# AI for Ecommerce. You can just ask. > AI for ecommerce. You can just ask. Hello Retail has spent years learning how products behave. Which ones sell on a Tuesday. Which ones people come back for. Which ones sell fine without a discount. That knowledge used to sit behind a dashboard. Now you can ask for it from Claude or ChatGPT. Here’s the program at a glance: Ask Claude how your store is performing Fix zero result searches without opening the dashboard Find products to feature in your next campaign Update the search template without writing code Your hosts: Philip Andersen CPTO Julius Altenburger Senior Product Manager Jacob Oettinger Chief Software Architect Recording will be shared after the webinar. - **Category**: Webinars - **Duration**: 40m 20s - **Published**: October 20, 2026 - **Views**: 28 - **Watch**: [https://helloretail.com/en/resources/videos/webinars/ai-for-ecommerce-you-can-just-ask-1/](https://helloretail.com/en/resources/videos/webinars/ai-for-ecommerce-you-can-just-ask-1/) --- ## Description AI for ecommerce. You can just ask.Hello Retail has spent years learning how products behave. Which ones sell on a Tuesday. Which ones people come back for. Which ones sell fine without a discount. That knowledge used to sit behind a dashboard. Now you can ask for it from Claude or ChatGPT. Here’s the program at a glance: Ask Claude how your store is performingFix zero result searches without opening the dashboardFind products to feature in your next campaignUpdate the search template without writing code Your hosts:Philip Andersen CPTOJulius Altenburger  Senior Product ManagerJacob Oettinger Chief Software Architect Recording will be shared after the webinar.   --- ## Transcript Hello and welcome to this Hello Retail webinar where we'll talk about MCPs and how you can do automations using Hello Retail's new MCPs. I'm joined here by Julius, our Senior Product Manager, and Jacob, who's our Lead Architect, to talk about how you can automate your webshop to get the most out of it. This is a pre-recorded session, but we will have a live Q&A at the end of it. So, add all your questions and we'll try to answer them as good as possible. How to ask the questions, I'll let you get into, Julius. Yes, so while we are going to be talking for the next few minutes, you have the chance to ask all of the questions that you have. You can add them to the right sidebar on your right side, where you will find the questions tab, where you can write in everything we have, and we will make sure to answer all of your questions in the end. And just before we get started with all of the content, also feel free to write in the chat from where in the world you are tuning in right now. We are really interested in hearing from where all of our webinar participants are from. So, without further ado, I would say let's go into it and let's talk about MCPs. Perfect. Thank you so much, Julius. MCP is kind of a technical term that got popular here a few years ago, maybe, max Max. Jacob, what is an MCP, just in layman terms? Yeah, you could say that an MCP is a way for an agent, an AI tool to interact with your product. So, where, as an example, if you were to go and adjust our search synonyms, you would log into Hello Retail, you would go into the UI and do whatever you need to do. So, if an agent needs to do the same, the MCPs are an optimized way for them to do that. So, it's us exposing tools to agents to allow them to work efficiently within our platform. And it kind of includes documentation for the agent as well, so it instructs it at the same time on how to use our tools. Super. So, does this replace APIs for all the developers out there, or how do you see that? No, no, it doesn't. They are related. It typically exposes the same types of things, but the MCPs should be more seen as a UI, but for the agent, whereas the API is something you use if you are developing some software that needs to interact with some other software. Yeah. So, that means that building automations will be much easier with MCPs, I assume. Yeah. Yeah, exactly. If you are building some tool that is used by you yourself to automate some process you have in your work life, then MCPs are an obvious choice for that. So, that you can use your local agent, like Claude or ChatGPT, and let that work through the MCPs and doing stuff for you. Whereas the API would be something that is built into some product on your end and interacts with our end. At Hello Retail, we have been building about 160 different toolings for this. And toolings are not like a measure of quality in any ways, but I think one thing that is really cool is that for a lot of people out there, when they think of MCPs, they think about pulling data out and just do analysis on it. But what you guys have built has also been a way for customers to update stuff. So, update the synonyms and do all this kind of stuff, which I think is super exciting. Julius, I know you have some examples to showcase. So, yeah, let's take it away. Okay, great. So, as an e-commerce manager, it's always a few different questions that you ask. It is mainly, how are we doing right now? What might be broken? And what can we do next to also improve our setup and do more? Because obviously, what we want is for more people to