Watch Now: Real Results from Real Ecommerce Agents: An Industry Roundtable
Watch the webinar below to hear directly from peers how they put agents to work, which use cases they picked first, how they’re measuring impact, and what they’d do differently today.
Transcript
Welcome and Introductions
Kat Flowers: Hello, everybody, and welcome to today's webinar. I'm Kat Flowers, the Director of Customer Success here at Constructor, and this is a really fun conversation we're gearing up for today.
If you're watching the recording back, or you're here live with us, you probably fall into one of three camps. You're an observer — somebody who's just keeping an eye on the agents market. You're an experimenter — you might already be running, or have been running, pilots and tests with smaller audiences. Or you're a practitioner — maybe you've got a couple of agents live and you're iterating on the experience.
And you're in luck, because I've got three practitioners here with me today who are going to be talking about their experiences launching and iterating on agent experiences with their customers. As a highlight: this is not a sales pitch, and this is also not a scripted webinar. This will be a natural conversation, and we will be taking questions in the Q&A box that you'll find in the corner on Zoom.
A bit of housekeeping before I do introductions. Please ask your questions. Even if we don't get to them at the end of the webinar, we will be sending out a one-pager and some resources for you that include that information. It also gives us good feedback on the things you might be looking for as you're navigating this market. And if you haven't registered — if you were able to just navigate to this link — please make sure that you go and register. I heard from a couple of folks that they were able to get to the live screen directly, so make sure you do that, so we can follow up with you after the webinar concludes.
Without further ado, we've got 35 to 40 minutes of conversation here, so I'm going to pivot over to introducing our panelists. Richard, go ahead and kick us off — tell the audience a little bit about you and what your role is with agents in your roadmap at Belk.
Richard Spencer: Right. So I'm the Chief Digital and Technology Officer at Belk, a department store in the southeastern United States. My role is as lead of the overall technology stack for the company, as well as the owner of the Belk e-commerce business. We've obviously been working very closely with different AI agents across the organization.
Kat Flowers: Awesome, thank you. And Sharon, go ahead.
Sharon Gautschi: Hi. I'm Sharon Gautschi. I'm the Chief Merchandising Officer for Rugs Direct and Lightopia. My role with agents — we are using agents to help our customers find the perfect product for them in the most important space in their lives, which is their homes.
Kat Flowers: Fantastic. Sarah?
Sarah Merlin: Hey everyone, I'm Sarah Merlin. I head up digital here at IndiMade Brands, and agents play a critical role in our ecosystem, both from an internal and an external perspective. Internally: how can we be more efficient, so we can spend time on the things that matter? And externally, similar to Sharon: how do we get our customers to find the right product that fits their needs?
Kat Flowers: Absolutely. Thank you all. Just so the audience can really understand the differences between each of your businesses — Richard and Belk have a very large in-store presence. Sarah, you all are serving a pretty unique market of customer. And Sharon, home design is critical, right? People have a lot of questions about that, and there are measurements and design choices that need to be made. So I'm excited to dig into each of your experiences and how you cater to those nuances in your businesses.
The Decision to Act: Knowing You Have a Problem Agents Can Solve
Let's start with the decision to act. That's a big thing folks want to know about — how did you know that you had a problem that could be solved with agents? Sharon, I think you have a really good story here. Do you mind going into that?
Sharon Gautschi: I've been with Rugs Direct and Lightopia for almost eight years, and throughout these years — especially in the beginning — a lot of friends would text me pictures of their space and say, "What's the right rug for this room?" And I would dialogue with them: is this the style you're looking for? Is this what you want?
I also assisted one of our board members in outfitting her home with rugs. When we went to a strategy session shortly after, she said, "How do we put this dialogue and this power in every customer's hands that visits our site?" But at that time, this kind of thing didn't exist.
So it sat on our roadmap. We talked about different ways to address it. Do we hire interior designers who will live chat? Do we train our customer service agents? What is sustainable and scalable? But then we started to see product agents out there in the wild and actually reviewed a couple. And we found out that Constructor offered product agents at that time as well. It was exciting, because there was finally a technology that caught up with this need that we had. It made it possible.
