LiLi Tile: Building the Self-Improving Ecommerce Site

[00:00:00] Mory Katan: The biggest thing with AI is, like, context equals output. If you don't have a robust context set, you don't have a really good output.

[00:00:09] Kailin Noivo: Welcome to another episode of The Ecommerce Toolbox: AI and Retail. Joining us today, we have Mory Katan, who is the CTO of LiLi Tile. And we have some really interesting things to talk about today. Welcome Mory.

[00:00:23] Mory Katan: Hey, how are you? Thanks for having me.

[00:00:24] Kailin Noivo: So, Mory, you and I first met a couple of months ago now. I think you were traveling in Amsterdam when we first launched our AI product, and I was kind of taking it to market. So, I was chatting with some execs, and you kind of came up. So, you and I synced together and maybe talk a bit about, first off, before you kind of got involved with us and ended up building your own AI brain that you call Owl with a lot of the Noibu data. Why don't you give us a bit of a history of your own career first and how you ended up as a CTO of LiLi Tile.

[00:00:57] Mory Katan: Absolutely. Sure. So, I guess a little about me: first-generation American. Once I was done with college, I took 5 years and spent it leading a development team at JPMorgan. I had a team of developers both here as well as in India, and we were overseeing various digital projects. And so, you know, that kind of got me pretty involved in the coding space. From there, I separated and started doing a lot more of my own things. LiLi has always been a part of a bigger family company that we have for construction materials. LiLi itself is a branch that we started around 2018 with my sister, and we kind of took a lot more of the design materials that we use, and we started listing them online as well. And so, you know, through that, we've been very active in trying to create an online presence, trying to be very tech-forward, and trying to always use the most innovative tools to bring the best experience to our customers.

[00:02:02] Kailin Noivo: Very cool. And you and I were pretty early on in chatting about, like, effectively recursive, self-improving websites, right? Like, how do you have always-on optimization. How does your site get better every day? Right? So, talk to me a bit about your journey with using AI and how, kind of, we got involved together and how you ended up building your own tool to leverage a lot of the Noibu insights.

[00:02:24] Mory Katan: I've always been, ever since the whole AI pocket became more popular, I've been very, very heavily involved. Even before then, I was doing a lot of machine learning elements. So, I've been pretty deep in the space. But the cool thing is I was very actively using a lot of different tools, though. You know, the problem with a lot of the tools, the AI things that exist, is like, you can ask it a question, and then you have to give it a lot of context, and then the answer it gives — sometimes it's relevant, sometimes it's not exactly spot on, depending on the context that you've given and the answer it gives is an action. So, you now have to take this new packet of code and do something with it. Right? You have to re-upload it to your site. And that process, it seems like a very tedious process, and it is; it's very painstaking. So, what we decided to do was to flip that on its head and try and innovate something that you can go through and, in one step, get the entire process done. Right? When you want to evaluate the site right through one flow, you want to go from evaluation to decision-making, to execution, to deployment, all in one study, in one flow. You shouldn't be going back and forth and back and forth. That time loss is time clicking. And so that's kind of what we developed with Owl. And we can get a little bit into that and what that system does. But, you know, it was essential for us to be able to look at all that data of what's happening online. Right? Noibu has an amazing ability for collecting all of that data, but you know, we have hundreds of thousands of people visiting the site every week. How does a human disseminate all of this? And so that's where I think AI makes a huge step: going through all the user session data and creating actionable insights, right? Is this, like, you know, if we know a certain button is driving higher conversion, why is that button not in a more prominent place? Can we test that out? Can we test it out on a subset of a few thousand and see if we actually have incremental growth? And so the thing is, oftentimes the human is the bottleneck. But when you actually give these binary rules to your AI, like evaluate the site based off of speed, go through and look at certain parts of the heat map, what it can do is it can derive decisions, it can act on the decision, it can deploy it, it can test it and then give you feedback. And so maybe we can even talk about what it did for the site once we implemented Noibu.

