Hello and welcome to today's webinar. What we're going to do is we are going to have a deeper look on the real use of AI. So no fluff, really a deep look on how it's done. While we are at it, we really want to encourage you guys to also participate in our webinar. So there's this Q&A tab. And yeah, if you have any questions, please drop those questions. Say hello. And yeah, especially at the... Aha, here we go. Here's the first one. Perfect. Hi, hi, Dirk. Yeah, and at the end of the webinar, we are basically going to have a look at your questions and we are going to discuss basically in detail what you came up with. Okay. For the people that don't know me yet, my name is Jan Schulte. I'm here Head of Consulting, 12 years with Magnolia. I'm also... nicknamed the Swiss Army knife of Magnolia simply because, yeah, I'm pretty active also still in software development, project onboarding, pitching, or like today, running a webinar. Okay. Also pretty important to note, if you need any additional background materials, there's a docs tab. No, there's a docs tab. So in the docs tab, you can basically feel, as I said, all the material around the content. And that you'll see today. And also, this webinar is going to be recorded. So it will be released in a few days. So you can then basically also watch this recording or pass it along with your colleagues. Okay. Now we really got a lot of folks. So definitely hello to Slovenia. Hello to Switzerland. Hello to Germany. Freiburg, pretty close by. Okay. Singapore. Hello. Okay. Well, this is really a big crowd today. Fantastic. So good to see you all. Okay. For the ones that don't know Magnolia yet. So what's Magnolia? Magnolia is a digital experience platform. That means with Magnolia, you can basically run your websites, your portals, your digital touch points. And as I said, what we are going to do today is we are going to have a deeper look on how you can really utilize AI. in a DXP platform. And we are going to talk about it. We are going to discuss it, how to bootstrap a project. And we are also going to have a deeper look in a real world implementation so we can really see how it's done at scale. Okay. And another thing after the webinar, obviously not right now, but we have on the top this small button that's called the Magnolia AI product. So if you want to basically also want to basically also try it out by yourselves, just hit this button and then you can just basically have a deeper look on how you can utilize AI to automate a lot of your typical tasks that you run in your DXP projects. Okay. I would say if I look at the numbers, it looks like we are now at a really good rate. Yeah. That being said, I would say let's kick it off by running a poll about your usage of AI. So let us know in this poll basically how do you basically see yourself in the domain of AI. Are you just a beginner, an expert somewhere in between? Just let us know. And that being said, well, I would now really like to welcome our experts on stage so we can really discuss how AI how AI is done in real life. No fluff. And here they are. Fantastic, man. Cool. I would say, how about you guys just introduce yourselves to our audience here? So as always, ladies first, Eleonora. Thank you, Jan. So my name is Eleonora Parlatore and I'm the Head of Creative Services and Operations in Global Blue. I have more than 10 years of experience in marketing and operations. as such. And my team is taking care of the B2C communication for all our shoppers in digital channels, especially in app, web, SMS and app push notification. I'm very pleased to be here to share my experience with Magnolia and thank you for inviting myself and Christopher. Fantastic. So, Chris, Will you just tell us a little bit about yourself? Hey, Jan. Hello, everyone. My name is Chris. I am a Digital Experience Platform Expert at Global Blue, part of the consumer engagement solutions team. My role focuses on enhancing Magnolia through new features and integrations while streamlining workflows to help our internal teams deliver content more efficiently and effectively. AI with Magnolia has started playing a master role in achieving these goals for us. I'm very happy to be here today. Fantastic. Fantastic. And last but not least, the one and only Marvinator Kerkhoff. Yeah. Hi, I'm Marvin Kerkhoff. I've been working for Magnolia and Arata Systems for almost 15 years now, developed a couple of modules and many of you may know me from our Magnolia blog, magnoliacentral.com. And, yeah, I'm happy to share some inside knowledge here. Okay. Perfect. Yep. And that's the thing with 15 years. I always thought 12 years, it's pretty long. I know it all. Well, it's going to be possible to basically overtake. Okay. Really, really good. I would say before we jump right in, let's have a quick look at our poll. So, okay. Not so much beginners anymore. That's interesting. So, yeah. We're getting quite a bit of intermediates and we have 50% of advanced users that use it regularly. This is, yeah, it's a really