I'd love to show you an example of the agent in action on Lightopia.
Kat Flowers: Let's see if I can call on Kai to play our Clio video — there we go.
Sharon Gautschi: There we go. Okay, thank you. So as you can see here, if you can read the print, it says: "I want layered lighting in my living room. How do I start?" If you don't have layered lighting in your living room, I do encourage everyone to go out and get it. And if you don't know what it is, you can ask our agent — you have multiple levels of lighting in your room, starting with overhead, then maybe a floor lamp, and then maybe a table or task lamp.
It's really exciting to see that this agent is able to answer this kind of very nuanced question. It's not just "what's the width of this pendant?" It's not just a spec question — it's able to answer this more nuanced question. So, really exciting to see that with agentic search, the tech has caught up with the need.
Kat Flowers: Absolutely. So if we're playing that back — you're able to put this consultant capability in the hands of everybody that comes to your website.
Sharon Gautschi: Yes, absolutely. They're able to not just ask spec questions, but really able to ask more about what's best for them, and learn. We push education and inspiration in our shopping journey — and you'll probably hear me say that more than once. You'll be able to learn what layered lighting is, or that you need damp-rated lighting for your bathroom, or what's the right kind of light for what kind of architecture. There's a lot of nuance here that the agent can teach you.
Kat Flowers: That's a great demonstration of identifying the problem in the customer and really solving for it. What a lot of the market is seeing is that everyone wants to go live with some form of AI. But the big question is: are we finding that need, and are we actually solving for it? Because customers remember. I know when I'm on a website and the agents meet me where I am and solve that niche need. If you're in apparel, maybe that's fit. If you're in your vertical, maybe that's design.
Bridging the In-Store Experience Online
If we pivot over to Richard, we've got a unique aspect of this in store, and creating an assistant-type environment. Richard, do you mind going into your differentiator, your customer problem, and how you solved it with an agent?
Richard Spencer: Yeah. We've long been working on this idea of how do we bridge the experience from in store to online? In store, you have such an expert-led, helpful environment, and we hear it from our customers — our customers love our customer service within the stores. Online has often felt much more transactional, and so we wanted to try to change that dynamic.
The idea was not just to have recommendation engines. Those kinds of things have always helped. But having an always-on associate available for customers to reference was something we really wanted to get to. This was one of the steps when generative AI first came out. We immediately said, "Here's an opportunity for how we can bridge that gap in a different way."
Answering Highly Technical Product Questions
Kat Flowers: Absolutely. And Sarah, do you also mind going into your experience, just to set the stage for the audience on each of your customer problems?
Sarah Merlin: Yeah. I'll back up a little bit for those who don't know what IndiMade Brands is. We are three distinct websites that are all within the craft and making space.
We were in the process of doing a Shopify migration. And when looking at what problems we could solve with the migration from a customer care standpoint, one of our top questions asked were very specific, nuanced product questions. For us, when you're talking about making a candle or making soap and percentages of ingredients to use, it's very scientific, and giving even a 0.1 percentage off could be very bad. We don't want to have chemical explosions.
So those questions that can be really nuanced would take upwards of weeks at times to get answers for, because our customer care team might not know. They would have to ask our product team. They might have to ask a manufacturer. But all of this data existed. We had all of these questions; all of the answers were tracked. And so my thought was, this seems like a really great opportunity for an agent — we have all the data available, how do we serve it to our customers? They can be self-serve, and our customer care team can spend more time on things that really matter and where they can make an impact.
So that's where we dove in with our agent. We put it on our product pages. It's similar to both Sharon and Richard: it's meeting our customers where they are. When you're on a product page looking at details of something, instead of searching or trying to find something on your own, you type a question, get an answer, and you can move on from there.
Support Questions as the Low-Hanging Fruit
Kat Flowers: Absolutely. Sharon and Richard, have either of you also found parallels in the customer care or support-type questions coming through that aren't necessarily assistive or consultant-based questions?