buy in our stores. So, for the next few minutes, I'm going to be an e-commerce manager running you through a few of the examples of how you can use the MCP to make your life easier. In general, all of the tools that the Hello Retail MCP has are built in a way that before something is edited, it will ask you for your permission on editing something. And in most of the cases, it will also just create a draft. So, for example, if it's editing a design on your actual live page that you can then publish yourself. But let's get right into it. One of the things that I so far did every month is create a performance report. For that performance report, I needed to go into different places, get different data points, put them all together in a joint report to deliver to my company. Now, I can easily do all of that within the MCP and it can help me analyze the data in the right way. So, what I'm going to do is I'm going to go into my favorite AI tool, in this case, Claude, and I'm going to write, Please create a performance report for the months, August and September, separating out numbers for Search, Recommendations, Pages, and Product Agents. So, now that I've sent this off, it will analyze the information from the different solutions and will come back to me with a comprehensive report of what it has found. The last part here is also important in the sense that it separates out the different solutions so that parts where both or multiple have been involved are separated out clearly by revenue and traffic. But while we wait, let's maybe talk a little bit further about MCPs. Sure. So, right now, I know you love your Claude co-work and I know there's also other preferences out there. Jacob, you built a lot of these MCPs and the infrastructure around it. So, what kind of platforms do we support at the moment? Yeah. So, the MCPs themselves, that's an open standard. So, in principle, we support anything that supports those standards. We're using Claude a lot and it works well in both Claude Cowork and in Claude Code. It works well in ChatGPT and in CODEX. Currently, it does not work in Copilot, but it's something we are working on adding. So, we'll get that fixed soon. Yeah. Perfect. And if people want to get started, where do they go to get started and set up like Julius, so they can do whatever Julius just showed? Yeah. So, I think the best thing is to start by going to our documentation at developer.helloretail.com slash MCP where you can have a look. Super. Yes. So, now we're almost there. We have been seeing that it has been working on creating the report for us and is actually already assessing quite a few different things that might be relevant for us. So, we'll give it a few more seconds and then hopefully we will have the report ready for us. So, it does take time, like you said. So, luckily you can spin up multiple chats at once to do different questions. I assume like this kind of, like this is something you want to ask every once in a while. Is a way for me or for you or our viewers not to having to do this manually every Monday, is a way to do, automate this in a way? Absolutely. Absolutely. So, one of the beautiful things about this is that you can automate these kind of reports or these kind of things to run on a fixed schedule. So, let's say you want this report every week, every month, and then you can just create a schedule within Cloud or chat GPT and it will automatically present the full report that you've set up for you. And I can also show that in a minute of how you do that as well. So, now we can actually see that we have some data in this performance report. And it's actually pretty comprehensive showing us that September has beat August on every of the site features. We can see the split by all of the information that we have on search. It even goes down into what have the top searches been, how they have developed, and quite deep into where do our shoppers actually go when they are interacting with our product on the different channels. It also gives me a way to know, okay, what are the things that we should do next? And I think what we're going to do is we're going to go ahead and do the first thing that is on the list here, which is fixing the 10 dead-end searches that it has found. Before we dive into the example of search here, we want to look at how to create this on a schedule. So, what we're going to do is we're going to take our prompt that we have been using before, and we're going to create a new task for Cloud Cowork in this case. And what I'm going to do is I'm going to put this prompt in, and I'm going to say, please run this report every first of the month for the two previous months. And what is going to happen now, it is that it will create a schedule task, which will know, okay, this is the month right now. This is the two previous months, so it will have these kind of placeholders in the prompt. And once I accept and set it live, it will be ready for me to build this report. So now I, of course, need to tell it that it is okay to run all of these things on a schedule. And I need to accept a few things here for it to run successfully. So, all these MCPs you're showing us pulling data from different systems and all that, are they new products? Do our customers have to buy