Watch Amplify: Agentic Commerce Live on demand

For twenty years commerce had rules. Build the store. Make the content. Publish and wait. Those rules are gone. Today there are 100 channels. 40 markets. Every language on Earth. Today, answer engines sell, algorithms decide. And the machines have started shopping. Millions of questions asked by shoppers and now by their agents, all demanding one thing. Content. More than you have, faster than you can make it everywhere, all at once. Every brand on earth is making more than ever before, and it sounds like noise.

A million assets, endless content requirements, no conductor. Welcome to Amplify, where the noise becomes orchestrated. Where your content harmony begins. Three numbers. As we start our day together today, we’re going to start with three simple numbers. The number 100. Content has never been easier to create. Technical teams are connecting to large language models, commercial teams are buying specialist AI tools and we are all producing more.

The five paragraph email from your colleague is now a 20 page white paper. The five slides you need to pre-read for your team meeting next week are now a beautiful 20 slide deck that you simply don’t have the time to get to. AI has removed the friction from creation, but every new piece means more work for people in this room. Something to review, adapt, approve, publish, and keep current. The new brief is 100 times the content.

The number six. A few years ago, a retailer had around six months to prepare for the next seasonal launch. Today, a trend catches fire on TikTok at breakfast. By lunchtime, customers are searching for it. By this time next week, it may be gone. The expectation has moved from six months to six hours. The team has not gained six months in capacity. They are simply expected to move at the speed of the moment.

The number 60. The board sees AI writing, translating and adapting content at a fraction of the old cost. Let’s take translation, for example. Work that once required a major agency agreement can now begin with an AI first pass in seconds. We’re seeing mandates come down from the board to reduce the cost of content by 60%, not simply an aspiration, but a mandate. And that mandate lands with content operators, e-commerce leaders, and the teams expected to make it real.

100 times the content. Six months compressed to six hours. 60% less cost. More demand. Less time. Less money. All at once. Now there is some good news, I promise you. This also represents a huge opportunity. Earlier this year, while trying to understand the true scale of content operations, I sat down with a senior financial leader at one of the UK’s leading retailers, one of our customers.

We were talking about AI and expectations from the board. So I asked a simple question. What does content actually cost you? There was a pause. The honest answer. They couldn’t put a reliable number on it, not because they lacked financial discipline. It was simply that that number sat in so many places. What struck me wasn’t that finance couldn’t see the total.

It was what it meant for the people being asked to change the operation. The board can set a 60% target, but it’s the people in this room that need to work out how to go and deliver that. And you can’t reimagine your content supply chain unless you know where those costs are sitting today. You can’t make an informed choice about what to fix first, or where AI will genuinely change the operation, rather than making one task cheaper. Without that baseline, we are simply guessing. So my question to the room is, do you know your number, and do you know what’s driving it? AI makes individual tasks faster and cheaper. Write the first draft. Translate the copy. Adapt the image.

But when each task becomes easier, the business doesn’t ask for the same at a lower cost. It asks for more. Workload increases. It simply shifts to a different place. The whole operation does not necessarily become cheaper or faster. Economists call this Jevons paradox. Task efficiency is not operating efficiency. One improves a step. The other changes what an entire team can achieve.

And the gap between them is where time, capacity, and money simply disappear. Remember the numbers. 100 times the content, at social speed, for 60% less cost. More demand. Less time. Less money. This is not an abstract economic puzzle. It’s the operating brief now facing content and commerce leaders. So we took the opportunity to build a model to make its scale visible. We call it CORE, Content Operations Real Economics. It measures the gap between what content costs you and what you think it costs you. For a large, representative catalog retailer, our model puts annual content operations spend at around £34 million, depending on the complexity.

The range is somewhere between 25 and 52 million. That includes people, production, agencies, tools, localization, rework and overhead. But that’s before any media spend. And most organizations are accounting for less than half of it. So CORE is not there to turn you into financial analysts. It gives you the evidence to ask better operating questions. Where is our capacity going and what could our team achieve if we released them?

Getting to the number is not your destination. It simply tells you the scale of the opportunity. We know the costs are material, and we know the market shifts that are going on all around us make it urgent. Now, it’s only fair to point out that we sponsored much of the research in The CORE Report. So some of you may be a little bit skeptical. So please don’t take my word for what’s driving all of this.

So I think we’ve explored that AI has removed the friction from creating content, and anyone can create content fast now, but it certainly exposed an operational constraint. I’ve shared with you some of the costs involved and the implications of inaction, and Chuck has shared with us where all this is heading. I’m now going to welcome to the stage a colleague who’s going to share with you what you can do about it. So please join me in welcoming to the stage co-CEO and CTO of Amplience, John Williams. So thanks, Jeremy. So you’ve heard that everything is changing, right?

E-commerce is changing. The way we’re interacting with commerce is changing. And you’ve also heard about the increased demand and the needs for content. So I’m not going to spend the next 30 minutes telling you things are changing. I think we’ve already kind of done that. Chuck did it a lot better than I could do anywhere. But what I want to do is talk about what can we change as a result? What should we do as a result? Oh, there you go. What I want to introduce to you today is something new. The Amplience Content Operations Platform. Right. And it’s been built and designed with our infrastructure and everything that you’ve already come to love. But it’s been built for the scale and the complexity of commerce. It’s built with a very different view and I want to basically talk through. Actually sorry about that on the slide. That always happens. I guess

