What if each of your customers saw exactly what they needed on your site?

[00:00:00]We all know that personalized experience can pay off in big ways, but what if you could continuously adapt things like search, product carousels, and more for customers even from a cold start?

[00:00:12]>> [music] [music] >> Welcome to one amazing thing about Malachite. Today we're talking with Sid Muthwani, co-founder and CEO, and Di Ling Holm, VP of product at Malachite, and they're going to be sharing one amazing thing about their platform with us today. Sid and Di, welcome to the show.

[00:00:35]>> Thanks for having us, Greg.

[00:00:36]>> Thanks. Yeah, thanks for having us.

[00:00:37]>> Yeah, looking forward to seeing this in action in a minute here. Before we do though, why don't why don't you start by giving a little background on both of yourselves and your roles at Malachite?

[00:00:47]>> Yeah, absolutely. So, I'm the co-founder and CEO of Malachite.

[00:00:52]My my co-founder and I came from Spotify where we built the underlying personalization starting in 2015 with Discover Weekly. And that has that underlying technology evolved into what you know of Spotify today as a personalization company. Everything is personalized on Spotify and it's all curated experiences for each user.

[00:01:12]That underlying technology and discover discovery is important is because discovery equals LTV. It's all about diversifying and expanding your taste and preferences overall. And the reality is with over a billion products that span music, audiobooks, and podcasts, it's hard to predict your intent, right, Greg? You could be on your way to work in the morning and you could want old nostalgic favorites to get you pumped up ready for your day, but on your way home you could just want a podcast, right? Or you could go to go to the gym for a workout and you have that same pump up playlist that you rely on every time, but then you go home, shower, and while you're cooking in the background you could just want an audiobook, right? How do we pick and choose the experience for you when we have so many options and hundreds of millions and now 800 million users around the world, um, and understand those intents in real time. And And the big breakthrough at Spotify was uh generative but for user behavior, right? So, let's just start with language models really fast.

[00:02:07]Language models have trained on the corpuses corpus of sentences on the internet and how words co-occur in sentences. So, all that means is it looks at every word in a sentence, it hides the word in a sentence and looks at the word before and after to predict that word, and then it looks at the answer and it goes to the next. And it does that for every sentence on the internet. What Spotify and TikTok ended up doing was treating every session as a sentence. So, you open that application and there's a sequence of events in every session. There's a lot of metadata and you're interacting with the product.

[00:02:38]But when we train on all of those sessions, we now generally understand user intent and your interaction with the product and can predict your intent to high enough accuracy in just milliseconds. The best way to understand the technology is TikTok because at the end of the day there's a billion user-generated content and a billion users. And every time a user opens application, they pick one out of the billion options, right? And as you engage, it just keeps getting better and better, right? And that's why you stay on TikTok and Spotify as long as you do.

[00:03:05]So, what we're going to demo is what we now do in retail e-commerce. Uh my co-founder and I spent the last year and a half building out a product a platform around this product specifically designed for retail e-commerce where we come in and we plug into all of that session, that user behavior data. It's 90% of the data that sits under each and every one of these enterprises right now in retail e-commerce. The reality is that's just used for dashboarding. You like just think about Google analytic events. What we do is we come in, we plug in to that data and we train on all of it and remember it using what's called continuous learning, which overcomes batch technology, which is pretty much what's on the market right now. So, then when any visitor hits that website, we don't need any history, no signing in, no third-party cookies.

[00:03:50]We're able to do what's called cold start activation and generate that user vector for them, which represents their preference and intent and immediately personalize. And it can power search recommendations, PLP, category pages, email, you name it, right? And not only does it adapt and learn in the session, it persists across sessions as well. So, if they come back, if and when they come back, then you pick right back up where you left off or it's just a completely new journey and intent altogether. So, Dai, you can share a little bit more about your background. Dai was our first customer at fun.com Halloween costumes.

[00:04:22]>> He's taken over the product with my co-founder Ian and Max. So, yeah.

[00:04:26]>> Wow.

[00:04:26]>> Yeah.

[00:04:27]Yeah, so came from brand side probably two decades total, but 10 years over at fun.com and Halloween costumes.com, which even if you haven't purchased from them, if you purchased a Halloween costume on Amazon, it comes from Halloween costumes.com. Just kidding.

[00:04:43]>> [laughter] >> And yeah, we we have used kind of those standard rules-based recommendation engines in the past and did an AB test against what Malachite has and saw a significant improvement. It was a 31% increase in RPV, revenue per visitor. And so then now they've since then I've left and I'm now with Malachite. I loved the product so much and how it's changing the industry and was able to help influence the road map there and and I'm owning that now with them, but fun now has taken search and PLPs next for them. So, they're starting to add all the pieces of the puzzle and are seeing just just great results. So, it's it's amazing. Honestly, this technology. When I heard about it, I was like, wait, wait, what?