them? Like, besides the setup that Jacob just talked about, anything else they should be aware of, or are they just good to go? They are just good to go. It's included within their subscription, and they have access to everything that they have bought and all products that they are using, with all of the data that they otherwise could also see in the dashboard of HelloRetail. One of the strengths of the MCP ecosystem is that while we expose a good set of tools for interacting with our systems, it is very nice that other tools also expose their capabilities. So, when you have an agent working in the middle of all this, and it has access to multiple systems, it will be able to combine those systems, and you can get even more value out of it. You could imagine pulling data from some analytics system and using that to affect how you configure products in HelloRetail or examples like that. There are really a lot of opportunities for just being more efficient in your work with your platform and the tools you use. So, basically, one idea could be that if you are sitting and doing a new campaign here for Black Friday, you can use MCP for Shopify to lower the cost of your products in the inventory. And then, at the same time, you could also ensure that those products are boosted in the search. You could create, I think you'll show it a little bit later, how you can do batches for it. So, it's really nice that how you can have one query, kick out work from multiple systems to automate across systems. And I think I agree that's some really powerful stuff you can do there. Julius, you had some other examples as well to show. I wanted to tell you a little bit about all of the possibilities that you have using this within search. So, what we're going to want to look at now is to see about synonyms. Because what we don't want is that a customer ends up on searching for something and then not finding what they're searching for. And there could be different kinds of situations, but sometimes it can be that they misspell something totally wrong if they write a word in a different language. And you, with the MCP, have an even easier way of looking at it and fixing it. So, what we're going to ask here is, please show me all searches that came out with zero results. And tell me where it would make sense to create synonyms for things that we actually sell. So, the really cool thing is, like, you just talked about how you can schedule things. So, the query you set, like, hey, let's figure out places where we need to do some optimizations every week. So, you can have it all ready Monday morning for you just to verify it and press accept, I assume. Exactly. How do you see it? Exactly. And it makes a lot of the things that you would otherwise do anyways by looking at it every now and then and then implementing changes much more like a regular part of your routine. So, it does all of the pre-work for you, presents you with the solutions and then you can look at it and think about if these are actually the solutions that are most fitting. And with that, then, can automate a lot of the things because the MCPs, they don't necessarily replace the human here. What they do is that they help you analyze a few things that you might not have spotted otherwise and that you might have skipped through. So, this is where they are really powerful. So, the interesting pieces, what I'm hearing for you is like, the shift of AI has changed quite a bit. So, in the old days, we, as humans, produced something and we had AI verified. But it seems like it's going the other way around. It's producing some stuff and us, as humans, have to verify the quality and then can ship it off. Exactly, exactly. And that just flips the whole way of how your work can be made a lot easier. Also, timing-wise, you get more time to focus on actually other more important things, which I think is a very nice thing. So, now we have our results here, where we can see that we have actually different kinds of queries that ended up with zero results. And it gave us a few different synonym candidates. It also made a judgment call here to say, okay, this might not be the perfect candidate and some things that might be of another fashion. So, what it asks now is if it's okay to add those nine synonyms above, which it has presented here, which I will say to yes, go ahead. And now it will get to work again. I'll need to confirm that this is allowed from my side. It will make sure to add these synonyms. And once the data gets re-indexed, they will also be live on the page. Super cool, Julius. So, now it's already done. We have 11 synonyms entries added across those nine different queries. And as we can see here, it's not live yet. It will get live after the next re-index, which is here for 1.24 local time scheduled. So, basically, what you showed us is that first, how can we build a report and take that report and actually make it actionable? So, I think that's a key thing for all of you guys out there. So, when you ask LLM to pull the data, it could be from search, recoms, page, whatnot, retail media, and it comes up with suggestions. You can actually feed it back to the system and say, okay, now you have the suggestion. Please just do as you're proposing and basically have the