Hi and welcome to Hello Retail Conversations. We have a really special episode today. We're doing something different today. We have an internal guest, a colleague of mine, Sarah Miguel Conain. We're doing this special episode because we just launched a new product just this week. It's a super special week at Hello Retail. We're very, very excited to push a new product called Product Agents out to the market. We felt it was timely to have a conversation about the data science behind the new product. There's a ton going on sort of underneath the hood in product agents when it comes to the product intelligence data we use and a lot of proprietary AI that is involved to make the product do what it does. So again, welcome to you Sarah. This is going to be a special episode where we sort of present more and talk more about what's going on in data science in terms of the new product called Product Agents. So yeah, welcome. Thank you for inviting me. It's a very cool project. Gladly to talk about it. Nice. Can you just start by telling us a little bit more about your role? What do you do on a normal Friday morning? Well, I'm a data scientist here at Hello Retail. I work in the research and development department and depending on the day where I work on different projects or different reports. But I said broadly speaking, I would say I look 50% into the past, what has happened and why. And the other 50% is deciding on the future. So yeah, like in e-commerce, there's big events such as Black Friday or Christmas or sales. And I try and figure out what helps, what have customers interacted with, what do users buy, why, when, and try and figure out why. And also, but also things that relate to our solutions. Like for example, we just launched a retail media recently. So trying to study the different campaigns, which worked, which didn't, where do we have to improve. Then that's a bit the past I keep on looking at. And then with all of that knowledge, then we figure out what we need to do in the, in the near future. So I never live in the present. No. You have a pretty substantial academic background as well. Can you talk just a little bit about that? Where you, what, what you were up to before you joined Hello Retail? Um, so, so I have a background in mathematics. I started in Spain. I'm, I'm from Spain. And there, yeah, I, I was more focused in the math part of it. I was not so much into programming there, but then um, I had to do an internship in a, in a company and that's where, uh, in, in Madrid. And that's where I started in this AI world. And I got more and more into it. And, um, so I then went into this, uh, program. It was like a scholarship and I spent one year in Japan working in a research center. And that was my big encounter with, um, uh, artificial intelligence and machine learning, deep learning. We used to call it more back then. And, um, yeah, so I decided that I wanted to focus more on that path. And, um, I came to Denmark when I was 18 and I was fooled by all the weather. I got like, uh, 30 degrees. It was incredible. And I said, I really want to live here. So, um, I saw that DTU had this, uh, business analytics, which was, uh, which would let me focus on artificial intelligence, but more applied to business. And, um, so yeah, that's what I did. I combined it with being a research, um, teacher assistant for one of the subjects. I did a really cool thesis about, um, fashion bot I called it. And, um, yeah, that's, uh, that's more or less my knowledge. I worked in another company before doing data science and now I'm here at HelloRito. Nice. Uh, there's obviously a ton of, um, AI chatter in e-commerce and in all of tech, um, at the moment. Um, I, I always felt myself that HelloRito is, is quite special when it comes to the, um, to how advanced and our proprietary AI is. Um, so, uh, which is again, why we wanted to, to talk to you, um, today about what goes on inside, um, the belly of, uh, product and intelligence and, and our new product, um, product agents. But sort of before we get to that, um, I wanted to hear your view on AI and e-commerce as a whole. How do you see the, the field? How has it emerged or where do you see it going? What are your thoughts? Um, I think e-commerce has been, uh, one of the sectors that applies AI as fast as possible. Um, we've been one of the first ones to add, uh, re, um, recommendation systems or, um, the thing is, AI is a term that started recently before we used to call it collaborative filtering or machine learning. So it had, it came in different names, but it was, it's under the same topic. So, uh, so yes, I think e-commerce has, uh, very soon leverage all this recommendation system or search or, um, um, not just so also in stock and in inventory and logistics. Uh, they, they've been one of, I'm not going to say the first one to apply cause I don't know who was there first, but, um, um, but yeah, they've always been very keen on, on leveraging everything they could. And, um, and now we can see it with, uh, generative AI, all these examples I said were more predictive AI trying to figure out what is going to happen and who is going to buy what and what's going to be the next trend. That's more predictive AI. Whereas now we're seeing like a boom in generative AI, which is like, uh, chat GPT or Sora, all of these ones. And, uh, I think e-commerce is, um, I can see it in Hello Retail, the retail media product we launched and product agents now it's, um, has, uh, combined both. Uh, it's not competition between predictive and generative. I think the combination of both is, uh, is, uh, is what is going to give more interesting products