[00:04:56] Kailin Noivo: Yeah, for sure. Let's maybe start off with, like, out of curiosity, how were you using AI last year, and how are you using it this year? And how big of a difference is that, like, how is that adoption? Because obviously AI has been around for a while. Like, ChatGPT came out in 2022. Before that, AI was called ML, neural nets, all that stuff. But in terms of really, like, I don't know, I'll speak for myself. It's only really been this year where it, like, 100x my productivity because even last year I looked at — my brother runs a Shopify store, and that's largely who we partnered with to build the AI product because he's a lean operator, like, 1 or 2 people. And last year I was trying to get a full view of the value of his inventory and his, like, gross margin on stuff. And I was using AI, and honestly, it wasn't that good, to be honest. It was over the Christmas break, and like, it was giving me confident answers, but I knew the answers made no sense. I'm like, I know there's not $40,000 of inventory. Like, that doesn't make any sense. Like, there's way more than that. So, like, how come you're giving me that? And then it would, it would find its mistake, and you're kind of like, "Oof, like, this isn't that good." Like, this is now in, like, November or December of last year. Where we started going all in was, like, at the beginning of this year. We started thinking about it last year as we saw the frontiers get better and better. But I'm curious, from your vantage point, like, what, like, fundamentally, how has your usage of AI changed in the last year?

[00:06:17] Mory Katan: Yeah, I mean, you're spot on. I think a lot of it, like a year ago, we were doing very menial tasks, you know, simple text compilations. It has always been really good at text. I think the math element and the visual element are something that came later. Now that we have this, like, AI race between a few of these, like, big boys. And I think that will continue driving innovation. And the cool thing is, as these big platforms are getting bigger, there are so many incredible online communities that are creating, you know, their own versions of things and making it open source so that, you know, you can run a lot of these models locally, right? You can buy a pretty good computer and run some basic models, too. So, you know, the shift is tremendous. What I would say for us is that how we're using it is different. You know, a year ago, we were using it for very menial tasks, you know, replying back to certain emails, creating certain documents. Now it can still do that, and it does that incredibly, but what it does more so is it has the ability to organize and catalog your context, meaning, you know, how much inventory you actually have, like spreadsheets of your financials; it keeps that all as pockets of content and context so that it can pull on it better and deliver more accurate results more consistently, right, because that's the biggest thing with AI is, like, context equals output. If you don't have a robust context set, you don't have a really good output. And what happens with most of the models, and this is where, like, Anthropic really shocked the industry a few months ago, is when they were increasing the token limit. Right? But it's really a matter of, like, how much content can you feed into it in order to get a good result out. And so, that's something huge that we worked on with Owl, where Owl is constantly taking the conversation and learning from it. Right? So, it will take a conversation that you have between, you know, this AI and you, and maybe Kailin, you're going through and saying, "Listen, like, I don't like it when you start the paragraphs like this, or I don't like it when you format the numbers like this." And it's going to say, "Oh, Kailin doesn't like that. Let me keep that in the back of my mind." And so it goes ahead and, like, logs it for itself so it doesn't ever make that same mistake twice. That's something that you don't really get with most AIs because they just come to you as a blank canvas, hey, you know, what do you want as an answer?

[00:08:47] Midroll: If you're listening to The Ecommerce Toolbox, you're entitled to a podcast-exclusive website audit. Go to noibu.com/podcast-audit for a free scan that uncovers the hidden friction blocking your conversions and shows you where you're leaking revenue.

[00:09:01] Kailin Noivo: Yeah. You know what's interesting too is, honestly, like, I'm seeing in a lot of these convos, and now that you guys have Owl, you must be seeing it too. Like, everyone's like, okay, I'm here, AI helping me run my websites here. They don't know, like, the GPS coordinates in the middle. Like, they don't have the, like, they need the path. So, maybe take us down that path in terms of, like, what are some real-life use cases that you've been able to implement that effectively work? Do you know what I mean? Like, obviously it's off the stage, and maybe it will never get there where you just press a button, go on vacation, and everything's perfect. Right? So, it's like we know we're not there. Like, where are we kind of in that journey and what really works well and what doesn't work well, in your opinion?