Richard Spencer: Certainly we see those kinds of questions come through. But to Sarah's point, for us it's a little bit more about how do we prevent bad customer experiences? Honestly, fiscally, it's lowering our return rates — that's a very expensive process for us. If we can help customers with fit, or look, or "how does this go with date night," or any of those contextual sorts of questions that these generative searches really help to answer in a better way, that's really been the big positive, as well as obviously closing that conversion. So in both areas, we see some real benefits. Sharon?
Sharon Gautschi: Yeah, real similar, just to echo what others are saying. I too really envision increased conversion, decreased return rate, but also potentially a shorter buying lifecycle over time. Right now it takes about eight days for a customer to buy a rug, and about 14 to 15 days for a customer to buy a light. But with the assistance of these agents, I can answer not only all of these technical questions, but any kind of question.
Surprisingly, a big question is returns — what's your return policy? So we had to make sure to add that in, to train the agent on that and the little nuances of it. If the agent can answer all of your questions in one go and you can have confidence, why wouldn't it shorten the buying lifecycle?
Kat Flowers: Absolutely. I might even say that that's a kind of MVP need — training these agents on your most basic support-type questions. It's low-hanging fruit to have an agent that answers those questions about refunds, or policies, or certain materials that might exist on the product that aren't necessarily consultant- or assistive-based.
So take note of that, for folks who are thinking about all of the capabilities of an agent and wondering, "Do I need separate agents for different things?" The majority of those agents can be trained on all of this customer query-type information.
Prioritizing Agents on an Already-Full Roadmap
It sounds like there are a lot of benefits here. We found the problem, we're solving the problem with the agent, and we're seeing a lot of early iterative results on success. So for folks who are thinking about launching agents, there's the question of prioritization.
This is a chunk of work. If we go back a couple of years, people were planning out their one-year, two-year, three-year roadmaps for their websites, and agents have just kind of come out of left field. They've shown up, and now it's, "Okay, we've got a year's worth of planning, or two years' worth of planning, that we already did. How am I going to fit this on the roadmap?"
So if we could walk through some of those experiences — we've identified the problem, we want to do the work, when are we going to do the work? How did you prioritize and measure the agent impact against other projects you had on the roadmap? This is an open question; anybody can go first.
Richard Spencer: I can start. I think the key for us was that we started looking at how we could potentially use agents, and we had a conversation with Constructor about different things. Kat, you were actually one of the ones who brought up that this was something you were interested in working on.
For me, it was very straightforward. I have a certain amount of bandwidth, but if I can leverage my vendor's bandwidth to get the same things done, I'm more than willing to do that. It required that I worked on parts of it myself, and some of the other leadership, to guide what we were trying to achieve. But it was a very small tech lift on our end. That was very attractive, to make sure that we got it done in a timely manner.
Kat Flowers: What would you say was the most time-consuming part on leadership?
Richard Spencer: Making sure that we understood what would make a difference, and filling out what that vision looked like. Some of the conversations we had with Constructor really helped to flesh out exactly what we were trying to achieve.
I had a very nebulous idea — like, it'd be cool to have an associate live in my customer's computer. So you've got to take that and ask, okay, but what are the most likely things they're going to help with? You start with basically search and recommendations. How do you make sure that happens?
We reviewed a lot of prototypes, gave feedback like, "Well, this doesn't work, I don't like this, it's too slow," whatever it is. And then we just disappeared and let Constructor go work on it. And then they'd come back and show us what they had achieved. It was a great cycle.
Launching an Agent Alongside a Replatform
Kat Flowers: Awesome. I would love to showcase — speaking of Constructor's agents, we've got a video of your agent, Sarah, and we'll put that up here on the screen to chat through. But you have a unique aspect of your agent. Did that require additional work in prioritization, or were there some things that you found along the way while you were prioritizing the work for the agent?
Sarah Merlin: Yeah. As I mentioned in the beginning, we were going through a Shopify migration. So in terms of priority, I had identified this problem as part of the migration. To me, this was a key differentiator for us in terms of launching the new site, and especially within our competitive space. So I knew exactly what I wanted this to be, and thankfully, when we were talking with Constructor, it was, "I have this."