and solutions. So, um, so yeah, I think, uh, What about if you talk with other data scientists? Well, what's, uh, what's the hot topic at the moment when it comes to a sort of use of AI and e-commerce, uh, on a, on a deeper level and it's totally okay, uh, to sort of geek out and, uh, talk about some of the more advanced concepts that, um, you're applying? I think, um, if I talk to other data science is, um, it's a science that you can apply to many places. So I have a bunch of friends from my masters or my, uh, bachelors, and we all work in different sectors. And I think one of the things we like the most is, um, how fast we can go from an idea to a product. So, uh, before you would have to spend a lot more time in the code itself and, and debugging and fixing. And whereas now from an idea to a solution is, is way faster and is way more interactive, I would say. So in that sense, we're happy about that part. And, um, when it comes to e-commerce, um, I think we do feel that everything goes fast. So you have, it's like a race and, uh, you launch something and you have to be thinking about the next one. It's kind of like a breathtaking race and obviously everyone wants to be at the top. So, um, so yeah, it's nice to be fast, but, um, but it also comes with, um, ideas must come quick and you have to be more creative and you have to be willing to try more things faster. But what if you went even deeper, uh, than that when it comes to sort of, um, uh, more substantial or sort of, um, hardcore aspects of, uh, the, the kind of AI that you're, um, applying in your, in your role here at Hello Retail, what, um, what's, what's the most interesting part right now? Adding AI right now is a must. It's not, um, I should or, uh, maybe it's, it's absolutely a must. So, um, so, um, so you ha we had to learn how to work with it. Uh, there's, um, there's a lot of testing, verifying, checking that the results are correct. There's a, there's, um, yes, it helps, it improves, but there's always like a 5% or a 10% that doesn't work. What do you do with that 10%? Um, how do you find out that what percentage is not, uh, accurate enough? Where do you put the trade-off? Um, all the data that you can gather now, there's also a big, uh, now that the part of the, of the models itself, it's, is way faster and way more, uh, approachable, then there's all the part of, um, gathering all the data and understanding it and the analysis you can do, you can do way further, deeper analysis and. No, but I can relate to that from, from the marketing side, um, that I work in. I, um, suddenly look at code, uh, all day long, basically invite code, vibe coding all sorts of stuff for, for us in marketing to sort of automate, uh, more of what we do. What's been super fascinating for me lately is that I don't only get to 80%, I actually get to a hundred percent with the projects I'm, I'm doing. So that's been a profound change when it comes to AI on the marketing side that all of a sudden it's not just interesting concepts or a little bit of help here and there. We can come from sort of idea to finished, uh, new piece of, of software that we use in marketing, uh, completely to a hundred percent sort of production grade. We can use it to build websites and so forth. So I, I guess it's sort of the same that it's come to that maturity, uh, um, in AI where you, where, where it gets the last 20% as well. Is that correct? Yes. I think, yeah, that's, uh, I completely agree with that. And also before it was more like you used AI to verify what you're doing and now you kind of verify what AI does for you. So, uh, so there has been a change there too in the way we work with it. It's only been there for not so long and the way we interact with it keeps changing. Yeah. So, um, cool. Um, we need to talk about product intelligence, uh, because it's, uh, in so many ways, the, the, the sort of the foundation of product agents. Um, Hello Retail, we've spent, I don't know, five years or so building product intelligence or, uh, proprietary AI. Uh, so if you were to sort of explain that to someone who doesn't know too much about it, what, what is product intelligence? Product intelligence is about knowing the behavior around a product. So, um, so this comes from when you're working in e-commerce, you're working with the customers and you're working with users. And generally speaking, there is a, there is a percentage of people that don't mind to be tracked and, uh, um, and they allow cookies. However, there is a part that do not want their data shared. So, but they, but customers still want personalization. So how do we work around that? Well, if I, if I don't know who you are, I need to know what you buy. They don't, they still want personalization, but they don't want to say who the person is. Then, uh, then what we do is, uh, we say, okay, then what's more impersonal than, uh, than a bottle of water. It's, uh, the product itself is, is not a person. So we can study the relation between the different products and how they correlate between them, how they interact between them, who's brought together with what, who is not bought together with who. Um, so in that sense, if, um, if you think in the end about e-commerce, it's a big matrix of, uh, of people, users in the columns and, uh, products in the rows. So, um, sometimes we're not able to, to track down the users. So instead of looking at what you've bought and what will you buy, we look at the product itself. And if you bought this product, then you're most likely to buy this others that also correspond. So, um, so yeah, pro I think product intelligence is all about product behavior and