[00:05:29]Coming from a brand side, you you usually hear all these AI buzzwords, right? And it's hard to tell what's really real and the fact that they can explain how it all works on the back end and and the math behind it.

[00:05:41]Yeah, it can't say can't say more about it.

[00:05:45]I I joined I left and I joined them.

[00:05:47]That's how much I love it. But I can show kind of how it works. Uh >> Yeah, let's let's do it. Yeah, why don't you why don't you share an example?

[00:05:53]>> here.

[00:05:55]So one of our other customers is um called Grunt Workwear.

[00:05:58]And uh this is kind of this is one of the demos that we usually will pull up kind of showing side by side. This was what their site used to be like versus what it is now with Malachite. So uh moving into this space here, you can see we have a user at the top and they're coming from Google. They've searched uh comp toe waterproof. They're a new visitor. And this is what the site used to look like uh when they came to Bronze versus now understanding where this user is coming from, that they're new, they're in New York, uh what they searched on Google, we're personalizing uh the the actual page that they're landing on here.

[00:06:40]Now, if you're using a different user, let's say this guy uh coming from an Instagram ad, right? Depending on what was in that ad, right? She is seeing something different. Or even this guy is coming from an email, so an existing customer has multiple orders.

[00:06:55]We already know what that he's interested in.

[00:06:57]Um and since we're fall workwear, you can see it's instead of just showing boots, which is their main product, but he's a returning user, understanding more about him too, and personalizing based off of all that along with where he came from.

[00:07:10]Um Beyond this, so that's just your product listing page. So you have had search to the mix, right? Here uh understanding all of the misspellings get you actually what you're looking for.

[00:07:22]And then finally, on your PVP or product detail page, uh same product, different recommendations for this underneath this product. One, that you may also like, which most people are familiar with, which is just a whole bunch of very similar items, right? But not um anything new. So you're you're basically bouncing between them. Now, if you're showcasing something that's more like complete the look, complete the uniform in their case, um you're going to be increasing your AOV, and people will be less likely to bounce instead of going, "Oh my gosh, $164, everything is that price or higher. I'm out of here, right?" They're like, "Oh, I can build this entire look, right?" Now, behind the scenes, so back this up, this user, just to kind of set the the the page here, the visitor user ID is right here, and that's our user vector, we call it. Um this one is for the returning two orders, uh that email, and you can see that information's coming along with their page view into the session behavior. On the right is their adaptive profile, so this is what we're learning about the customer as they're coming through the website, and you'll see this change as well as the in-session behaviors as we go through. So, what's happening in if you imagine products on a on a map, all of your product catalog on a map, uh some products are more similar than other products. And what the vector does is it identifies how closely those are related and kind of groups them together, right? Um and there may be groupings like based off of color, so right now you're seeing um by category, but they could also be similar to color. So, this is almost like a a three-dimensional map, if you will. So, as I go through, now we're seeing a hover, if you look at the session behavior, two times on this shirt. Uh they searched cold weather work outfit under $200, right? So, it can understand that context. So, going through all of these steps that they're having on the way, they're dwelling on an item, they uh have an add uh to cart here, and then a bundle. So, as they went along the journey, so you're going this cold start when they first get here, they hit the PLP, they're showing this outerwear, the hoodie, and what they're interested in, they're searching that out outer cold cold weather work outfit. So, these products are all getting closer to the visitor, and that's was actually personalizing the page.

[00:09:39]So, as you're going through and you see all those products are shifting depending on what the user's intent and their behavior. And that's what you see on the pages here based off of what they're searching for and what they're clicking on and what they're hovering.

[00:09:52]All of those different user pieces.

[00:09:55]>> Yeah, yeah. Nice, nice. Well, yeah, cuz I mean, you know, most uh search results or merchandise is let's just say not that smart. Right, it's based on some very simple relational I mean, I've done a little bit of that, you know, like it's simple relational rules because probably because you had to you have to guess for what anyone anywhere might Yeah, so this is Yeah, this is amazing to see >> Yeah.

[00:10:24]>> No more customer also bought, customer also added to cart. Every customer is different, where they're coming from, what they're searching. All of those are very specific intents.

[00:10:32]Um so, yeah, being able to actually personalize based off of those instead of just what some other people had done previously.

[00:10:38]>> Yeah, yeah. Right. Well, and and I think, you know, one of the most annoying things is the you know, probably that one user already had three pairs of boots and, you know, the one scenario was going to just show them more boots. And, you know, no offense, but they looked relatively similar to each other. So, you know, so how many how many of the same kind of boots do you need?

[00:11:00]Although, I do have 100 black shirts, I think, but, you know, that that aside.

[00:11:05]Um >> And and you're a specific user and >> Right. Right.