good crowd that we have here. Awesome. Good. Let's start actually with the why. Why the usage of AI? And for that, I don't know, basically what motivated you and how do you basically, yeah, started with everything around automation with AI? Thank you, Jan. So, at the beginning of our journey, honestly, we tried to figure out ways in order to optimize and simplify, especially, our content workflow. And honestly, this is something that it motivates today as well. And we were focused in three main pillars that was, of course, consistency, speed, and cost saving. So, what happened is that during this journey, we focused on translations, first of all. As you can imagine, our website has a lot of different language barriers, language sections, and 12 now, but in the future, even more. And we thought, okay, let's focus on this first quick win in order for us to maximize our speed and save costing. The second one is the metadata. And the third one, of course, is regarding the alt text. Yep. That definitely is just, I mean, knowing CMS in so long is exactly those pain points that typically cost a lot of time. And therefore, yep, this is really a perfect point to start. Definitely. Yeah. So, if we speak about concerns, because always at the beginning of an AI journey, there are obviously some kind of concerns, challenges, and then there's also the, yeah, doing the first steps. And therefore, Chris, so if you think about the initial hurdles or tech, governments, quality, and so on, how did you tackle that? And also, how did you overcome it? Yeah. So, the initial concerns when we started to introduce AI into Magnolia was mainly around, as Eleonora mentioned, quality and consistency. Early AI generated translations often lacked that linguistic nuance or contextual understanding required for content, and especially our bespoke content. And given the diverse languages that we have, as Ela mentioned, we have 12 and we will have more in the future. However, we have refined our prompting strategy to improve the quality, and this has improved it significantly. We will continue to improve this prompt from today to make adjustments to improve the output. Today, mainly AI has become a trusted co-pilot for us to produce content faster without compromising on accuracy or brand tone mainly. Yeah. Yeah. Yeah. Yeah. It's, um, also looking back, it's pretty, pretty crazy how everything evolved. I mean, I can really recall, hey, there's no auto translation from Google and from Microsoft. Yay. Yeah. Yeah. Exactly. Did certain things, but compared to what we are capable to do right now, I mean, it's crazy. Really, in those very few years, it's, uh, absolutely mind boggling. Yeah. Um, let's really, start then to talk about, um, some quick wins basically that were achieved, um, right at the beginning of the AI journey. And with that, um, Eleonora, um, so can you tell us a little bit about the first, um, quick wins and also, especially on what you get, what you got back from using AI and what was a global impact on the overall workflows and output and everything around that? So if we are discussing about the quick wins, I would say that translations is what we are discussing about actually. And of course, at the beginning, you can imagine that was one of our biggest bottleneck. Um, every single time we had to do a variation, a change on our translation on the website and on the app, actually, um, there were a lot of back and forth with, for back and forth, sorry. Uh, I'm Italian. I don't know if you can, I guess, I think you can. Just a little. And what happened is that, yeah, um, there was a lot of, uh, email with the external agency, a lot of, uh, additional costs actually. And, uh, quite sometimes that we were waiting to receive the translations back. And not only that, also, uh, as soon as we had the receiving of the translation, some editor in our team had to go inside the, CMS and start importing all this transaction manually. So you can imagine how much time we were wasting, honestly. Um, this is, I think one of our, later on, I can also discuss a little bit about the return of investment, but this was one of our main concerns and the first things that we were looking when we were discussing about workflows. Mm. Yeah. Definitely. Okay. Yeah. Um, also Chris, I mean, if it comes to metadata is also some, some of those hotspots where we typically see that a lot of energy is lost. What's your take on basically also here automating things? Yeah. So metadata generation has traditionally been manual and it's often an overlooked task as the importance isn't always fully understood, especially from our internal teams. Uh, uh, uh, uh, our teams will focus on building, uh, uh, our teams will focus on building, for example, marketing campaigns. Uh, they need to build the landing pages, the email templates, uh, the targeting all needs to be put in place. Uh, metadata can quite easily be forgotten, um, in the prioritization of everything. Uh, with Magnolia's AI driven, um, generation, metadata can