It was a decent amount of work from my team internally, going through all of the questions and making sure we trained it and curated appropriately. This was a big thing for us, especially with some of the scientific pieces of it.
Before I get any deeper in terms of measuring success — I'm a little bit of a nuanced use case here. I didn't know how this would convert or impact conversion. So I didn't want to put that as the goal; that to me was a happy byproduct. The real goal here was reducing the amount of questions my customer care team got for all of our brands, which I can happily say we cut in half — and growing across the board since launch, across all three of our brands.
But what's really cool is there are a lot of unexpected things that came up from launch. As you see in this example — this is on one of our soap product pages — someone asked a question in Spanish, and our agent was actually able to respond contextually in Spanish. We noticed that a small but growing share of customers for this specific brand were asking questions in Spanish, which led us to actually prioritize doing Spanish translations across the site experience.
Getting this launched was a bit technical, because there was some work from Constructor, some from my internal team, and some from our development team as well in terms of getting this live and running. But it was relatively seamless, and the quality of what we're getting from the data — we look at this every month and ask, how can we improve? What can we be doing differently? What information are people looking for that's lacking?
Similar to the lights piece, we sell a lot of fragrance oils, and conveying scent online is really challenging. So we've gotten a lot of really good data out of this in terms of what we can be doing to improve and change our experience. That's a really good example of how we've done implementation.
Building the Business Case for an Agent
Kat Flowers: That's fantastic. And I have a question for you, Sarah. Understanding that you identified the customer problem, talked with the team, and decided to prioritize the agent — what if this was somebody on your team? Imagine a world where you're not thinking about agents, but as the leader, someone on your team is bringing you a business case. What are the things that you would be looking for in an agent business case in order to prioritize the work as their leader?
Sarah Merlin: I think that trifecta is win-win-win — win for the business, win for the customers, and win for our internal teams. This was one of them.
A lot of times with agents, it is an investment, depending on your infrastructure, and you need to be able to prove out that whatever you're doing is going to move something. What is the metric you're going to move? And ultimately, how is that tied back to revenue? In my case, I tied it back to savings in our customer care tickets. But to Richard's point, it could be conversion. Or to Sharon's, it's certainly that time to purchase is lower.
But really: what problem are you trying to solve? Is it a made-up problem, or is it real? Is it helping internal teams, customers, the business? And really the financial impact — what needle are you trying to move with this? And then, how can we either MVP it or think about a plan to get to a definition of good?
Kat Flowers: Totally. Sharon, I'd ask you the same question — overall prioritization, of course, and how you knew you were going to fit this on the roadmap when you did. But also that similar question of, if you weren't the one thinking about agents and rooting for them, what would you be looking for from your team for solving your customer problem with agents?
Sharon Gautschi: For us, we were really excited that this tech existed and came to light. Fortunately, I have a very nimble team, and we opened the path. We certainly have a very full tech roadmap, but we opened the path. We said, "No, we're going to move forward with this. This is the future of e-commerce."
So it was easy to get enthusiastic about it and make time and space for it. I love the way Sarah put it — that if it's a win for the customer, it's a win internally. It's sort of an easy answer. This is the future, and so it's just a matter of when to apply agents to the site for the customer, not really if we wanted to do it.
Moving Early in a Fast-Growing Market
Kat Flowers: Richard, you shared a similar sentiment in the past — that you want to be one of the first to be doing this and you don't want to miss the boat.
Richard Spencer: Yeah. We know that generative AI has a lot of noise around it that's really positive and brings attention. This is something that for us, we looked at strategically. If I look at the total sales numbers that are influenced by it, especially in some of the different ways it's being done right now, some of those numbers are not super impressive — but the growth rate is very high.
So for us, we wanted to be able to try some different things and learn about different things, so that we're ready. If you're starting to get ready when someone else has already figured it out, then you're going to be too late, and it's going to really adversely affect where you're at. I'm super excited to see some successes, but it's also equally important just to prepare my team and prepare my organization to start thinking about some of these things and where they will lead eventually.