what I want to talk about is the difference between tasks and creation. Because actually when you create something you can now create product descriptions. You can create translation. You can now generate images for God’s sake. And you can, as Chuck already mentioned earlier, you can do buying guides and you can do FAQs. And increasingly we can now research and create content for AEO or GEO or for these new discovery channels. Right. You can do all of that and you can do it all with AI. And that is just fantastic, right? It’s incredible the quality of content that you can produce using AI, and many of our customers today are using that. We’re doing that today, but I think there’s something hidden within all of this, something really important, which is that creation is just a task. And that’s kind of what we said earlier, that actually it’s just a task, right? It’s an important task. But if you think about it,

before you create something, you actually need to understand what you’re creating. Someone needs to research it. Then it needs to be tested to make sure it’s the right thing. Well, here we go. Then you need to make sure it’s the right thing, and then it needs to live somewhere. It needs to live somewhere structured and managed. And once it’s there, you can then publish it, deliver it. But ultimately you need the outcome that it reaches the customer. So you don’t need a system that just generates content. What you need is a system that actually takes the content through the entire operation. And it’s really, really interesting in e-commerce, as you know, we’ve been in this a very long time because we’re not talking about one article. We’re not talking about 20 campaign pages, we’re talking about tens of thousands of products, if not millions of products, millions of assets, multiple channels, multiple languages,

variants for personalization, for instance, and increasingly, writing content for channels, for customers who’ve never even started their journey on your website. And that is the state of content operations today. If you’re running e-commerce, that is what content looks like. And if you think about it, there’s another thing that we’ve been learning, if you think about it, is that you guys, all our customers, You’re not just waiting for someone to turn up with an agent. You’re not just waiting for AI just to turn up and do the job for you. So you guys are building agents already, and some of you are buying agents already. And if you think about it, merchandising have agents for writing product copy, and the marketing teams might have some really nice cool agents or software for generating their images that they want. You have your internal teams building their own agents and getting excited by it, especially for technologists like me. You had translation

international teams using translation software, and increasingly we have agencies. We have SIs that are building their own agent platforms. And increasingly we also have lots of new vendors popping up that are providing specialist agents and specialist software for particular use cases, and these are really good. Some of them are amazing, like they’re really good. They’re making the task faster, they’re making it cheaper, and it’s really good. But that’s not the only problem. That’s not the whole problem. You’re not solving the whole problem. What you need to ask yourself is what comes next. Because once you have these new agents and you have this new software and you’re making the tasks easier, what you’re doing in many respects is creating new things for people to organize and orchestrate and coordinate. Someone has to, you know,

manage the model. Someone’s managing prompts. Someone’s managing a new supplier. You’re managing new brand instructions for all of these different models. And coming to that, you know, you’ve got, you’ve got a system where you had not necessarily the best content supply chain. Most people haven’t. And you’ve put these things in place to make things faster. But actually what you end up doing is fragmenting your content supply chain even further. And what’s really interesting about this is if you look at the scale in which these individual agents and tasks can do what you’ve actually done is made it even worse, because now you’ve got these isolated things running around creating all sorts of content and attribution and everything else at scale, and it’s not connected together. So you end up with a much bigger problem in terms of fragmentation. And if you think about it. What about consistency? Do each one of these things think about your product and your brand in exactly the same way? What happens if the models change? And we all know these models change they

change every five minutes, and they never work in the same way. So what happens when they change. And you have to update the models in all of these different places? What happens if the prompts change? What happens if the requirements change? What happens if the way that you display the brand changes and you’re going to do a whole new campaign? And what happens if the team that built it in your tech team got really excited and have gone somewhere else, and now you’ve got to maintain it. And then there’s a bigger question. Which one? Which one do you do first? What order does it come in? What context does each one of these use? How all of these things are going to orchestrate together and how they’re going to be connected together. And really, what you’re talking about is people having to do more things. People ultimately have to understand this and people have to coordinate it. So what you’ve done is you’ve made each of the individual tasks

faster, you’ve made them cleaner, you’ve got lots of really good content coming out. But ultimately, as an organization, you may have actually created a lot more things for your teams and your departments to coordinate. And ultimately, people are still the orchestration layer within these organizations where you’ve speeded up the tasks. But I think there’s something deeper in all of this, something really deep in terms of how we have been building enterprise software for a very long time, and that’s this. If you’re working on something, you’re probably looking at the screen, right? You’re looking at it like, just think about that for a minute. If you’re in a CMS, sorry, if you’re creating content, you’re in a CMS, right? If you’re managing assets, you’re in the DAM. If you’re translating content, you’re probably going to be in some sort of translation system. In fact,

if you’re doing any task at all, you’re in the system that is doing that task. And, you know, if the whole industry of enterprise software and building it has been on this actual assumption that you design the software because somebody is going to be looking at the screen, that actually the person who is operating it is looking at the screen, but that is no longer the case with AI. That is no longer the case. You don’t necessarily have to look at the screen anymore, and the person operating it doesn’t have the screen because before it was if someone isn’t, if someone isn’t doing this, if someone’s not looking at the screen and someone’s not operating the system, then there’s no work happening. But that changes with AI. And here’s another thing. Another question you should ask yourselves that feeds into this whole thing around enterprise software is if the AI is doing the work, why do you have to watch it? I mean, actually, I like to watch it sometimes because it’s good to see what it’s doing. But why should you have to watch it? Why can’t you do

something else? And that’s the question we asked ourselves is like, what? What should this look like? And this is ultimately the question we came up with, which was, what would a content operation system look like? What would a platform look like if we built it from the ground up. If we designed it for people and AI to work together, what would that look like? And this is what we come up with. There. I’m sure you’ve seen plenty of these bots in lots of other applications. Darren and I did consider it. Not really. And the reason why that doesn’t work is because another assistant doesn’t actually change the operating model. And a collection of agents isn’t a task either.