[00:09:46] Mory Katan: That's a good question. So, what works really well when you set this up, the first thing that we really like to do is let Owl learn. So, it goes through all of the systems that you are already using, it goes through all the conversations you've already had with customers, it goes through your cadence, it tries to learn as much as possible in order to really understand the nature of your business and how to best suit it and how to give recommendations that are useful. So, you know, something that we used it for, which is very surface-level obvious, is we integrated it into our Shopify site with Noibu, and what it's constantly, constantly doing is improving the Shopify site for speed, for efficiencies, doing A/B testing in terms of testing for conversion rates. That's one facet. The second thing that we did is that it looked through our entire CRM. We were using a third-party provider. And so what it did was it read through all of that and, basically, recreated a CRM that's more custom for us and for the nature of our business. Right? So with LiLi, we're dealing with handmade beautiful tiles that have a lot of variations that have, you know, demand variations based of, like, where we're importing it from. So, a lot of these factors are things that we want to keep organized. The conversations with customers, we want to keep everything organized. Our marketing and any touchpoint. So, Owl created a CRM that was meant for us so that we can actually follow up with the customers. It created an inventory tool which actually logs and organizes all of the inventory and gives us predictive purchasing decisions based of that first link, of Noibu, right? So, we look at how many times did people hit a page, how many times did they add them to cart and add that to a decision factor of whether or not we should order it and how much we should order it. Other things that it does is that it looks at our financials and creates simpler dashboards, you know, pulling in QuickBooks. We have an entire warehouse inventory, like, pick, pack and shipping process, like the boards that you see in the warehouse; all that was created by it. And we have, like, you know, monitors all over the warehouse to make sure things deploy. Yeah. And I could go on.

[00:12:00] Kailin Noivo: On one end, you now have a connected system that pulls in your financials, your website conversion data, your order history, your cart data, some of the information relative to, like, your lead time of different SKUs. So, okay, you now have a connected system where all the data is connected. Makes sense. On the back of that, you could build dashboarding; you could build different views and, like, recurring queries. Makes sense. On the topic of website CRO, you mentioned a few things. So, number one, site speed. Taking Noibu's real user metrics and effectively having an agent always find speed opportunities, and then on a weekly basis or whatever cadence you want, speed up your site. From there, you mentioned something, like, I think you were chatting about, like, self-healing sites, so like solving technical issues, compliance gaps, things like that. Great. Now, where I'm very interested to dive deep is on the, like, A/B testing side because that's something that, to your point, everyone wants to prove incrementality, right? So, from UI/UX, designing, making sure you're pulling in that heat mapping data to come up with A/B testing ideas and serving that over the wire. How has that been? Like, have you guys been able to redesign any of your PDPs or any of that? Like, how is that? Because I feel like that's something everyone always wants to do, but it's just either timing, bandwidth, tooling, resources; like it's just, historically, before AI, it was really hard to like do the research, come up with the idea, write the test, launch the test, look at the analytics. So, yeah, maybe talk to us a bit about, like, the on-site CRO.

[00:13:40] Mory Katan: Yeah. No, that's a huge thing because we actually went through two very big edits in the last few months, and I'll talk you through that. So, the first one was LiLi, right? And we onboarded Noibu on a Thursday afternoon. By, like, Saturday, the Owl had already, you know, made a whole bunch of its own decisions, and Sunday night we launched. So, again, 4 days. Our site speed went from 3 seconds to 1 second. Our add-to-carts over 30 days went up 500%, and our general conversion went up almost 200%. We did the same thing again just two weeks ago on another brand, and we used to have a developer, a designer, a project manager and then, obviously, like ourselves overseeing this, and it would take at least two months to take a site from ideation to deployment. And then, obviously, a lot of the factors within it are like gut feelings, like, oh, where should we put this button? You know, what looks better? Things like that. We just did a website redesign where just the project manager was working on it, and in two days we had a design; the design was approved, it was implemented, and it was launched live on Shopify. 2 days. No developer, no graphic designer, just the project manager and, obviously, ourselves. And you know, we're already seeing a drastic increase and impact in this new site. So, it's really about that; it's about the cost savings both in time and in manpower. Right? Oftentimes, at least what we find in our business is that the developers aren't really the creative entities that are pushing ideas forward. So, it's kind of like you're losing momentum, going back and forth and back and forth, and hey, can we move that button? And hey, can we, like, you know, redesign this page? It's time lost, right? Your focus should be selling.

[00:15:33] Kailin Noivo: I love that. And to your point, like, all of these things have historically been gated. You couldn't run 20 A/B tests at the same time before.

[00:15:40] Mory Katan: Right.