how products interact with each other. So when a user comes in, then we don't need to know who you are. We just need to know what products you interact with. And we have, we already have a feeling we don't need to know who you are. We just need to know what is it you want. Um, when, when I look at our product intelligence data, I'm super fascinated sometimes by how a seemingly similar, uh, product, uh, has very, very different purchasing patterns than another adjacent, uh, product. Um, so we, we know an awful lot in the data about the purchasing patterns for every single, uh, product and, and their categories. And I, I always just felt that it's a total gold mine for, um, for e-commerce stores when it comes to knowing when to promote a product or knowing what products to recommend and so forth. Um, so how, how does all of that sort of function behind the scenes and in data science? Can you sort of reveal bits and pieces about, uh, what, what's actually going on for us to, um, have that data analyze it and, and sort of make it useful? When we talk about product behavior is the machinery behind it is creating a vector space. If you, a product has a number and for that shop, this product is number 10, 10. Um, but in our vector system, this product is a point in our database. Our vector, our vector system has like a more than 700 dimensions. So it's a bit difficult to imagine, but, um, but what we do is represent this product. It's not anymore product number 10 for us is a black t-shirt for men to do sports. Um, so if it's part of a cluster this way, when, um, for example, this, this, uh, shop creates, um, has a new product, product number 20, and is also a black t-shirt that does, um, for men who run, then it will be in the same cluster. This way, all the information we had from product 10 can be shared with product 20 because they have the same characteristics. They, they, they are in the same vector space, the same spot in our vector space. So, uh, for example, one of the biggest problems of looking at user behavior is the cold start is what we call cold start. And is that when a new customer comes in, we know nothing about it. It's like when you start dating someone, like, uh, don't know if, uh, they're sporty, if they, uh, have, um, brothers, I don't know, like, you know nothing about that person and you have to figure it out. Well, this is what happens with many web shots that we work with, like new customers come in frequently and, uh, and we know nothing about them. However, this does not happen with, uh, with, uh, products. If you add a new product to your catalog, we know who is similar with, and if you're similar to another product, then all the information we have with, uh, with this other product, you can share it with your new one. So this way, when a new customer comes in, we look with just a few clicks of use, we see what products they're interested in. And these products, they'll have relationships, they'll have correlation. So we're moving that we know from exactly the collective data sets and thousands of stores and so forth. Exactly. From that vector space. So all the connections that, um, that product has with other, that other people have made, then, uh, we can apply them to you. So we don't really need to know who you are. We just need to know, we just need to know the product you look at and we can figure out a lot of information from you. So we can still personalize, uh, your shopping, uh, interaction, but we don't have to know who, who are you. Yeah, that makes sense. Another thing is that we see that our web shops, um, is 30% of the products that bring 80% of the revenue. So, um, so there are... Across the board for all... Across, yeah, across all our customers. ...that we, that we sort of have data from. Yeah, it is true that we have a very broad selection of customers in the sense there's people who sell boats and people who sell clothes. So, so we do have a broad overview of, uh, but yeah, the general numbers. Yeah, but in general numbers, 30% of the products being bring 70% of the revenue. And another number that surprised us was, um, 50% of the products that are active now in six months, they will no longer be active because the inventory keeps changing so much. So all this information from the products that you gained when you had it in your store will be lost if you're not able to find the relationship between what you had and what you bring new. So it's not just about, um, we don't want to know who the user is because of data tracking and it's not just about the data privacy why we choose product intelligence. It's also because product intelligence and understanding the product behavior behind it. It's, uh, is, it's just, it gives so much knowledge to what's going to come. And a hundred percent, you won't lose all the information you gained before. Yeah. Yeah. That makes sense. Something that really fascinates me about our new product, uh, called product agents is that we're sort of possible. We're, we're, um, all of a sudden able to produce these brand new types of emails that you are just not able to produce in any other way in any other system. It's a really sort of a new emerging innovative, uh, solution in, in many ways. So, um, and we're, we've built a product that will sort of, um, tie into your email, um, provider like a Klaviyo for starters, um, and it'll sort of enable you to send one-to-one personalized emails for the individual, um, based on what they've been looking at, what they've been shopping in the past that we're sort of enriching with all of our products intelligence data. So I want to, um, explore what, what is going on