Right. I’m sorry. Isn’t an operation either. We have to do something fundamentally different. So this is what we come up with, which is the Amplience Content Operations Platform, and it gives you everything you need to create to manage to orchestrate and deliver the content you need for commerce. Shall I show you? Do you want to see some stuff? This is. This is a lot of work from these guys in the front here. But before we go into that, let’s have a look at something. This is probably a bit old world, right? A bit old school. I like old school. This is a wireframe of a page. And we’re going to focus on the banner there. The banner reason I’m going to focus on banner is because we’re going to create an autumn collection with this banner right now. Think about it. What’s the first thing you’ve got to do? This is like a quiz here for my customers. What’s the first thing you’re going to do with a headless CMS when you want to do some work? Yeah, you’ve got it. You’ve got to create a content type. Because unless you’ve got a content type, you can’t put

the content in right before. You would need to get a developer to build the content type, they’d have to look at that design. Work out what’s needed. Build it and then go. Yeah, that is a content type, right. And it has to match the design. But now you don’t need to do that. You can just write a prompt and give it the design. There you go. It looks much better on the big screen. And you can give it the design. Now what Amplience is doing is it’s actually looking at that design and it’s interpreting it and it’s designing the structure. It’s building the schemas, all those wonderful things that, as technologists like to talk about, is doing there in the background. Right.

It’s really understanding it. And then finally you have your content type already. So what would have taken, say, the best part of a day for a developer to create a content type? We’ve just done it in a matter of minutes. Well, actually seconds. And we’ve got a content type just from a prompt, which is really, really cool. But I wasn’t here to talk about tasks. I was here to talk about operations. This is really cool. It’s made that process super efficient. And not only can you do it from a wireframe, it actually works with Figma as well. So you can do it with Figma. So it is really cool. It is real task efficiency. But as I said, e-commerce is not a task. It’s more than a task.

So let’s ramp it up a bit, shall we? So let’s build an autumn campaign. And the first thing you need to do with an autumn campaign is you need to get the images right. You need to have some images to put in that banner. I’ve got a content type. Let’s put some let’s put some images in that banner. So I’ve asked Amplience now to find some images for my autumn campaign. Great. There you go. Now it’s going to add it to a collection. So anybody who’s using the DAM. This is a bit of a new feature there. Now you can now add it to the collection. Just point that out. I like that, but this actually shows you the actual realities of operations in content. Because look at those images. They’re not the same, right? They’re all different shapes, different aspect ratio, different backgrounds. It’s not very easy to work with, but that’s what it’s like in reality.

So what we’re going to do is give Amplience an opportunity here to do some work. And what we can do is standardize these images. And look, we don’t have to do one at a time. No. Come on. We can actually do lots at the same time. And of course we’ve written a flow in Workforce to standardize images. We can we can access that now and we’re going to standardize these images. So we’re going to take them all and make them into our sort of brand style. And there you go a bit of flickering there. That’s good. But now we’ve got the images right. What are we going to do next? It’s not really good for a banner. So why don’t we create a banner? We can use an image as a context. We can use it as the context to actually create what we need. And there we go. We have now basically asked it to create a banner image. And there we’ve got it right. This is a proper lifestyle image. I could see that on a banner. And I’m not a marketeer, but I quite like it. So now we’re going to use it in the banner. But of course it’s in the system. Actually we can do other things.

We can add some metadata to it, which is great, right? Because we might want to find it later on. But also this is e-commerce. So an image on its own is not really that exciting. So why don’t we make it shoppable so we can do it out of the box, make it shoppable. Now we’ve got an asset right from nothing. We’ve now got an asset we can use in a banner. We’ve made it shoppable. It’s all nicely enriched and a whole bunch of things we can do with it. But that wasn’t what we were really. That’s all nice and good. And you’ve seen probably these demos before and other things. Right. But we want to create a banner. Right? So let’s now create a banner. actually, I don’t want to create a banner. I actually want to create multiple banners. I want to create a banner for each of my markets because I have more markets. So what I’ll do is write a prompt and I’ll give it a brief of all of the markets and what I need for those markets using that image. And it will basically understand what we need. It’s understanding what we’re trying to do. And it finds that there is an actual workflow already for this and it builds it out. Now, the critical thing here that you need

to understand, real critical thing you need to understand here is that this is not some content that we’ve created in an AI platform, and now it’s stranded in the AI platform and you’ve got to go, oh my God, now I need to get a developer to write some APIs, pull it out, stick it, copy and paste it, do whatever you need to do right. This is actually living in the CMS, the place where you manage your experience. It’s doing it there. It’s not. It’s just seamlessly there. It’s just part of the same platform. Now we’ve got our banner. So sometimes you need to direct the system of what you want it to do. Right. Sometimes you want to go in and change the content yourself and do an action yourself. Sometimes you want to delegate it. You think, oh, I can’t be bothered. I’ll get AI to do that because it’ll do it better and I’ll get it done other times you need Amplience to bring you back in the operation, because you need your personal expertise to help make a judgement call on a decision. And this is the interaction model that we have designed the entire application

with this whole new platform. The whole interface is designed around this new interaction pattern of working with AI. Look, the CMS still matters. The DAM still matters. What has changed is the assumption that you have to do every single step, swapping between all of the applications. So that was quite exciting, right? It’s like we’ve now got a way in which we can get Amplience to do lots of work. Let’s move it up a notch. What about if instead of giving it lots of work, we give it an outcome. And what I want is to basically create inspirational landing pages that are GEO-optimized for a set of categories. So that’s what I’ve done. I think a little bit. So it’s found the categories we’re looking for Understands what we need to do. And if you look there, it’s actually found a flow. It’s found a flow that we’ve already built that would do this work for us.