be created, uh, with one click now, uh, pulling key information from the page. Uh, this will meet SEO standards. Uh, this can then review, regenerate this content if they want. And publish instantly. This automation turns the previous tedious task into a quick, reliable step. And it also helps, um, our teams build good habits and maintain consistency across all pages. Um, another thing, Yarn as well is the fact that it's still, I think, very exciting to see a mundane task that can potentially take hours previously now takes a second with a click. Uh, there's still, uh, there's still, there's quite a lot of magic about AI at the moment and how quick it acts. So just a lot of time. Yep. Yep. Um, yeah. Um, you completely subscribe to that in the end. Uh, but on the one hand, it's super magical. On the other hand, you get also used to it, to this new level of speed. And this is how it feels to me. Um, as you said, stuff that typically takes me a very long time. Um, it's just. A mastered advantage for us is being able to focus on more important tasks. Instead of spending two hours populating metadata, we can now spend that time doing more valuable tasks or even upscaling ourselves. So it is, it's really valuable to us. Yep. Exactly. This is also how I see basically AI in general. The stuff that I typically do is stuff that I don't enjoy anyways at all. And then just focus on the cool thing on the creative part. So we're coming up with more content, more code, whatever it might be. And if we can also build on top of that, what Chris was saying is now, honestly, the, the editor that is taking care of the page creation and the campaign, the market campaign creation, uh, can, uh, do it in a, as Chris said, a very easy way, quick way, but also without asking the help for other experts that maybe at that specific time has to be focused on other activity. So it's really not just, uh, things that it makes, uh, it makes the work faster, but even better and more accurate without asking for external resources. Yep. Absolutely. Um, yep. Um, as I said, also for me, it's the exact, um, same, um, perception. Um, it's just for working with the system or also with getting additional information where I would have otherwise asked, um, around the team and whatnot. Um, AI can really deliver a lot and automation, getting the right things, um, uh, uh, uh, done and delivered. Um, so Marvin, um, basically if you have a step on the, um, immediate, um, benefits that we had when you used AI, what was it for you? So, uh, you mean the third quick win would be, I guess, if we talk about in this scenario, um, that, uh, image captions and alt text creation, um, Usually this is something that also is forgotten after you set up your website. Most of the times then maybe after a couple of days, people are just not doing anything around it anymore. And yeah, this is something that we automate regularly now with the AI tasks that are automatically scheduled. And yeah, now you can also then trust these outputs good enough that you can say, okay, this is really something that you can publish more or less without even look at it and automatically translate it. And yeah, especially for things like regulations, like the European Accessibility Act, you just need those things, right? And for that, I think it's just a pretty simple solution to make it possible for everyone to participate. I'll just add to that as well, because obviously this has been something that me and Marvin have recently integrated into Magnolia. And the fact that you can now upload an image and it will generate alt text just automatically without even clicking a button is just super handy. You don't even need to think about it anymore. You just know that every image is going to have alt text. Yep. Yep. Indeed. Because yeah, as Marvin said, on the one hand, and also I've seen the story so often, it's really absolutely necessary. And everybody just writes off, yeah, for each, every single image that I'm going to upload, I'm going to create this alt text. And then after a few months, well, 40% doesn't have, don't have any alt text at all. And then just even you don't have to care about it. You just get those alt text in all the different languages. Again, huge time saver. We had another case. And at the beginning, I could not even believe the numbers. But once you run the math, it's absolutely true. If you look at the project, in the end, it was just seven years that are completely automated away. And again, this is time that you can use basically to do stuff that is really impactful and not describe an image that are there anyways. Okay. Cool. And time, honestly, sorry, but time is, you know, cost savings. For example, if I can also give you a little bit of fresh news in Global Blue. So we just finalized a math exercise in order for us to understand the return of investment. And actually, thanks to these features, we are saving a lot. For example, for translations, we usually go to these external agencies, what I was mentioning, before, and