work done without having to click five different buttons out there and really making all the optimizations a breeze or even more of a breeze to do. And now we're doing a demo and trying to keep it fairly short. But if you are in doubt about any of the things it's suggesting or you wanted to expand on something, it is a conversation. You can continue asking, why do you suggest this thing? And it will come up with some reasons for why. And then you can decide if it makes sense or if it doesn't make sense. But it is very cool to use the tool to explore the data that's available and then be able to apply the end result of that to your platform in an efficient way. Yeah. And it is fairly efficient and becomes even more so if you have multiple stores. For instance, if you have multiple locations and you kind of want to do optimizations to more of them, it becomes even more efficient because you can use the same session to interact with multiple stores in our platform. So that means that you can optimize the synonyms for different languages, of course, and I assume also the sciences will show in a little bit. One of the key things we worked at the HillaryTor for a long time is our product intelligence. And that has been MCP enabled a bit. Jacob, could you with a few words just explain what is product intelligence? Yeah. So when we get product catalogs indexed into our system, we'll do an analysis of the products. And along with all the data we have collected over time and the models we have trained, we can analyze those products and figure out what type they are and what things they go well together with. And yeah, so that's part of it. But also properties of the products like what time of week do they sell the best and is this a good candidate for being lowered in price or not? So will you sell more by lowering it in price or does it actually not make a difference? So all that intelligence is what we extract from the data we get in using the models we've trained. Very cool. So Julius, how do we use that in the Cloud Code as you're using right now? So what I could do now is, for example, that I have my newsletter that I'm sending out now for the next quarter. And in this newsletter, I still have space for a few different products to be shown. So what I could do is to ask the tool to give me certain products that sell well in Q4 of a year. So what I'm going to do is that I'm going to ask, I am currently working on our Q4 newsletter. Please show me a few of my products that would work well in this newsletter based on their Product Intelligence data. So Julius, while you're thinking about it and pulling the data from our AI models, a lot of times we are assuming it's like we get stuck. We get not writer's block, but I don't know, learn block. We don't know what we are going to ask of it, of good ideas. If we want to be inspired for new things that is not just what we're showing today and what you can use our system for and the MCPs for, where should you go use? We actually have quite a few examples of what you can use it for within the developer documentation. So if you go to developer.helloretail.com/mcp, for each of the different tools, you will find a few different products that you can go ahead and try out to see how it actually works, how it also works with your data and your shop. So now we can see based on the product intelligence data that there is a few different products that it would suggest for us to include as their share of quarterly sales falls into Q4. And how do we read this? We need to look at, okay, where are things actually higher than 25% because if they would be 25%, it would evenly sell all year. So we now could actually just take these few different products that it suggests and include them in our newsletter. And that could be done in different kinds of ways. If you're using newsletter content from Hello Retail, you could also automatically have these products selected for your newsletter sense in there. So that means that I'll just continue the chat and let it know, hey, please create a new newsletter campaign and add those products in, is that correct? Exactly. Very cool. And I think there's also one more very important thing here to remember, which is that the Hello Retail MCP just doesn't work in isolation as well. You could also say if you have connected, for example, to Klaviyo to say, okay, create this campaign in Klaviyo for me. Or if you ask it for which products sell well with a discount, if you're having your Shopify connected to say, please lower this product directly in price alongside. And all of these things can be orchestrated together in a full system. So it seems like we can do so much with it. We've talked a little bit about how getting data out, updating data. Jacob, I know one thing that you've also looked into is like, how can we use the NCP to update the visuals of our different solutions? So let's see what you have to show. Yes, definitely. We have this demo store that Julius has also been working on the data for. And it already has our search configured. You can see it works. But we will pretend that we are closing up on Halloween and that we would like to adjust the design to reflect that we are running a Halloween campaign and we want to sell some products. Before