Do we want to go with it? I think we do. Let’s just see what’s going to happen. Right. If we zoomed up a little bit here. Because at the time. But you can see it’s no longer a task. It’s no longer a single piece of work. What I’ve done now, what Amplience has done is it’s actually created multiple pieces of work, all from a single task. And if we move forward with that, you can see it’s going to go through. It’s going to process. This is all inside AI. It’s now processing those individual pieces of work. And again what have we got. We’ve got the inspirational landing page all in the CMS ready to go. And Chuck would be pleased about this, we’ve got some FAQ questions there. So it’s nicely done for GEO as it should be. And if you look at the top, it should immediately answer the question that you’re actually trying to represent with this particular piece of content, because that’s what you need to do if you want answer engine optimization. So everything is there and it’s all fully managed and you can change it. That was one single request and multiple pieces

of work, but it didn’t leave it locked away in the AI and the AI system. It actually delivered it as finished content, which is really, really, really exciting to me. It’s really exciting. These guys have worked tremendously to get this working together and it’s just it’s incredible. But that was really simple. Does everyone think that was really simple? I thought it was really simple. Like how like how hard it can be now to create content. Super simple. Really simple. But how simple is it really? How simple was it, really to get all of that work? Because there’s some things going on underneath. Let’s take a look. So what looks like a really simple request on the surface is actually a lot more going on underneath. Right. A huge amount. So we start the process and we gather everything we know within the business. What is the content that we already have? What are the assets we already have. What’s the product. What’s the attribution. You name it, we’ll gather everything up. So we’ve got all of that corpus of information, but that is only half

of the story. That’s only half the story. What you want to know is what’s going on outside of the business. You know, what are customers searching for? What is actually returning in search? What are AI engines currently surfacing? What are the questions that we should be answering? Those are the things that you need to really think about. And that’s what we’re doing in this part of the process. We’re pulling that information and that gives you so much more than a prompt, right? What it gives you is evidence and understanding that where the product gaps are. And as Chuck mentioned before, it’s fundamentally important that you understand where the gaps are, how you’re going to write content, and where the semantics are that you’ve got to create. And to me, this moves on to oh, there you go. This moves on to the next bit, which I think is the most important bit. It takes that context. It takes all you know about your product. It takes everything that you’ve learnt in the outside world and it brings it all together. So instead of saying nice, big prompt gone, write some content. You’ve seen those demos, right? “Write me some content.” This

does so much more than that. It goes what content should exist that doesn’t exist? What is going to fill that gap? What is going to make a meaningful impact on the journey that our customers have? It asks those questions because it has the evidence to do so, and once it knows those things, it can create the content and create all of the assets it needs for your customer experience. And the next thing is creation, I said, is just it’s just one task. Once it’s created, you can now go and enrich the assets. You assemble the page how you want the page and put it into the right content type and then put it into the content management system. Now as we’ve seen before, and now you’ve got content in the system that manages the experience without having anybody to worry about all of this piece. Right. So as I’m a technologist and a software engineer by trade, I like some of these kind of laws. Right. And this one’s particularly important for this.

And what it’s really saying is this is that in every application, in every piece of software, in every system, there is some implicit complexity that you just can’t get rid of, right? You can’t code around it. You just can’t get rid of it. It’s implicit. The only thing you can do with that complexity is to decide where it lives and who will deal with that complexity, which is what the operation platform has actually done in this instance. Right? Because we’ve moved the complexity away from the user, we’ve moved it into the operation, we’ve moved it into the agents and the AI. And ultimately, I guess the people who developed that first layer. But it just makes it simple and it gives you something that you can actually operate with.

So what we’re seeing is one simple request, an entire content operation underneath what looks like one single piece of content is, in fact a chain of research decisions. Management, governance and delivery. All together, all managed in a single unified operation. This is how I like to think about what an operation brings together. So it’s a little model that I use. So agents in this world, they perform the tasks right. They’ll create the content they do research whatever you want them to do, the models, ChatGPT, etc. They provide the reasoning, they help with making the decisions the agents are going to make. That’s the operating system for the agent. In some respects, the systems provide the context. It’s the systems surrounding it that provides it with the data, the interaction, the tools, the skills, everything that the agent needs. The system provides that

data and that content, that context and people. People provide expertise. And I don’t know if any of you guys have noticed? I’ve not. And I’ve tried not to do this. I’ve not mentioned the word human at all in this because although I might be the species of a human, I’m actually a person in reality. And it’s what’s really important is that we’re all experts in this audience today. We’re all people with deep expertise, and that’s what’s important. That expertise is part of the operating system. So talking about people, I’ve talked about reviews. I’m not sure you’ve seen anything yet.

And this is my favorite part of the demo. But I just want to get something straight with everyone first. Right? People are not in this process because the system is failing, right? They’re not. People are in this process because they’re providing expertise. They’re providing deep expertise because some decisions need expertise. And that’s what that’s what people are about. And this is what we have built. This is our ethos. This is what we built in the review screen. So let’s have a look. So in the review you no longer just do approve and reject etc. You can actually regenerate it right. You can actually now do some changes and regenerate. And there you go. I know you’re looking at going that was really quick. That was really cool. So guess what? I’m going to rewind and let’s see if I can get that to go back because I really like I like this bit. So as I’m doing the presentation I’m sorry. We’re going to have to watch it again. But look watch this.

So it goes in there. Right, click that. It goes in to that request. This is for someone to review and look at it and go, that’s not a student living room at all. I want a modern loft apartment. Boom, let’s regenerate it. And there you go. That image has now changed. I know you guys have seen lots and lots of vendors do room generators and stuff like that. This is inside a review screen. It’s regenerating it from a context. To me, that’s the most incredible thing about this is like, we put so much effort into making sure the reviews were for experts, not just for a human to press the button like the old Homer Simpson thing when he was in the nuclear power station.