we saved that for just bulk translations pages, we have saved more or less 70% of our costs. And not only that, because we are discussing also about speed. The team is doing what in the past was requiring weeks, just in few minutes, actually seconds. So it's also on the action level, we see that we are saving a lot, a lot. I think it was 80% of our workflow time. Well, 70% cost savings, 80% workflow time. Okay. This is really massive. That's crazy. This is really impressive. Absolutely. Yep. Okay. Before we move on, and since we really discussed on how you can save a lot in the real world, not in theory, let's quickly run a second poll to ask you guys out there, how you are basically using AI right now. Besides just using chat, you can use your browser, obviously, which everybody does. Okay, cool. While the poll is live, let's maybe, obviously, everybody knows if it's about using AI, a lot of the magic lies basically in the prompts to make sure that you get accurate results. And therefore, let's take a bit of time for some reflections and advice. So Chris and Marvin, can you tell us how to prompt like a pro? Yes. So when it comes to prompting, context is everything. Obviously, we do know this. Large language models perform best when they clearly understand your intent, your audience, and the tone. The best advice I can give here when it comes to prompting is take the time to test and learn. Test and learn, see the output, and then adapt your prompt. Over time, this approach will help ensure that the output aligns with brand glossary, tone of voice, and it'll become more accurate. Here, I can go through each focus and show you our global prompt with a slide. So this is our global prompt. If we move on to the first section, the first thing to do is to explicitly define the role of the output. So here, we're telling it, you're a professional translation engine. So we get it off on the right foot. Next, we'll explain the function along with dynamically inserting the language ISOs for translating from and to. So we'll always use English as a master, and then we can insert the other language. Next, we'll further explicitly demand it only returns the given text translated and not to explain or answer or expand on any of the responses because previously we've had times when the headline is a question, say, how do you shop tax-free? This will sometimes get answered. So you need to really be explicit in your demands. Next is the preservation of HTML tags and variable placeholders. This comes from the ITN workspace. And rich text fields. These as an input to the LLM will always have HTML tags. We want to preserve these in the response that we get back. And sometimes it will eradicate these. So we make sure that we put this in there. Next is don't translate certain words, anything within your glossary and general words such as brand names or store names. Next also is long and short form text. We have told the LLM to basically treat short words. words such as one or two words to treat it as UI or like buttons, labels, tooltips, and then apply standard terminology to it. For long form text, we treat it as a normal sentence. This allowed us to get a better response in the end. And lastly, one thing that I do want to mention is something that me and Marvin worked on recently is that we added another placeholder variable at the end, which we've associated with an instruction field within the translation extend model. This allows us to add extra context. This allows us to add extra context. So this is a global prompt that gets assigned to every single request that we make. But then we also have the ability to add extra context to each individual request that we make. I hope that this helps people create and craft their own prompts. For sure. I think that was the everything that you need to know about prompts and never dare to ask. So basically, if you really follow this pattern about being specific, down to the latest detail, down to the latest detail, that definitely pretty good advice also for basically all domains, especially translation, but also on content creation and so on. Perfect. Really good. By the way, before we move on, don't forget, I know this side, there's this Q&A tab. So really, if you have any questions, please drop us a note so we can then just basically discuss later on your questions around the real world usage of AI. Okay, let's continue. So, okay, let's continue. So, Eleonora, Marvin, so, as I said, the first steps are typically the hardest. So what's your advice on your first AI project? So, how to get going? Yeah, I would say start with simple, easy tasks like, for example, SEO metadata generation. That is pretty simple, out-of-the-box solution with Magnolia. You install your module, you install your module, AI accelerator, and then afterwards need to connect it to your model, and that's mostly it. Then you have your dialogues where you can generate any type of text, especially for SEO metadata. That's pretty, pretty simple. And then go ahead