Kat Flowers: Absolutely. And Sarah, did you have anything you wanted to add to that?
Sarah Merlin: Yeah, I was going to say, I think that's something that is a really good takeaway for people listening — that it is really unique to our three experiences. When you see AI and agentic, the first thing that comes to mind is a chatbot, like a help center chatbot. What's really unique about all three of our use cases is that we're finding unique problems that are enabled with agentic. I think that framework is really critical to why we've had so much success — or at least why I've had so much success.
An Unexpected Win: A Bilingual Agent
Sharon Gautschi: Can I say that I was really impressed by the IndiMade clip — that your agent is bilingual? I thought that was amazing. Is that something you had to turn on, or was it already there? And by the way, just really smart to take that simple learning and then apply Spanish translations to all of your products. What a great learning that was. Is that something you knew already? Did you have to turn it on? How did that work?
Sarah Merlin: That was a total happy accident. When I was looking at the data, I saw the questions were being asked in Spanish, and I thought, what happened? It just so happened — it was very serendipitous — that the language support was released across languages about a week before I happened to analyze the data. So no, there was nothing additional we needed to do. It was just turned on.
And I'm really curious now. One of the latest things we released is the ability to do two-way chat. So you could see in the example that not only could you type a question, but then you could follow up. The really interesting insight out of that is — I actually thought at one point that the more questions you asked, the less likely you would be to convert, because if you're having to ask something a lot of times, maybe the response isn't attuned. Actually, it was the complete inverse. The more questions someone asked, the more likely they were to convert and engage.
Which does make sense — you're going back and forth and engaging. It's just not the expected behavior. We're already seeing a lot of really interesting learnings out of that. It was honestly a beautiful happenstance, I will say.
Customers Don't Always Know What to Ask
Sharon Gautschi: I do think that customers sometimes just need more ideas or more guiding, so the two-way sounds really good. That's something we started to have a little bit of in our agent. That's why, too, when we were looking at product agents, one of the companies we talked to said that some high percentage — like 80 percent — of the time on the product agents, customers are clicking on the suggested questions.
I really do believe it's because customers have questions but they don't know what to ask or how to ask it. And they see one and think, "That's a really good idea," and they're getting their information that way.
Sarah Merlin: Very true. I don't know, Richard, if you're seeing something similar, but what Sharon is sharing is exactly what I'm seeing — which again was not at all what I would have expected. Our customers really know their product, or they're making something very specific. I would have thought that the suggested questions weren't as important as they were. But yes, I'm seeing the same thing too.
Clarifying Search in a Large Catalog
Richard Spencer: There are a lot of different use cases or experiences. Our customers are going through something that is not too dissimilar, which we found with the way that we are approaching search: because the search becomes a little more clarifying, the customer ends up with less.
Our site has a lot of products — we're a department store, we have a lot of products. So sometimes what we believe is happening, and the data sort of correlates this (it's hard to get causality), is that if you search for something like a yellow polo shirt, you'll get so many results that it becomes a little overwhelming for customers. They don't really know how to buy, because they don't want to buy the wrong one. It's this weird thing where you have too many choices.
But because the agent helps clarify that search down to recommendations, instead of having four pages of shirts to look through, you have 20 shirts to look through. It makes it easier for them to find the actual thing they're looking for and feel good about it. And when they feel good about it, they buy it. That was an unexpected advantage. It makes a lot of sense when you think about it. We're doing some testing to make sure that our theory is truly correct. But that's exactly what we were talking about — by doing this kind of experimenting and looking at things different ways, you get some really cool ideas.
Key Iterations: Placement, Naming, and UI
Kat Flowers: Speaking of that — what are some of the key iterations that you all made? There was an initial release of the agents for you all, and you set out with an initial hypothesis. What were some of those early learnings, and then the key change that you made? Was it a UI change? Was it a naming change, or a placement change? Sharon, go ahead, you first.