It’s more built for people like us. People are experts who want to drive the operation. Yeah, you can see. Look, it’s got a chair. Da da da da. All the products you want that come with that room set, and there’s nice little workflow we could show you in the demo. Anyway, that was the bit I love. Let’s get that bit out of me. So, when it comes to agentic, the first thing I think we all need to understand is just because something is agentic, and we’ll keep hearing that word over and over again, it doesn’t mean it is uncontrolled. And actually, like I said before, the person in the process. So the human in the loop person in the process, I should maybe trademark that the person in the process is there because they have deep expertise and they can help drive the right decisions. And they understand your business more than you do. You understand your business more than the machine really does, and understand your brand and understand the emotional value of it and how you’re trying to reach people. And that’s why they’re in there. And that’s a fundamental part of it. Now, if you sat and you probably have sat

through lots of AI demos being wowed by and think, oh my God, this is amazing, how’s the world going to change? It will finish here. I’d be finished now. I could go great. And everything goes to Jeremy. I’d go to the pub, but it’s a bit earlier and I’d finish it right. But no, that is not what we tried to achieve here. We wanted an outcome. We wanted to launch this whole new campaign around inspirational pages. And we’ve got content. Look, we’ve got loads of it now, all of the categories, we’ve got assets that connect to the content, but it’s all enriched. It’s all amazing. So actually, I’m not going to let you go yet. You’re going to have to watch the next bit because what we want to do, and ultimately what we need to do is publish all of this content. I’ve done it in a direct view. This time I didn’t get AI to do it for me. I thought I’d do it myself. But there we go. It’s now published. This is now on a live site, and you can see the same sort of content, all the nice GEO and AEO pieces to it to make sure it’s all going to work and it’s delivered live, right?

This is not stuck in some AI system that now you’re going to go on and get the developers to work out how we put it in the CMS. It’s all there. And for me this is the fundamental difference, right? This is the fundamental difference about this because this demonstrates something that no one else is doing right now. It’s the difference between an outcome and an output. Right. This previously you see outputs. This is an outcome. And that’s the fundamental difference. So far you’ve seen me create content. You see me create assets. You have also seen me manage the content and manage all of those assets. And you’ve seen orchestration. You’ve seen how people how AI and how systems can be orchestrated together. And you’ve also now just seen to delivery each one of these four pieces, these four parts is all part of a single operation. You need all of these pieces together for a single operation. And that’s what our platform has done. That’s what we’re

trying to show you here. And if you if, you know Amplience from the past or whenever we started work together, I know we’ve got some customers here who have been with us a very long time. And I really, really, really appreciate that. It’s amazing. But if you look at where we if you come at different points in our lifetime, you’ll think of Amplience as maybe a headless CMS. For retail, for e-commerce, you would think of us as a DAM. You might just think of us as the, you know, the dynamic media infrastructure delivering billions of content variants all the time with huge scale. And increasingly, some of you might be thinking of AI and agents, and none of that is wrong. That is all true. That’s the things that we’ve been doing. But it’s no longer how we are describing Amplience that’s no longer who we are today. What we are is a content operations platform for commerce, and we give you everything you need to create, to manage, to orchestrate and deliver your content at

scale. Built on our fully scalable infrastructure that we’ve had from the beginning of time. So I’m going to leave you with a couple of thoughts. So you don’t have to listen to me ramble on much longer. But I strongly believe that the winners in this whole new world about agentic commerce that, you know, Chuck, has articulated incredibly well. The winners won’t be the ones who can just generate tons of content, right? The ones who can just keep generating loads and loads of content. In fact, if you do the wrong content, you’re probably going to lose. They won’t be the winners. The winners are going to be the ones who can orchestrate that content better. So on that, I would just like to thank you all for spending some time looking at this new, amazing product that we are. As you can tell, tremendously excited about. And again, thank you and I will see you all later and enjoy the show.

So our next speaker is going to give us an inside look at how JD Sports are thinking about that exact foundation, understanding exactly who they want to connect with. Building content that genuinely resonates, then using AI to take it further. So please join me in welcoming JD Sports’ Director of Digital Trade, Craig Egan, to the stage. Okay, so Amy, looks like the first issue was the font didn’t pull through, but we’ll rock through this in a content first style. And we’ll ignore the fact that my font being typical JD, everything has to be on brand. We have our own specific fonts. It’s all good. It’s fine. We’ll work with it. Amplience and the team asked me to be here today to basically run through the journey, as they called it, that we’ve gone over the last two years trying to undertake our content strategy into the world that we all now know as e-commerce professionals. I call it

a headache. I think most people in the room would have called it a headache, but we’ll use journey because it seems like it’s a nicer term. And content for us is the thing that connects our brand partners, 64 of them that we sell across sports and fashion with our end point consumers, and it’s got to resonate. That used to be a lot simpler for us in the old, in the old times that we’ll all remember fondly. Now it’s a lot more complex. So to show you the kind of ethos of how we want to go about and tackle our content strategy, I thought, I don’t want to sit here and talk to you guys about it. So I’ve created a bit of a video, content first, which I want to run you through now, because I’m assuming that most of you hopefully will have heard about JD or know about JD. But none of you all really understand what goes on behind the scenes when it comes to our content strategy. So I’ve got about a 52-second video for you to enjoy, but hopefully we’ll bring up the energy in the room a little bit as well.

JD Sports know that youth culture isn’t created by adults set in boardrooms, but by young people at ground level. Our role is to propel youth culture forward on every corner of the world. Forever Forward will also power everything we do next. From how we treat our team to our lens on the biggest events in sport, music and lifestyle, Finding and elevating the next movement in fashion and culture to a visual identity that shows up wherever we go.