[00:11:08]>> Malakai would identify he really enjoys black shirts. Let's keep showing them.

[00:11:12]>> Right. Fair enough. Yeah, yeah.

[00:11:14]>> And what the user vector can do is it can understand if you know what you're looking for, that that complete the look strategy is because the likelihood of that user wanting the boot in that situation is very high. So, let's prioritize a complimentary add-on. But, if you're looking for more diverse options, different colors, it's going to understand that and and change the recommendations accordingly.

[00:11:33]>> Yeah.

[00:11:33]>> Right.

[00:11:34]>> Yeah. Yeah. No, I I I love this. Yeah, and I mean it's it's I always I also think it's really interesting to be able to apply something from, you know, different in you know, from Spotify, you know, from that kind of world and apply that to e-commerce. So, you know, it's it's you know, in retrospect it sounds like a like a no-brainer, right? But I I know it's you know, that takes that's that's what innovation is. So, that's that's a >> Absolutely.

[00:12:00]>> I love it. I love it.

[00:12:01]>> It took us It took us a decade to do it.

[00:12:03]Lots of compute um and time resources and expertise, but but yes, uh you know, just I guess sitting with the level of scale that we were experiencing at Spotify allowed us to be able to bring this to retail e-commerce, right?

[00:12:17]>> It's a big mind shift for the merchandisers too that have been hand curating and pinning and setting up rules just to make sure that the products um that they're hoping sell are actually shown to the customer. Now, they see something like in recommendations and being a customer first, I saw this where it's like that I don't know if that's right, you know, or had a great example at Halloween costumes where somebody was on the pink Power Ranger page and they came to me and they were like, "These recommendations look horrible." And it was all pink costumes. They were like, "They should be superheroes, not the Power Rangers." And I was like, "Normally I'd say yes, but the data is proving otherwise. What did you click on before that?" And they were clicking on all these different pink costumes. And so, the intent is they don't care if it's a Power Ranger, they want a pink costume. So, they're showing We were showing all pink costumes. So, it it's a definite mind shift from that like hand curating the like complete the look and that kind of thing.

[00:13:13]>> Yeah, and I guess along those lines just what does, you know, cuz a lot of I think a lot of great technology enables the humans that used to do very manual tasks or let's just say repetitive tasks, you know, it enables them to do something better. So, you know, how does this What does the merchandiser get to do when some of this other stuff is kind of taken care of for them?

[00:13:37]>> Yeah. Well, we do still have rules available for folks because we understand sometimes there's a product miss and you have stale inventory you need to sell through and the only way to give it that visibility if there is no intent behind it is to boost it, right?

[00:13:50]Or to pin it. So, those are still available. So, merchandisers can still do that. But, um now they can focus their time on merchandising potentially emails, right?

[00:14:00]Uh or maybe maybe they can help on paid social side.

[00:14:05]Maybe you don't Maybe you're expanding your catalog or you're adding a new domain and you don't need to do another hire because now you have more bandwidth. Um maybe they can set more strategies. Um things to come with Malachite, for sure, some some smart agents that can recommend instead of pinning in and boosting based off of a particular product idea or category is RPV, right? So, let's boost based off of that or maybe our AOV has dipped. Um so, let's make sure that we're focused on that and setting rules that will help us increase our AOV, right? So, it's more around business outcomes versus pushing a particular product.

[00:14:41]>> Exactly.

[00:14:42]>> There's a lot of like it's it's able to balance many objectives across the teams for each user as it's making data-driven decisions, right? In milliseconds. And so, if you're looking to promote a new product line or maybe suppress a a product that's actually suddenly low on inventory, right? It's able to we we our merchandising tool gives the team the sliders in order to influence the personalization for each user and then that user vector is considering all of those different intents of the teams when making a decision of what to actually show the user itself, right?

[00:15:16]So, there's more control in your PPC. It's It's like the merchandiser's there in the session, moving the shelves around, right? And as they And right now all they get to do is predefine rules and everyone just sees the same thing. It's kind of like walking in the TJ Maxx and it's just it's chaos, right?

[00:15:34]Like chaos and and now you walk into TJ Maxx and the the shelves are coming up to you and they're they're optimized for each person, right?

[00:15:43]>> Yeah, it's like a personal shopper instead.

[00:15:46]>> Yeah, yeah, love it. Well, hey, thanks so much for showing this today. For those that want to learn more, what's the what's the best way for them to to do that?

[00:15:54]>> Yeah, our website, we offer free demos.

[00:15:57]We're happy to be reached that way or you can find me at Sid Malloy Pike.

[00:16:02]>> Wonderful, wonderful. Well, again, I'd like to thank Sid Malloy Wani, co-founder and CEO, and Di Lingholm, VP of product at Malachite for joining the show. You can learn more about Sid, Di, and Malachite by following the links in the show notes.

[00:16:15]>> [music]