in, uh, from a data science, um, point of view in the new product, uh, called product agents. What are some of the sort of the key AI features that you've been involved with building? So for me, this, uh, project agents is like the perfect mix of a predictive AI and the generative AI. So, um, in the predictive AI is all the product intelligence. The, we know what products the customers wanna is gonna want and when they want to buy it. And now with all this generative AI, all those LLMs, Claude, ChatGPT, or all of these ones, we can create the content personalized to them. So, so we're mixing all the knowledge we gain from the product intelligence and the relations and what is brought together with what and the upsells and the, and how our products related alternatives to each other with the generative AI that is coming strongly. So, um, so we mix them together in, uh, in product agents. So as a data science is nice to see, uh, a mix of both worlds. Yeah. What about the, the product intelligence part that we, we talked about, where does that sort of come into play in product agents? Um, I think, uh, thanks to our product intelligence, we're able to give way more personalized emails, but also more spot on messages in the sense, um, what we talked about the inventory that, uh, in fifth, in six months, 50% of the inventory will have changed. So all this, uh, information from the products is lost. So if someone has bought, um, a mascara and in six months, this mascara is not longer available. It doesn't mean that the customer itself is not going to buy it anymore. They want to buy a different one. So with this, um, alternative, us knowing the alternative and what other things we can suggest them, then we can still make the customer buy mascara again in our shop. Or another case is, um, I think that's the one I like the most is the price drop. Sometimes you look at your product and you're not willing to pay that price, but I would love it if, um, they make a sale or an outlet and it dropped 20% in price, then that will be the threshold I would be likely to pay. So if I get an email saying the product you viewed has dropped in price, then I would more, I would, is more likely that I see it rather than if I go back again to that web shop and see that, uh, I view that product and now it decreased in price. I'm not going to do that. I want, I want an email that makes my life a bit easier. Yeah. Yeah. That man, that makes a lot of sense. I think one thing that fascinates me personally is that there's so much hype around, um, uh, shopping agents, um, at the moment or agentic commerce assets sometimes called, but it's all very focused. A lot of it in a way on, on the shopper and sort of helping them or giving them agents, but we're actually doing something different. We're giving the merchants agents that, that they can use to improve the, the overall sort of shopping experience for, for their shoppers. So it's sort of like merchant side agentic commerce, if, if you will. I know that you've been involved with things like, um, adjusting tone of voice per language, um, which is something I wanted to talk about. I found that super fascinating to, to hear about. So what the product in, in the new product agents in the, in our dashboard, uh, you as a customer can sort of select different, uh, tones of voice to adjust, um, the, the email content that is being produced. Obviously e-commerce brands will be super, super, uh, keen to send out messages that don't have, um, sort of crappy AI slob content, but, um, messages that are really to the point in a nice language, that fits their overall brand, uh, guidelines and tone of voice and everything. Um, but what I know that you sort of noticed that, um, selecting friendly means something different in different languages, for example. So can you talk more about what that was all about? What, what went on there? Yes. Uh, the challenge behind it was that we have a, a broad, um, selection of customers. So we have, uh, in different countries. Yeah. In different countries. So when we create emails, obviously we wanted the tone of voice to align. So if you're a medical company, then you want to sound trustworthy. But if you're, um, if you're selling games, then, uh, you might want to sound way more friendly and approachable. So there was a, there had to be a tone alignment. What I realized, um, there also, the, the other challenge was the languages. And I thought, okay, then it's just like, say this in a friendly version in the Spanish language. It doesn't work that way. Cause a friendly, a friendly Spanish person is not the same as a friendly German. For example, the way we speak or the, or the jokes or the, um, or the slang there is in different languages is way more different in, if I compare Spanish to English to Danish or, or even the formal one, there's, um, in Spanish, we don't really use the, the second, the pronoun for your highness. No, we don't use that, but there's other languages where, where they do. So, so it's not just about, this is the tone. This is the language. It's kind of a matrix in the end cause, um, what does friendly in Spanish mean? What does friendly in German? What does formal in Swedish? There's, there's a whole bunch of combinations that we had to take into account. That's super interesting. What, what kind of data science work is going on to actually make that happen. So, um, so it's generative AI that's creating the, the messages itself, but then there's, uh, there's lots of verification and, and quality testing. So, um, so manually test, manually checking what works, what doesn't, what are the problems? Yes. But then