with some more advanced things. But it's still pretty simple, is the asset caption and alt text generation. We did it as a third part because we had some other issues that we wanted to tackle first. But I would advise to do it next because also there, it's pretty simple. You also can connect it directly with translations. So it's probably not translation per se. It's more generation per language from an image. And the last but not least step would be then to go into the translation steps. As you saw from Chris early on, this is a more advanced story because you need to go through your text and check and check and check if this is good or not. And depending on your content. And depending on your content base, this could be massive with a couple of persons involved. Especially if you have different languages, you need to contact people that know this language by heart. So yeah, you always need to go through this and improve the prompt itself. Yeah. And then you probably also want to code something here and there, especially with this translation stuff. It's also a thing that you want to control over a specific workflow. So it's not only just hit a button and wait. So it's also to check who is involved in proofreading, who is involved in this and that. So this is mostly a more advanced story. But overall, yeah, that's it. On my side, I would say maybe it can sound quite obvious, but as Marvin said, start with easy features, something that can be added on a manual in an easy way. And this is true, actually. But also, you have to keep in mind what are your objectives, what are your needs. You have to have a clear brief and detailed brief of what are your needs in mind. Because Magnolia and AI will give you a lot of flexibility, a lot of flexibility. You are going to build something in a tool that your team already knows how to use. So you just have to be conscious of what are the needs, the objectives, and just go straight in that direction. Otherwise, the schedule is the limit, isn't it? So sometimes it's difficult just to be focused on specific features and tasks. I subscribe to that for one of 1%. And this is really one of my biggest problems at the moment. You can do absolutely everything. So it's really about finding a clear goal and then also sticking to it and not getting carried away. Yep, definitely great advice. Yep. I mean, no, we spoke quite a bit about it. I would say let's take the chance and have really a look behind the curtain to really see how a real world AI-assisted Magnolia installation looks like at scale and not like in a demo project. Okay, great. Let's start with the Pages app. We can go to the specific page that we want to generate SEO keywords for and hit the Edit Page Properties dialog. Inside the Meta Description, we have two fields for the keywords and for the description. For those, we can now generate these texts. And then we go to the Assets app. Because in the Assets app, we can also generate the captions. And the Assets app and the alt texts. So I upload a Thailand image now, which then, after a while, sends the data to OpenAI. After it's getting the data back, it will save it. It will then hit the next translated version and go on and go on. It will does it for two fields, the caption and the alt text. And, yeah, this will take a while. And, yeah, this will take a while. So let's do this thing. And we go to the next step. Inside our i18n app, we have a lot of keys. And we also see that one of these keys is actually not translated. And for that, we can now go to the Instant Translation All method, which then sends over everything to OpenAI to translate it. We can also use other models here. So, yeah, everything is translated. Perfect. We can then hit Publish. And that's it. So let's go back to the Assets. Check if everything is there. So we see everything is here. Caption, alt text. And we also should see it in different languages. So next up, let's go to the Pages app again to become a member. This page is actually not translated. So if we want to go, if we check, for example, the Spanish version, we see that all the text is still in English. So we can go to the QuickTranslate method, which then sends all the text over to our specific model. As I said, it could be Gemini. It could be any kind of model. We are totally free here. And, yeah, since this is a couple of texts and also a couple of hidden things, it will take a moment to translate all the text. text and page properties and so on into Spanish. But then, yeah, it refreshes it. And we should see that everything is translated into Spanish. And also inside the Edit Page Properties dialog, we should see that the keywords and the descriptions also translated into Spanish. And this is how it's done in real life. And not just as a concept. Yeah. And as you said, especially, I mean, all those things that we saw, they just obviously compound up drastically because there's a lot of editors that have to do, in the old world, send those mails or hit APIs to long-running processes