Sharon Gautschi: One big consideration when you're integrating this is where you're going to put it on your site and how you want it to play next to classic search. From what we understood from the team, plunging a customer into agentic search without still having classic search could create a bit of dissonance. And we didn't want it to feel like you type in the search field and think you're going to go into agentic search, but you go into classic search, or vice versa.
So there was a lot of thinking about how we do this on desktop and mobile. This is kind of in the weeds, but we made some decisions like: if you type into the classic search field and click on "Ask Clio" — our agent's name is Clio — then it will take you into agentic search. But if you type in classic search and hit enter, then it will take you into classic search.
On mobile, "Ask Clio" was presented on load. But we felt like there was too much going on. It cluttered up the heading of the mobile experience. So we tucked it under, so that when you click on the magnifying glass that has a couple of little agentic stars, it will open up both classic and agentic, and then the customer can select or choose their journey with what they're comfortable with.
It was all these little decisions like that throughout the experience that you really have to think through. How will the customer be comfortable with this? You don't want the customer to have to be retrained to use your website. That's the biggest downfall in UX. So it's all the little decisions like that. That was a specific example. Richard?
Richard Spencer: We went through a variety of different tests and different approaches. Where do we indicate it? How do we name it? Initially we were like, let's glom on to AI. I forget what the exact name was, but it was something like "your artificial search agent."
I will tell you honestly what we decided we liked for a lot of it: we just stole from some of our competitors. What we experienced, we thought, actually makes it feel much more comfortable to customers.
One of the things that we really liked about the search that we did recently with Constructor is taking it to where — and it's a little different than what you did, Sharon — when someone types in something that's a very generative sort of statement into our search bar, instead of them having different buttons to push, which still exists, we just basically choose for them based on that behavior. That's basically us copying what Google's doing. If you ask Google certain types of questions, Google gives you a generative answer and then their regular search answer below. We looked at that and thought, well, that seems to really work well. So we went in that direction, and we'll see how it goes.
So you see the initial load, and then you have more of an agentic entry point that you can choose. We went through different iterations of this as well. Initially it was less interactive — it was more directive — and we felt that when people get into a chat flow, they want to be more interactive. So we made those changes as well.
And just like anything else, you measure and you see if you're going in the right direction with these changes. The great thing is Constructor has been super responsive. Sometimes they look at me like I'm crazy, but they don't usually say it out loud. So I appreciate that.
What Partnership with a Vendor Actually Looks Like
Kat Flowers: Luckily, I've never said it out loud, Richard, in the past couple of years.
That's a really good point, though, Richard. For folks who are here that may not be doing agents yet, or are executing on them and wondering what this work is like when you're working with a vendor partner — it really is an equal investment from both sides.
Speaking from the Constructor perspective, we've got great playbooks and processes that our teams run through, and I think that's a great way to get agents live. We really want to consult. You want your vendor to be able to consult with you and offer advice on what's best in class, or what type of direction we can offer based on what your vendors know about your customers — and likewise, you're making decisions on your side for your business. What should this UI look like?
Customization is inevitable with agents, because very rarely will an agent come out and just work for every single e-commerce website. Branding is different, tone is different, the problems are different. So the investment of time from the e-commerce and leadership side is so critical.
Naming, Trust, and What's Next
Sarah, talking about your agent — any key iterations that you made, or any initial launches where you were like, "That didn't quite work the way we thought it was going to," or "We want to make some changes in order to solve maybe a new problem that we incidentally came across after launching the agent"?
Sarah Merlin: One of the key decisions that we made was how to name it. A lot of our customers are makers, and we did get feedback around AI and hesitancy around it. So we were very intentional about saying, this is our AI assistant. We didn't name it — it's just the AI assistant. I think that also built trust with our customers: this is AI, but it's okay for you to interact with it. So that was a big one.
The two-way chat was a very clear next iteration. We were seeing such good interaction with it that I wanted to be able to expand that. AI in search is yet to come. For us, it's more of how Richard talked about semantic search — that's a big part of it. But also, how do you tie product and content together? I think that's a huge opportunity for us.