the MCP tools, the way you would do this is that you would adjust the template code in Hello Retail. But now you can use your agent to do this. What we want to do is adjust the design on this on our demo shop to include some bells and whistles for our Halloween campaign. So we'll start out telling the agent to do that. Let's adjust our search template to reflect our Halloween campaign. I've already prepared a campaign page on my Shopify site so I can tell it about that. The campaign page is live at... I've prepared a small graphic and... I've prepared a graphic for the campaign. So I'm just giving it the URLs to these assets that I've prepared. And I want to put the graphic in the corner of the search template. And I want to make it linked to the campaign page. So the agent will now fetch the template and it will look at the code and it will do the adjustments and it will make it possible for me to view those adjustments directly on my site. So because the shop is using the managed platform from us, it is possible to preview these design changes live on your site without publishing them directly to your customers. So it's a completely safe way to do this change while still being able to preview it visually. Yeah, so it looks like it's done and we'll go have a look at it. So it has now scheduled out a plan for how to do this change and I will let it know how to... Let it know to just go ahead and do it. So while you're typing and thinking about, I think it's pretty cool that, okay, you're asking it to update the search. But again, as we talked about before, you can change stuff out. So you can imagine that at the same time, you're also having it update the boostings for your different pages. You could create a retail media campaign for that specific product or campaign page and basically have a chain in different ways of working in just one career. I think that's pretty cool for this one as well. Jacob, right now you're doing it for search. Is it different for, let's say, recommendations, it could be pages or whatnot? No, it would be the same workflow. And the cool thing is that it is now doing developer work. So it knows about how to do JavaScript and HTML and it has instructions on how our APIs work and what's available within the tooling. So it really allows someone who is not a developer to do the type of changes that would have required a developer previously. And then if you're a developer, you, of course, have more knowledge about what's possible and how to best do things. So that gives you even more powers to do quick and efficient changes to the setup. Exactly. And if someone breaks it, we have an FCP endpoint to our audit log so you can figure out who actually broke it. Yeah. That's pretty cool as well. So now it is done and we can see that it has created a draft of the search configuration and we can now preview it directly on the site. So as we can see our maybe not that pretty Halloween graphic is in the corner. And if I click it, I get to my Halloween sale page. Yeah. And let's say that we want to adjust this for sale badge to also reflect our new Halloween spirit. And maybe I want it on the right hand side instead of the left. Well, there. So I think what you're also showing Jacob right now is like since it makes it so easy to do visual changes with MCP, I think what we're going to see is a lot more people are going to do these more seasonal stuff. We're going to see more badges that's going to be colorful in the season or kind of the spirit you have right now for Christmas as well. But like smaller holidays, I think we could do some interesting stuff around campaigning as well for this because it's going to be so easy to do that across all solutions. Absolutely. And I think there is also a big thing about actually having the possibilities to do some sparing with the tool to what kind of changes you could do. And there might also be some different ideas to what else to do. So you can really get this holistic view and this project that goes all around your shop to prepare for Halloween, for Black Friday, for whatever kind of event that you have. So basically what you could do is you could say using our product intelligence you showed before, give me anything that sells well around Halloween. Let's find something that sells well when it's discounted 10 to 20 percent. We don't want to give too much of our margins away. Give those list of products. And if you have a platform that's connected with MTP, hey, please update those products, the prices for those products. And suddenly you have the campaign running, maybe even with some visuals that Jacob is showing as well. Yeah. And also, as we as we're doing here with some batch that is on there, you could automatically have it at the sale batch in the right way and the right look that you want. So let's have a look at how it actually looks. Yeah. So we have now updated the sale batch to a nice pumpkin orange and it didn't require any particular knowledge of me in knowing how our system specifically works. It is just plain English explaining to the agent what visual change I would like to see. And again, remember that it can be an iterative process. It is not the most pretty sale batch in the world. So I would probably continue working on it. Maybe