to certain agencies. And so, you know, certain agencies. And now you get it really in a few seconds, which, yeah, well, we heard it 80%. That tells already the full story. Okay. I mean, now that we basically master the art of AI and automation and automating those tasks, if you reflect and take a few steps back. So was Magnolia the right fit? Why was it the right fit? Yeah. What are your thoughts on that? I think it was absolutely the right fit? Why is I mentioned it a little bit earlier, I guess. The adoptions of AI inside Magnolia was very smooth. We are using exactly a tool that our team knows already how to address. There are no other new tools that we have to implement. And there are absolutely not difficult trainings that has to be put in place. And this was very useful for the team because we just give them a demo, explain them how to do translation or metadata. And in a few hours, I would say at the same day, mostly of our team was already able to do this specific activity by themselves. We were just checking, doing quality control, check also the tone of voice, but it was it was easy. So this is why I really think that this can be a good fit if you are using already this type of CMS and experience platform. Obviously, to add to that on top of Eleonora's comments, for me and Marvin will probably back me up here, one of the biggest advantages of working with Magnolia, especially around our AI initiatives was how simple and seamless the technical implementation was. integrating these features didn't require a huge amount of additional time or resource. We could actually squeeze it into our sprints that we currently work on that already have large scale projects happening. These could work alongside. Also alongside this was Magnolia's flexibility and openness to collaboration. They were always gathering feedback, providing advice, and they understood our goals. They worked with us very closely during this whole initiative. And it made the entire process feel like a natural extension of what we're already doing, not a separate complex project. Yeah, and I can jump in here as well. So I can just give everyone an advice that do not, because I know it from my implementation, implementations in the past, just try to involve Magnolia as well inside your processes, right? Because sometimes it's just, because sometimes it's just, it's just Magnolia is somehow the company that delivers some kind of product, but it's not that, right? So you can really collaborate on that, especially for that topic. I had a lot of discussions with Jan, with Sebastian already. We also developed something, then we made this available on Magnolia Central, for example. Then you, get back to the board, get back to the board and integrated this. So this is now part of the actual product. And yeah, that is something I can just give everyone the advice to do it, right? So Magnolia only knows your thoughts and problems if you tell them the issues. Yep. And that being said, again, just to encourage it. I mean, obviously, AI is evolving at a breakneck speed. Suddenly there's Gemini 3 and it's really a start or challenge the throne of JGPT and so on. So from that regard, every week there's something new. And really, if you have any questions, problems, ideas, and so on, really just reach out. We are all together in this journey on, yeah, applying AI more and more to be more efficient so we can focus on the cool things. Okay, cool. Again, gentle reminder, don't forget this Q&A tab. So you can just visit the app if you have any questions. Okay. So as I said, it's in the end, as so often with portfolio, it's all about collaboration. And I can only really just encourage everybody to reach out. While we are adding, again, this recording will be released pretty soon in the upcoming days. A little bit of shameless self-promotion. I want to also tell you that we are going to soon going to run a webinar on how to handle basically everything around Jero. So as I said, the web is changing really, really quickly. Most of you guys have seen it. AI is now deeply embedded in Google for search. That doesn't mean that Jero is not relevant. It's still highly relevant, but the term of GEO also becomes much, much more important. So that's basically the optimization for all those kind of chatbots that are used by more and more users. And yeah, we are going to tell you a little bit about how you can basically optimize your content to make sure that you can optimize your content. that you're really also there on top of the list if somebody uses those AI-driven Google services or obviously also the other search engines. Okay. Good. That being said, I would say it's time for Q&A. And again, right now is the time to drop a question if you have any. Good. Let's see what we have. Okay. Oh, that's a technical one. What's the meaning of slash function of a $s in the prompt that we have seen? Chris, maybe you can jump in. So this was the variable placeholders that we use. They're assigned to, the first one is assigned to