The future has always been shaped by those who refuse to stand still JD. Forever Forward. So as the guys mentioned, I’m looking after everything when it comes to trade content, merchandising, copywriting and content. Obviously at the heart of all of that. I’ve been at JD for about ten, just over ten years now, And my job when I first joined. Used to be a lot more simple. JD has been on the high street for 40 years, predominantly in the UK. Super organic level of market saturation. We had a very simple job to get our brand and our products into the hands of consumers. Shoot a campaign, keep it clean. Keep it consistent. Keep it on brand, take it through the segments. Be super targeted. Work out how you want to resonate with in one geography a lot simpler, and then work out your marketing spend and how you want to push that through the channels to make sure that your end content hits the consumer exactly when and where they need to see it. Now, through the success of the UK, they decided that we want to give Craig and the team a slightly broader remit. So over the last couple of years, we’ve decided that we want to

scale and really go after Europe. So now that content strategy that used to be super linear and relatively simple in today’s terms anyway, now spans across 13 European markets where brands, trends, weather, platforms of choice all change. And if we wanted to shoot one simple campaign, that’s out the window. Our costs, as you heard from this morning’s sessions, then quadrupled and content cost became a real challenge. Resource didn’t grow as fast as the markets grew naturally being in a shareholder owned business. But we needed more outputs. My job in the team’s job was to make sure that these outputs weren’t only on brand, but were also being seen at the right part of the journey for all of the consumers that we wanted to connect with. Naturally, we didn’t have the budgets just to go after a whole market, France, for example. We went into it first with a lot of content we produced in the UK, controlled the brand tone of voice, controlled the outputs, made sure that we tick the boxes from the Nikes, the Adidas, the Under Armours to make sure that we were representing their brand in the best possible way. And guess what that content did

when it hit the streets of France, flopped, didn’t resonate, got very little cut through, and the more marketing money we pushed at it to get it out there just cost us more money without the conversion that came through off the back of it. Terrible. So when it came to then analyzing the new world, which I think the team of analysts did a really good job of kind of identifying this morning. It’s messy. There are thousands of points of entry now into a customer’s journey. We’re somewhere in that middle part where customers can now consume content on social platforms and checkout directly. They can shop through GEO, they can shop through LLMs.

It’s a pain in the ass. So how do you manage to break up 64 brand partners across 13 territories produce content for it that is entirely on brand for what you want to stand for as a retailer, and get it into the hands of your consumers at the right part of their journey. And that is what I wanted to kind of talk to you about today with the Amplience guys’ support, because they’ve jumped into this journey with us right from the very off. I thought we were going to scare them off to be fair to them. So internal credit, they got right on board and they were looking at How do we take what JD currently do and give Craig something as a tool that’s going to scale what we can do, but keep the integrity? So this is the question that I kind of asked the teams, but I also asked the guys at Amplience as well is how do we create content at scale that connects with our consumers in a way that people expect JD to show up for the markets that we’re known in, but grabs attention for the markets we’re not known in. Oh, and FYI, I wanted to make sure that none of this is churned through AI as slop and it stays on brand.

That was the question we took to the teams. That was a question we took to Amplience. So we’ve been partnered with Amplience now for, I think it’s just over, again, 10 to 12 years. And as John said earlier, we used to use them solely as a PDP asset depository, manage all of our PDP assets, housed them all in one place. And over the last two years, as we’ve turned our tech stack into a more headless one of the future, we’ve decided to go all in into that CMS and we create all of our PDPs and studio in-house, all of our campaign artwork and imagery in-house. We have third party brand campaigns, and by that I mean these are the brands and the campaigns that are driven by the brand partners that we have, that JD influence. But ultimately they control and then the influencer assets and the content and the partners and the talent. I’ll talk about later that we also create all fed into Amplience as a CMS, all work through a lot of AI workflows.

That team can manage in one place, so I can get the outputs that we need to show up on every single one of the platforms, and all the different content types that you can see there below. And in that ever changing world of what good looks like or what gets cut through, we’re able to really drive a specific and bespoke workflow that can allow us to basically take the strategy into single markets, to really bespoke customers in communities that hopefully resonate and test and learn. If it doesn’t. And I think that’s the beauty of this. Every single thing that we do before we even get anywhere near briefing content, has to go through these three areas. Head office. Before we sit down with anybody and we start briefing or creating or shooting or driving costs. I asked the team to have a look at with this specific campaign or this specific request and this specific market. Here’s the audience. Do we understand them? What are we trying to achieve with the content and who we’re going after? We know with my France example.

Spreading things too thin and trying to get the whole market doesn’t necessarily work. You burn through cash and funds way faster and your reach doesn’t go as far. There’s a little phrase that we use at work, where you can’t make everyone like you, but you can absolutely make a tailored audience love you. So we want to focus on those pockets. Where are we going to meet them with this content? What are we driving here? What is this a social play? Are we going into a physical space with a community? Are we going to do this through our own channels? And we’re going to do this from on a more global level with a brand partner. So where are they? How are they spending their time? Where are they going to discover these content or this content? Where are they discovering trends and how do we interject ourselves into that journey? And I think someone mentioned it earlier, can’t remember who it was, let’s meet them where they are. Let’s not go and try and bring them to JD. Let’s take JD’s product and brand proposition and service level into where they are, and then we tailor our storytelling to make sure that we can amplify whatever we’re doing with the best possible outcomes, all on all within our own brand, DNA and USPs. But we do it through tactical content and tactical partnerships, because that gives us that leverage in markets that we’re

not well known or we’re not as trusted as maybe locally local individuals, talent or influencers. So from a JD perspective, our target audience has been refined. It’s pretty simple. We have one key set mission from a consumer perspective, and that’s to win with the youth in every market, in any region that we go after. We want to win with the youth and we want to understand who they are. We want to define the youth as a 16 to 24 year old just for this room, just so everyone understands they’re the consumers of our product. We also have to create content that connects with the customers as well. The people who are buying on behalf of the consumers, but the people we really want to get to with our content is to influence the youth. So from a three pronged attack, we do it through cultural integration.