We have hundreds and thousands of recipes that are shoppable and that feature all of our products. For these makers, how do you find the inspiration for what you're doing? But also, ask not just a search term but actual semantic questions, and then get a result that's powerful enough to surface either product, or collection, or any kind of blog content data appropriately.
In a world where maybe 10 years from now everyone will be shopping via an agent and not on a site — what makes you come to your site directly? That's really ownable: being that point of brand contact for that maker, for that very specific thing. That's what I think about in terms of what to prioritize and what to work on moving forward.
ROI: What the Numbers Actually Showed
Kat Flowers: That's fantastic. I'm keeping my eye on the clock, because we've got seven minutes left, and I feel like we could go on this topic forever.
I'm getting a lot of questions in the Q&A about ROI, and so everybody wants to know about that. What are we seeing in terms of high-level numbers, iterations on metrics over time? And also, if any of you have a metric that you didn't think would be impacted but was impacted — what are your high-level stories?
Richard Spencer: I can provide a couple of quick ones. When people enter into the generative search or generative chat, our conversion rate is a little more than roughly double right now.
Some of that's intent — some of that's because they're interested enough to ask questions and to be engaged, so they're probably more likely to buy. But we've done some testing, and some of it is also just about restricting and guiding people to the right products, and the right products at the right moment.
The one thing that we didn't necessarily plan on, although it makes a lot of sense, is that we saw return rates for those orders were slightly lower than our others. So we're trying to figure out how to double up on that to improve overall return rates, because returns are a huge negative for our business. Some of it's hard to avoid — we have customers that will buy four dresses for an occasion and then return three. They try them on, they want to see how they look. So we're looking at how we can continue to improve in those areas. But we definitely have seen lower return rates, mostly probably around fit questions they ask the assistant. No one likes returns, I promise.
Sharon Gautschi: None of us like returns. I can chime in. We don't share a lot of metrics very often, but I am happy to share a little nugget for the product agent on Lightopia. We have seen an over 50 percent increase in add-to-cart rate and purchase rate. So it's been quite impactful.
And thank goodness, it makes sense — our hypothesis made sense. The customer has technical questions and they're not eager to reach out to a customer service agent. We have lighting specialists and experts that you can call, and they will outfit your whole home with the right lighting that is not going to blow up or be wrong for the space. So yes, that's a really nice metric.
Sarah Merlin: On my side, when you think about ROI, it obviously costs money to put an agent on your site. So I was looking at, could we reduce the questions to customer care enough to cover the cost of this? And actually we reduced them to double the cost of the agent. So we're making profit off of it technically, depending on how you cut the data.
That's a really good way to look at it. Is the conversion lift rate, or the add-to-cart rate, going to cover the cost of what it is to both develop and implement it? Depending on your customer base, it might be a challenge to prove that out so quickly — or you might be pleasantly surprised.
How Customers Actually Received the Agents
Kat Flowers: "Depending on your customer base" — it's such a good line, because there are so many factors with launching agents that depend on your customer base. One of them being: how will customers receive this?
What we know is that there are a lot of mixed reactions to interacting with AI in general, even outside of e-commerce. So whenever you think about your customer bases, were there any challenges in getting them to adopt the agents that you put on site? Or did you find that, for your particular customer bases, knowing what we know now about them, that it was easier than you thought?
Sarah Merlin: It was much easier than I thought, honestly, because you always hear the most from the loudest people — Kat, I'm sure you appreciate that. We had gotten a lot of noise about "I don't like AI, it's taking away from my creativity." But truthfully, we got zero complaints on our agent at launch, and it did way better than I ever would have anticipated in terms of engagement, conversion, and how people were using it.
I think why that was so successful is because it was in the right spot. We put it on our product page, when people are thinking about specific product questions. We met them where they were. And that's how I think about search and moving forward in the roadmap: what are those points in our customer's journey with us where it makes sense that we're solving a problem in that time of need?
What's Next: Where These Teams Are Taking Agents
Kat Flowers: And if we were thinking about the one thing that you want to do next with agents, are there any specific functions that you all have your eyes on?