when you're talking about creating 1000 emails at a time, it's not, it's not possible to be looking at every single, every single email in every single language. I'm not, I'm not that, uh, I'm not that smart. So in that sense, um, we had to create, uh, some verifications, some, uh, um, when it comes to repeating words, uh, that was one of the problems or, um, sometimes when you want to make it friendly, it's too childish that we also don't want. Uh, sometimes when it was formal, it was too formal towards legal terms and, and that we also don't want. So there's been lots of tweaks and, and little changes that, and subtle changes that made, um, it was like a snowball because the more restrictions you put to the, how the language has to sound and how the, the phrases should be constructed and the vocabulary you have to use, then there's less space for creativity in the sense. It doesn't want to be so friendly. It doesn't want to be so formal because it also has to maintain all the other rules you have. So there's a bit of a trade-off going on as in everything in life, but, um, but yeah, verifying all of that quality testing, um, different languages that I don't speak. Um, sure. That's, uh, that's been part of the challenge. Yeah. What about the whole sort of on, um, um, um, the agent side? Um, there's a ton of, uh, work going on to sort of select the right product to promote to the individual, um, where we obviously apply product intelligence. Um, could you maybe explain a little bit more about, um, how we are now applying product intelligence in that whole process of, um, we're discovering intent from an, um, a returning buyer. Um, we want to maximize the revenue that we, uh, sort of allow the, the, um, the e-commerce business to gain from selling again and again. What, what's actually going on behind the scenes in terms of how do we find those products that we should sort of pair with the individual shopper? As we know, there's different products and there's different, uh, users. So, um, product intelligence is where it comes in handy and, uh, and helps us figure out who wants what and when. So, um, for that, you really need to understand the product you have and how it relates to the rest of products you have in your inventory. So, um, for example, I would say there's two lines within the products. Are they consumable products or are they, uh, are they a one-off? For example, if you buy a sofa, then more is, you don't want to buy another sofa in a week. However, if you buy a shampoo, then you're more likely to buy it. So understanding if you're a replenishable product or not, that would be one of the, um, I would say the first, um, division I would start doing, uh, and depending on which one it is, then you activate different triggers. So, um, so if you're buying, um, food for your dog, then most likely you're going to keep on buying it. Maybe you change the flavor. Maybe you change the packet size, but for example, packet size was, uh, package size was, uh, one of, um, the things that, um, product intelligence help us with. Because if you buy a one kilo one, it's not the same as buying two kilos. So that helps us understand when do we have to buy a, send an email for replenishment of that product. Uh, so that's more the replenishment size, but then for the, for the other products that you also want to sell, then showing alternatives. Some people go in, but they, but they didn't buy in the end because they're checking different options as everyone should. But, um, you found a similar product that they should be interested. Maybe, maybe, you know, by the products they viewed or clicked that they're more into colorful, colorful clothes. And they looked at this black sneakers. And then you say, but you didn't manage to see this green ones I have. So, so because you know that you're more into, um, colors and, and you want to some shoes, then this is where product intelligence makes the relationship and sends you the alternative, uh, to what you have seen. So, um, so it's more about product intelligence helps us see what does the customer actually want and when does it actually want it. Yeah. And then we can do that without sort of asking you to set up all sorts of convoluted flows Exactly. In your ESP or, um, whatever you're used to, to build the, the, the flows. Exactly. And then also there's a, there's a part of, um, maybe you clicked and you viewed and you bought, and there's some products that are replenished. And then there's some others that you would not replenish. Then there's a mix of, uh, emails you're going to be sending. And that we also, that we also fine tune. You don't, no one wants to receive 20 messages from the same workshop every day. So, um, knowing when to, when to send them and exactly what emails to send, then that's, uh, and filter what emails you're going to send. That's also a product agent thing. Yeah. I think, um, uh, I said before that we're, we're essentially building, uh, agents for the merchants, but, but they are also serving the, the shopper in the end. The shopper just wouldn't know that they have had an sort of agent run for them in the background. Um, so where, where do you see it? I mean, where, who do the agents belong to in this case? Is it the merchants or the shoppers, or do they sit in between them? I think the agents help the shop understand what does the costumer want, but it also helps the shopper to find the products they actually need. And I do see a future here in, um, um, they talking about all this hyper-personalization. And I think before, if before it was the, the shopper who had to go to the shop and buy, now he can go