it could even be a small pumpkin even inside the product tile. So there's plenty of options to continue working on it. But I think the example illustrates that you can fairly quickly do some changes that would have been difficult before the MCP tools were. And of course, the LLMs were available. Yeah. So I think what we have seen today so far, we've seen how we get data out from the system. We've seen how we can take the data we've gotten out from the system, automate the workflows. We can even schedule it. We've seen how we can do visual changes and how we maybe we didn't show it, but we also could change it. So doing maybe some knowledge gathering from the product intelligence and using that for visual changes as well. So super exciting. And so thank you so much for listening. And thank you so much, Jacob and Julius for joining us. And we'll see you around in the Q&A in just a bit. Thank you so much. Thank you. So, Jacob. Yes, you do. Okay. Yes. So basically the PyMTP is available to anyone right now in the beta. And it's available to anyone who has a paid account. It doesn't matter which of our products you have. Just if you have any of the paid accounts, you can utilize it for free at the moment. So, let's see what the next one is. Julius, do you know the answer to the question? Yeah. I think it's a very good question here from Frederick. So, what I've seen is that the difference in quality can be there in between the models, just as the models get more smarter over time. But in terms of the application and what it can do, I wouldn't say there's any difference in quality. It's just that the things that it can derive from the information it gets from the Hello Retail MCP can, of course, differ depending on what model you use. But there is no real recommendation from our side what model you use, because I think they are all very good and work in the same way. I could maybe add that specifically for doing analysis of data, then maybe it is worth paying for a bigger model, a more capable model. Whereas more grunt work like updating a lot of data, I don't think it makes a big difference using Opus or Sonnet, for instance, will be able to do it just fine and it will most likely be quicker at doing it on top of being cheaper to run. So, I think it depends a bit on what you're trying to do specifically. Yeah. Yes. So, Annika, just reach out to me. I think it's been in the chat. It should work out of the box. So, you should be able to write your domain name. So, myshop.se or .dk. And then they should actually be able to use it. I know our Delivery Services tech support team is utilizing this actually in terms of using it to copy the size between domains. So, I haven't seen any issues, but please just reach out to me and I'll see if you can help. So, Copilot. Yeah. So, right now it works with Copilot as being annoying that it doesn't follow the standards of the MTP protocol. But we will make sure that it works. So, I hope within the next few weeks we have something for you guys to be able to utilize it for as well. Julius, do you want to answer that one? Yes. So, with the MTP you also have access to the activity log and that has all of the information about changes that have been made. So, you have the possibility to see what has changed and with that also revert changes that have been made by the MCP. But generally, also as we said, before anything happens, the tools would ask you to do anything if they are allowed to do a thing on your behalf. And also again, the configurations that are live on your site, like search configurations like we were looking at, they are in a draft state. So, you have a search that is running live on your site and then when doing modifications, the agent will make a copy of that live configuration and into a draft. And that draft is the thing you're working on. So, you can discard the draft if it turns out not the way you wanted it and you'll just be back at your live setup, which is what your customers have been seeing all along. So, there is really no risk. You can easily start over, discard the draft, create a new draft and try again. So, specifically, it changes to the design and the setup on your site. It's not that risky. Yes. I think that were the questions that we had from the audience here. If there's any more questions, feel free to post them now or otherwise. I think we'll follow up or send us an email afterwards and we'll make sure to send you the information and also always trying to enrich the documentation with more examples and more information on what we can do. So, thank you so much for joining and we'll see you all soon. Thank you, too. --- ## About Hello Retail Hello Retail is an AI-powered ecommerce personalization platform based in Copenhagen, Denmark. We help online retailers improve search, recommendations, merchandising, email, and retail media through our proprietary Product Intelligence engine. - **Website**: [helloretail.com](https://helloretail.com) - **Video library**: [helloretail.com/en/resources/videos](https://helloretail.com/en/resources/videos/) - **Demo**: [Book a Demo](https://helloretail.com/en/demo/) - **AI information**: [helloretail.com/en/ai-info](https://helloretail.com/en/ai-info) --- *This content is optimized for LLM consumption.*