Sharon Gautschi: Oh yeah. Do you have another hour?
For me, I just think that agents are a no-brainer. They're the way of the future. Why would you go into classic search and just type "brown rug," when you could type, "I want a light taupe rug that will go with my dark blue couch, and I need it to be kid- and pet-friendly"? Why would you just type "brown rug"? I think our kids are going to grow up asking what classic search is — they won't know what it is.
Ultimately, what I'm looking for is — I think we'll see it iterate, and we'll be able to have customers describe a space, and then the agent will outfit it to their needs and furnish it just how they want, and the customer will be able to easily say, "No, not that one, I want something different."
I would love to see virtual walkthroughs. Let's say I'm talking about all home furnishings — walking through a virtual home, being guided through different products and seeing them. Almost like a virtual showroom. "Okay, how about this one?" "Yeah, my couch is just like that." Or, "No, swap it out, I'm looking for a different kind of vibe. Show me something different."
So really bringing, as Richard has said before, that store associate into the experience, and having it personalized for every single individual shopper. That's what I'm looking for in future agents.
Kat Flowers: Sarah, Richard, how about you both?
Richard Spencer: For our end, maybe I'll answer in a little different way. Some of the big areas we're thinking about agents — even within the Constructor platform — are agents to help our employees do their job better. So agents that help with finding bad product or bad data, or maybe helping with categorization updates. Or more general questions like, how can I maximize selling clearance? I'm trying to move more inventory and get rid of some older product, and have it suggest, "Well, you can create a new category," and then you can say, "Yes, please do that."
So just these agents that will do analysis and help you run your business. Things that a lot of us are already writing into our mail programs. Sarah, I think you talked about having a chief of staff agent. That's the same idea — within these products, I think that's the next iteration for a lot of us: how we can reach further into the depths of our analysis and operations.
Sarah Merlin: To add on to that, Richard, I think about this from an internal and external perspective. Yes, I have an AI chief of staff. I highly recommend everyone building one, and I have an army of agents that I really want to figure out how I can deploy across my own org. I think there's a huge opportunity for finding efficiencies and workflows and how people do things internally with AI.
But externally, from IndiMade's perspective: how can we be the go-to resource for our makers? There's a lot of opportunity, depending on if you're a creative maker or you're more optimized on business, and through that spectrum — where do we inject into the experience to make it easier, not necessarily for them to buy all the time, but to use our platform for creation and ideation and discovery?
We're similar to Richard in that we have tens of thousands of SKUs, and so people tend to generally find a scent they like and stick with it. If it's selling for them, that's great. But how can we push discovery through AI is a huge opportunity — something I'm already starting to think about in terms of the holiday season upon us. How can we use AI internally and externally to provide better experiences? I could talk about this for hours, but that's my high level.
Closing
Kat Flowers: Absolutely. Those are all fantastic focus areas, and we could go on for hours about these subjects. Thank you three so much for investing the time and sharing your experiences, for our folks who are here live with us and have hung on. And thank you to those of you who are watching back and have made it to the end of the recording.
Whether you find yourself in one of the three buckets that we mentioned earlier — maybe you're just learning, maybe you're experimenting, or whether you're a practitioner getting your reps — I hope that you've found some advice in here.
My key takeaways from this conversation are: you need to find your customer problem, make sure you have a good entry point for your agents, build a hypothesis, and keep an eye on all of your metrics, because you might have some unexpected benefits. Get comfortable with customization, be ready to meet your customer where they are, and iterate on the experience over time. And of course, choosing your vendor partner through that experience matters.
If you want to talk about agents and you want to learn more about our Merchant Intelligence Agent — which, to Richard's point, is about how agents help our staff members be more efficient and get actual insights out of their product discovery platform — you can head over to Constructor.com. If you're not familiar with us, look into Merchant Intelligence Agent and fill out a form so we can chat more. Or you can investigate some of our resources around Product Insights Agent, AI Shopping Assistant, and ASA in Search, which are the agents that we showcased here today.
So be on the lookout for the recording after this session concludes. And thank you so much for your time today for this conversation.