in, he, she can go into the phone and, uh, and shop online. So I think the shop is getting closer to the shopper. So, um, now it's, uh, now we're in a point where you as a, as a customer, you have to go into the online shop and see what you want to buy. But I think the near future is more the shop coming to you, knowing what you want to buy and making it available and easy to see for you. If, um, for example, if you were to explain this to a friend of yours, um, how, how will this matter to a shopper on online? Um, if they're, you know, buying stuff online, uh, this new product agents thing, why will that make a difference for, for you as a shopper? So before when you wanted something as a shopper, you had to go to the shop. Whereas now you literally have the shop in your phone. And I think this next step, if you mix it with hyper-personalization is, is the shop literally coming to you and selling you, this is what you want. This is the price we have it. And this is exactly when you want it. And it's going to make, it's going to make it even easier for you, this shopping experience. Yeah. So you're getting to a point where you won't just get, um, like, um, generic product recommendations in your email marketing, but ones that sort of fit your preferences exactly. And at the right time and so forth. Exactly. Yeah. You don't want to be browsing in order to find you more or less always want the same type of clothes. You want the same type of products, maybe changing them a bit. So, so this pie, um, this shop coming to you directly with what you know, we'll save you time. We'll make it easier and it'll be a click. Yeah, that makes sense. Great. We're serving a huge variety of different stores in different countries and different sectors and industries and so forth. Um, so that obviously means that, that different e-commerce businesses will want to communicate in quite different ways, depending on what types of products they sell and what type of company they are, what kind of brand they want to communicate and so forth. How, how are we sort of handling that in emails, in product agents, um, all of that variation that they will need? So, um, so yes, we talked a lot about personalizing the email for the user based on what they bought, what they've seen, but, um, but there's also personalization from the brand. And I think that's important too. It's not just, it's a one-to-one communication. It's not just telling users is also from the brand that it's important to add. So, um, so yes, we align with a tone of voice. We align with the language we're talking about, but also our customers can align their brand voice into it. They can, they can say what it means for them. Be friendly. What, what, um, messages they want to get into their emails so that it comes from them to their users, not, not a general friendly message. Yeah, that makes sense. I think a lot of brands will just be super, super sort of, uh, keen to, uh, retain the brand, um, uh, sort of promise and tone of voice and so forth. So we, I think what matters to me as a marketer is that we're doing absolutely everything we can in this product to, um, let, um, our, our customers communicate in the exact same way as they've always done and what they've used to build up their, uh, e-commerce brands over time. Cool. Um, before we, we wrap up, um, I want to ask you a question about your sort of your own personal favorite e-commerce business. Is there a store that really stands out for you that's doing something interesting? Yeah. Um, I'm going to go close to home. It's called S'more is, uh, is a running brand in Spain. It recently just started and they don't even have a shop. So what we talked about the shop before you had to go to a shop and now it's online in your phone. Well, I can see more and more e-commerce directly starting with an online web. You don't have to have a shop now. And, um, the way this company started was an influencer, a girl who runs a lot and who does a lot of sport. And, um, and instead of, um, she created a community of people that are like her. So I don't really need to know exactly who you are, but I know the products you're going to like. And she created like this, uh, fashionable running brand that I think, uh, she found a cluster of people and, uh, she did the products for that cluster of people. And I just thought it was a, it was a nice way to get yourself into the market. Yeah. Yeah. That's great. We've had other guests on, um, um, Hello Retail Conversations that talked about the importance of the, the sort of the DNA of the store that you really differentiate to that community you're serving, uh, and not don't just be get another generic online store. Exactly. So in product agents is something we take care of in the sense we want to make sure it comes from you, not from, uh, from some generic, uh, store you said. So, um, so yeah, I think maintaining the DNA in this world that is getting broader and broader, it's, uh, it's still important. So, um, so yeah. Yeah. All right. Um, thank you so much. It was super interesting to hear a bit more or a lot more actually about, um, what's going on behind the scenes of our new product launch. And, um, um, uh, thanks a lot for, for joining and talking about your daily work as well as everything that you've been involved with in, uh, the, the sort of the development of product agents. Thanks a lot for inviting. And, uh, thank you for, for watching. Um, so we'll be back next time, uh, with a conversation with an external guest, but again, today we wanted to do a bit of a special, uh, because of the launch of product agents. Um, so thanks again, Sarah, for that.