[00:15:41] Kailin Noivo: Like, you could run a few, maybe. Right? And then coming up with dynamic landing pages for your ads and then taking the behavioral data and then launching an A/B test and, like, iterating it every week like this. I love this idea of, like, self-improving websites, and like really what you need is you need the data, and Noibu has all that behavioral data and all the data that we chatted about. You need, obviously, to connect your Shopify or your whatever, which, like, your cart system, and then you need an orchestration layer, and some people are building this in Claude, Codex; some people are building it in it depends on your own system, and then you have people building it specific for retail like you guys. Like, you guys are a retail-specific harness to effectively deploy these end-to-end workflows, which makes a lot of sense.

[00:16:28] Mory Katan: Yeah, I mean honestly, Owl spans far, far wider than just retail. We've also done a few service companies. But yeah, to your point, it's really about having that operator who can take it from idea to execution in a very rapid way without losing time, right, because you do want to iterate very frequently. You want to try out new things. You want to gain access to your data. The biggest thing that we're seeing is a lot of our customers, you know, they have a CRM so big, hundreds of thousands of people — they don't have any way of querying that entire database quickly. How many people, you know, customers in the last 30 days didn't receive an email from us? Blast out an email. Right? Most of the time, those are functions that take a long time to create, automations to build; this is like opportunity lost.

[00:17:15] Kailin Noivo: By the way, that is like kind of crazy. Like, you can't even query your own data. You're being held hostage.

[00:17:21] Mory Katan: That's what I'm saying. And like I have companies that, you know, onboarded for us. They took like 4 months to onboard Salesforce. It took them about 3 days to onboard our Owl with WiseBird, but you know, they didn't have ways, like, to query data in Salesforce; you have to create a lot of, like, specific functions. You know, it's more painstaking to get to that result. With a system like Owl, you can query 4 or 5 data points that have no correlation and just cross-reference them. For instance. I'll give you something. For us, what we do is that we cross-reference locational geographic, we cross-reference whether or not, like, there is a populace of pools, right? Because, like for us, for custom materials, pools is an impact. We cross-reference, like, how many times they visit the site. There's a lot of various factors that are like living on different platforms that you start mushing together to get better context. And this is something that, you know, big multi-billion-dollar companies do regularly because they have a team for this. But when you're running a team of, you know, a handful of people, there's no way of doing that.

[00:18:27] Kailin Noivo: No, fully aligned. And as we're looking to wrap up, Mory, maybe tell us a bit about — anyone listening, they're always trying to, like, straight up, I'm just getting asked almost on a daily basis through LinkedIn and when we go to trade shows, like, everyone wants to implement AI; everyone's boss is asking them how to do it. Like, I'm a retailer. We have tens of thousands of listeners on these podcasts. What do I do next? I'm an ecomm store. Like, I want to deploy end-to-end workflows to drive my conversion up and improve my operations. Like, what's the next step I should take?

[00:19:00] Mory Katan: I mean, the easiest next step is reach out to us, right? We are WiseBird Systems. But if you are interested in doing this on your own, I definitely recommend diving deep into the AI realm because a lot of these tools definitely help get you from zero to one. I will say that something like Owl took many years of ideating, and it comes ready out of the box with a lot of this like knowledge that you know, we all bring, which is a very, very important factor of it. And then also tying it together with good integrations like Noibu is critical because you know, you have data, you have the ability to understand data, and then you have action, right? And so marrying all of these things together is critical. So, I think the easiest first thing to do is just reach out, right? Have this conversation. Let's identify some workflows that are taking you a long time and can be repeated by AI. And if you're spending more than an hour a day doing admin work, you should probably evaluate that, right?

[00:20:01] Kailin Noivo: Sounds good. This has been awesome. Appreciate you hopping on, Mory. And please let me know if there's anything for you folks listening that Mory or I can help with; you can find us on LinkedIn.

[00:20:13] Mory Katan: We're here to help out.

[00:20:14] Outro: The Ecommerce Toolbox: AI and Retail is brought to you by Noibu. To find out more about Noibu and how we unify error monitoring, site performance and experience analytics to uncover growth opportunities and skyrocket your revenue, visit www.noibu.com. That's n-o-i-b-u.com, and then make sure to search for The Ecommerce Toolbox: AI and Retail on Apple Podcasts, Spotify or anywhere else podcasts are found and click subscribe so you don't miss out on any future episodes. On behalf of the team here at Noibu, thanks for listening.