We do it through community driven first lens, which I’ll give you some examples of in a minute. And everything is underpinned by our sport, music and lifestyle pillars because that’s what JD plays in. We don’t want to go too broad. We know exactly where our sweet spot is and where we have authority, and we know exactly where we do. So the question then became meet them where they are again 3 or 4 years ago. Really, really simple. There was limited channels. There was limited ways of discovering. And the customer journey from discoverability to hopefully transaction was a lot simpler. Now it’s not. We undertook a load of research with the guys at Kantar and Google a few years ago, and had a look at some stats around where are these customers browsing and discovering products.

And I think, again, John mentioned earlier, where are they, where are they spending their time? And I was astonished to find that the shift away from traditional media into short form video, the likes of TikTok has now overtaken Instagram in a much, much shorter period of time. And then YouTube, with the average 18 to 34 year olds spending over 125 minutes a day on YouTube, which gave us a great deal of insight for how we were going to potentially penetrate that market and that platform from a brand awareness perspective. We’ve always been really good at that lower funnel, creating content that drives that final click when the customer already knows what they want and they’re searching. But now customers are transacting in different places. They’re discovering their trust is all over the place. YouTube is where we decided, do we want to show up ourselves and be a retailer that says, hey, buy this product? It’s really cool. No, there’s no authority in that. And I think there’s no there’s no we can’t be genuine. We’re a retailer. Everyone knows what we want. We want them to buy our shit. Obviously. So we partnered and we took this insight and we worked out. What does that look like? It used to be celebrities that we used to use endorsements for.

Now we find local creators. We find local people with authority in these communities who can tell our story for us, share about the products that we sell. Creators really are today’s celebrities, and this one shocked me when I first got it as well. So 98% of online viewers are more likely to trust a YouTube creator or an influencer than they are a traditional celebrity. Gone are the days now where we use the likes of professional footballers in the Premier League to push some of our content stories, or musicians or athletes. We partner. It’s a hard it’s a harder one to kind of get right, because you need to make sure you find influencers and talent that genuinely align with your brand proposition so that they can create the content the way that they want to create it.

That still resonates with the consumers you try and target, and therefore it feels more authentic and genuine when it hits the end consumer. But this really did shock me when I first got it. Also this as well. And again, this goes back to trust on these platforms. And I think we deal in that 16 to 24 year old market where they’re the most fickle, but also most tech savvy consumers that we possibly deal with. Gone are the days where there’s loyalty to a retailer, there’s loyalty to a trend, or there’s loyalty to specific products or wanting to be cool. We need to show up at the right place at the right time, and over 70% of the content that we produce now isn’t actually produced by JD Sports. It’s produced on behalf of JD Sports from a myriad of network of influencers across those 13 markets, who know how to show up in the markets that they work in with their communities, from micro to macro, to make sure that we can be in the right place at the right time. We still have to approach these communities differently. We still have to show up in a physical way as well.

We partner with local sports events like Ball Out, a three-by-three basketball tournament that happens in the likes of London, Liverpool, Manchester, Paris, Amsterdam, for example and partner with some of our brand guys. So we’re in a physical space and the give back is there. The alignment with the community is there and the understanding is there. We amplify the voices. We partner with running clubs. If anyone knows anyone in the running club, they’d love to talk about it. It’s the perfect thing to kind of amplify their voice alongside our brand. No offense to those in the running club. And then festivals as well. Like music festivals, like the spots that they’re already wanting to go to, where they want to be seen, where they want to engage.

Our brand can align with these things again in that sports and music and lifestyle pillar. And like I said before, everything had to be done through a tailored storytelling approach. We couldn’t get away, as much as we wanted to, with shooting one campaign at one cost, with real people in rainy Manchester, and expecting that to hit the streets of Milan when it’s sunny and warm and have the same cut through. The trends are different. The climate is different and people want to feel like they’re being heard, seen. And we understand as a retailer what is important to them as a consumer. So the three main areas and the three pillars that we do everything through is JD campaigns. They’re always on. They run on six week increments. These are the ones that we think about. We create, we shoot. We tailor what our message wants to be to the consumers.

We have our brand campaigns that give us that global voice, that global reach. Partner with the likes of Adidas here in this minimal example to make sure that we’ve got that authority as a retailer of choice when it comes to who the brands want to be stocked in. And then the most important one over the last two years, like I’ve mentioned, is local creators. They’re the people that are really going to drive that message into the bloodline of everything. That is a heartbeat from a community perspective. They give us the authority at that real local level with a little bit of product seeding. They create the content. We give them an idea. We let them run with the brief. There’s very little guardrails in terms of that, because we’ve handpicked a set of creators who are brought into the JD family to make sure they understand what we’re trying to achieve. And then it’s over to them.

Everything that we do now is still shot from a UK perspective, but we’ve added a European lens as well and that gives us, with Amplience, the content creation to be super focused when it comes to regionalization, whether that is across Europe or even further afield. Thankfully, I only have to look after Europe. It’s already a headache. I don’t need to worry about any of the markets that we’re in globally, but it gives us that ability to churn this content out at a massive scale format to what we need to, where we want to show up and get the brand campaign that we want to live and breathe, living for longer in the six week increments that we continue to roll this out.

