may actually be set up potentially leapfrog, right? If we're talking about that context around The inventory, IVR, all those applications w- with, along with Magic Apron.
If you see Anthropic and OpenAI dealing with that context of the user, right? They're they're collecting all this memory and context around them. Every now and again, they have to jump in and provide th- that, that disclosure, "Hey, we have this memory feature. Do you wanna turn it on or off?"
How do you do that? It feels like you are set up to have this great context and to switch it on. And I'm just wondering what your approach is there to win this race.
Angie Brown: It's a broader question than any, one retailer. I think the north on knowing what our customers need and how we can remove friction is still a great guidance to that.
When we think about memory and context and things of that nature, so those are some of the next frontiers that we have to continue to productionalize. One of the examples that we use is I've asked you, I, I told you I need a pet-safe product. Why the next interaction am I having to remind you I need a pet-safe product?
And you could just think how that grows into a total home and things of that nature of what you've bought before, knowing your brand preference, knowing that you like Milwaukee tools and how then, we're serving up better recommendations. All of that just comes around to how do we make this easier for the customer?
And definitely having that knowledge on a broader realm makes it easier for the customer to do this.
Sam Witteveen: What are some of the other things that you see? What do you see as the future? Okay, you're talking about memory on cu- on customers and helping out customers in that way. What other things do you see agents enabling going forward with what you've been Doing
Angie Brown: I get super also geeked out about what happens behind the scenes.
We talked a little bit about the delivery agent. I love the opportunity as it relates to assortment, space planning. Today we have 2,200 stores, and as you're really thinking about each one has 30 to 40,000 SKUs that sit in there. When you start to think about the productivity of those SKUs, the assortment of those SKUs, what's selling in what markets and what's not, this is a long-standing retail dilemma that we've always had.
But man, the ability, like, to apply AI into that, and not just then AI, but moving into agentic AI, because it's one thing to come up with the perfect assortment. It's another thing to take that perfect assortment and translate it into a beautiful planogram. It's another thing then to put that planogram and assign it to a store and understand all the product moves that have to happen on that, and then baseline against that, store labor that want to utilize in order to reset that planogram, and then do it on a seasonal basis or on, as products are changing, whether that be, as products are coming into trend to do so more frequently.
A lot of those processes today are underpinned by heavy subject matter expertise, heavy human data analysts, entry into CAD systems and planning systems and things of that nature. You start to streamline all of that and turn it into, "Here are your signals. Do you accept? Here's what this is gonna mean.
We'll update your planogram and get it going for you." And it's just going to, enrich that f- flywheel around better and better store experiences contextualized for the customer base. At the end of the day, I want you to come into a store, find exactly what you want, know exactly where it is, and have the exact right amount of inventory in for it.
And when I can use all of this data in a richer way, it just helps the ability for us to do that.
Sam Witteveen: Angie, it sounds like you've got a lot of things that you're working on going forward. Thank you for joining us on Beyond the Pilot today.
Angie Brown: Thank you very much again for your time as well.