a language. So we have our master English language. The second one is assigned to the select language, which is the language that we choose to follow up and translate into. From our 12 languages. And the last variable placeholder that we have was towards that field that I described at the bottom. It's an instructional field, part of the translation extend application workspace. This allows us to give extra context. So they're just placeholders that we then can input data into. So in special scenarios. In special scenarios, you just need this extra instructions for chat GPT, especially. This brings us probably also to the other topic that there's already a question for regarding dictionaries and so on. So you can probably do this by just inserting this inside the prompt, especially inside those instructions. Or you just use things like the deep. So if you rely on deep. So if you rely on deep. So if you rely on deep. So instead of open AI, Gemini or whatever, then yeah, you can just use this as a placeholder there. Yep. Okay. What else do we have? Oh, okay. Are there any remaining challenges or especially if it comes to reporting and visibility? Yes. I mean, it's especially if you have a big content and you want to translate it all the way around the website or the website. So if you have a big content and you want to translate it all the way around the DxP itself. It could be on the pages app, assets, somewhere else. Sometimes if you just hit and wait, it's challenging to identify which parts are now translated and in all languages or not. So I can give you some smaller example. For example, we had a certain scenario where a specific phrase in a specific language was was not correctly translated because it was harmful for the OpenAI API. And then it just didn't result into anything. So we just had a simple text field without anything in there. And to identify those things, it's sometimes a bit challenging to get it. So that is something that we could improve as well. I think that's what we wanted Marvin to work on in the future was obviously to get some kind of output, some response. If something does fail, then we know what to action, why it fails and what to, because if we're sending off massive batches, especially in the i18n workspace, you're talking thousands and thousands of keys. You can get lost in the data sometimes and it's really hard to pinpoint. Yeah, absolutely. Okay. Okay. No, we got quite a few more. Okay. I'm missing this kind of a repetition. Does Magnolia have something like a translation memory? So that identical strings and then the question ends. Again, from my point of view, the translation memory is more or less a thing of the past. So if you just work with traditional agencies, obviously having this translation memory saved you quite a bit of cost. But in the age of cost. But in the age of AI, it's, from my point of view, please, I'll be correct me if I'm wrong. From my point of view, that's one of those concepts that are not so relevant anymore at all. I mean, well, I would say no, it's probably still relevant a bit, because sometimes you just want to have certain things not translated. As we saw it in the prompt, I guess, Chris showed the specific things that the specific things are just English and they should be English or just specific things should be translated in a specific way, because probably also you have some artificial names, right? So names that are not existing in any language and you don't want to translate them or you just want to translate them in a certain way. And therefore, I think it makes sense to have something. But you need to give this context to an AI. Yeah. Yeah. As well, sorry, if I add these, it's true that AI learn, but we do learn as well. And this prompt is not something that is set in stone, isn't it? So if we, if you see actually that there are some wording that are not translated in the most perfect way, still grammatically correct, but not how you would like them to be, then you have to go back to the prompt and just change the prompt and AI will do their magic. So to go back to the original question. So to go back to the original question. No, there is no dictionary in Magnolia, but you can build a certain content app, which then where you can fill up all your text in there and then add this to your prompt. So yes and no. Indeed. Has, how has this new AI workflow been adopted by a wider content team? I guess that's for sure for Eleonora. For myself, yes. So how it has been adopted? So we started using AI in different channels. So we are sending out emails and SMS using AI AI content creation. And this is helping us a lot because as I said, consistency is the key. We want the message to be consistent in all the channels that we are using, isn't it? So the team start not just using inside Magnolia itself, but using the capability in order to create copy text in the other channels. And at this stage, because we are still, you know, improving our workflow as well, just using the content that has been created in these channels and copy them in the right workflow that we have in our process. I also just say that adoption is very easy in the teams just because of how quick it is. Everyone loves how rapid it is to create content. So getting the teams on board was not a hard task and it's helped them tremendously. Mm-hmm. Stop. More questions. How does Magnolia behave if the source text changes in the meantime? Is the translation then displayed as outdated is the question. So there are multiple ways to handle it, I would say. There is one thing where you say, I leave it like it is. There is no out of the box solution to just auto-generate everything. So that you say just organization-wise, if someone is changing something, then it needs to be translated by this person. But what you can do as well, similar to the AI translation for the asset captions, is just that you just say, if something is changing, so there are observers available inside Magnolia, where you can say, where you can say, if something is changing there in this, then you can just hit the translation again. But it needs to be proofreading and it needs to be published as well. So this automation is possible by, yeah, if something is changed by an editor. The other option would be to have some extended workflows. So if something is changing, detect it, maybe then send an email. So you can send an email to a translation team or whatever you have to just do this, whatever you like. So in this case, Magnolia is very flexible, but it does not provide you an out of the box solution straight away, but you can simply do it. Right. So doing such things like observation is really simple. Maybe half a day, day of development and that's it. Everything is there. Right. So that is pretty simple. Yup. Indeed. Indeed. So often with Magnolia, for the different content production workflows, what we really found during the years, there is no one size fits all. Each and every team is different. They are always different strategies and so on. And therefore there's not only one way, there are many, many ways. And you have to basically determine how we want to tackle that. And then absolutely, as Marvin said, then it's just about plumbing the, the features together. We, we used the out of the box workflow at the beginning. Uh, Chris, Ela, you, you know it, um, the way you had to batch everything into one translation. Then you hit the translation. Then it goes to the translation, goes back. You can have a look at what is translated and then click another button to import it. That is possible, but sometimes it's just not quick. Right. And we just need a faster way to spare. Right. And therefore we say, can we just skip it? And yeah, as you saw in the demo, we just, uh, hit one button and then it's translating everything, um, with this instant translations. Uh, so yeah, you are still using it, right? So in certain cases you want to have it. It's for the larger batches. We use the, um, or you, we batch everything together and send it off. But with a small translation keys with a short sentence or one or two words, we can hit instant translate. And it's much faster. Exactly. Yeah. Crazy fast. Yup. Um, okay. And there's another one from Kuma. Um, what AI-driven scoring features should we prioritize to ensure all content, including text and images meet our quality and optimization standards before publication? Hmm. Okay. Um, I cannot answer it honestly. Without any inside knowledge about this, uh, uh, the, the, the, so that is really hard because without any knowledge about the quality, uh, gates that you have, um, it's interesting for me to, to, to, to, to say something, but honestly, I have no clue. Yup. I suppose it's, it's just reviewing inter, like we have the ability of globally to obviously review some of the translations internally. Um, it's a lot of multilingual people at the company. Um, so the proofreading, um, so the proofreading, we, we don't have an internal score ourselves, but, um, if we do need to get some kind of feedback, we will speak to native speakers, um, to adapt the prompt basically going forwards. Yeah. Doing quality, quality control and tone of voice check. Yeah. But honestly, as Marvin said, it's depending on what are the, the, the, the goals, what is, uh, you want to achieve always, always start with the needs. And then find the solution. This is, I know quite obvious, but sometimes because we've tried out a solution before can be, yes, challenging. Yup. Indeed. Okay. Um, since I don't see any more questions and we are slowly, but surely running out of time, then I would basically, um, start to, um, wrap this Webinar up for, um, for today. So, um, thanks for everybody that attended. That was quite a big bunch. Um, and obviously, um, a big, big, big, um, thank you for you, Eleonora, for Chris, for Marvin, for also taking the time, uh, to basically, um, yeah, show our audience how it's done in the real world. Thank you for inviting us. Thank you. Thanks, Jan. Thank you, Magnolia, thank you guys.