But we went one further last year, and all of the data that I showed you around how customers resonate with content now has changed. And they don’t necessarily always just want to see flashy brand adverts pushing a product that’s clearly a commercial sales target. We didn’t want to create content in a boardroom anymore, which is one of the things that we kind of challenged ourselves on. So how do you actually give the youth a voice to amplify their communities through your content. To build a meaningful connection with our business. The most commercial time of the year was Christmas. So last Christmas we decided, let’s get rid of Anthony Joshua. Let’s get rid of Cole Palmer in the mainstay. Let’s not utilize some of the big names that we’ve used before, paying millions of pounds to shout for our brand.

Let’s just give 286 kids an iPhone for three months and time to go and capture their lives in the product that they live and breathe by. So I want to show you this now because this was a risk. I’ll be honest. Like we didn’t know what was going to come back with this. But I’ll give you a little snippet now of if anyone who didn’t see it. This was mobile first. It was only published on YouTube, and it was shared more than we’ve ever had before because genuinely, the people you’re going to see in this aren’t paid actors. These are people from the communities, people who live and breathe sports, music, lifestyle and just celebrating the youth rather than criticizing, which I think happens a lot in the press these days. So.

You just lost and lost in our own worlds. I’m Mario. Getting my GCSE results in a week. So take me back. Yeah. Go show them. It’s a group sport. I’d love to become famous if that’s not a thing. Probably just. I’m a pharmacist. Who can. Fail. You can make mistakes, but, like, really? Who cares? Can I ask you a question? Please ask it. I’m scared. I don’t know what.

Wasn’t the full clip, but if you do want to see the full clip, it’s on YouTube. But what I’m really proud to say about that campaign is that was shared and viewed over 52 million times organically on YouTube into so many different streams that we could then re crop and make it live for so much longer than just a Christmas campaign. There was no Christmas trees in this. There was no fairy lights. There was no present given. This was just a celebration of the youth. And I was putting our trust in them at the most commercial time of the year. As far as our peak goes, we put our money where our mouth is. Like I say, huge swing could have massively missed.

Thankfully it didn’t. And I think that advert from memory was third most watched Christmas advert or advert at the time on YouTube, behind the big boys like John Lewis, which I’m not going to compete with anyway because Christmas is theirs, isn’t it? And then this is really the only bit that I wanted to talk about in terms of AI, because I felt like there’s been much better voices in the room. From a technical perspective, who would talk to you about that content scalability in AI? So I spoke to Brían and the team at Amplience and said, well, how do we how do we update everybody from a JD lens without me giving away some of our secret sauce too much? Otherwise I’ll end up in jail at head office. less theory and more practical usage for how we do this. So AI is definitely the answer. It’s not the only answer. It’s one of them. And we wanted to control the AI. We were very risk averse two years ago when we started this journey, when it came to AI and content creation, specifically from an ethical standpoint. Very, very, very concerned from a genuine kind of content creation standpoint. Did we want to go down this route? Could we trust to go down this route? Could it keep on our brand tone. So we brought it in-house basically, and we

worked with the team to work out well. If we don’t want to go to an agency and outsource this level of responsibility on the one thing that underpins our brand, which is our content production. So we built a model internally and it is the answer, but it’s definitely not the quick answer. And I just want to walk you through this kind of process now. So we didn’t even start by shooting any product. We took what I like to call as the iconic duffel bag and we thought, right, well, can we put this to the test on something that is less sensitive when it comes to how do we how do we build an AI model that can do genuinely genuine quality content creation from an AI perspective. This process that you’ve seen in front of you now took six months to reach maturity just on one single carrier bag. So to get to a point where we could get a sign off from some of the studio guys I’ve got in the room today who are our brand aficionados when it comes to content creation, and winced at the first point where we said we wanted to use AI to scale and make our lives easier and really bought into this process. So can it help? Absolutely. Is it quick?

Absolutely not. If you want to do it right and have control yourself, but when you get it right, you can start having fun with it and playing around. Now we can take shots so we can again ask the guys that I showed you in there in their demo earlier. Different backdrops, different content styles, but the quality of the product and the way that the AI model understands the textures, the specifics of the shoe meant that we could show these to our brand partners, and they thought we’d actually shot these ourselves. They had their product guys go through with the fine tooth comb around brand marking, stitching, textures, materials and couldn’t necessarily detect what was AI. Not these specific ones, obviously, because no one’s hung a shoe from a from a telephone line. But we showed them a lot. We did a bit of a guessing game around AI or not. Quite fun, actually, to watch them kind of scramble to work out what we created with AI and what was genuinely created from a photographer.

And the process is pretty intense. It doesn’t necessarily solve all the resource problems at the studio, which the studio guys were quite thankful for because they thought they were going to get replaced by AI workflows. But it just meant that the production, the photography, the videography, all of the design, the stylists and everything else were brought much further up the process. So now the art direction, the styling happens as a brief or as a prompt. We build that in to everything we’re doing in the CMS and everything that we do with our AI model internally, we can then utilize real models. So we’re not having to use AI models all the time, which again, ethically isn’t great. And dress them up in a myriad of different ways. And then we can also retouch and change the backgrounds out and make sure that that finished article is as polished as it would be if one of our photographers in the studio went out onto the streets with a real model with the camera, took it, sent it back to the studio, retouched it, reformatted it, and then published it back into the CMS, which again, we haven’t got the time for anymore.

To John’s point earlier, and Jeremy’s point earlier, we need this done in less than six hours now. Fashion is fast paced, trends are fast paced and we need to move with the times. So as an output for you guys, like, I just wanted to give you a few different examples of taking one physical shot that was done, which was the original photo you can see here. And then being having the ability to swap this out. So if I’ve got a new balance trend happening across different markets, I can still take that one shot, swap out the trainer, make it resonate from Liverpool to Manchester, Paris to Marseilles and all the way over to Austria if I so want to. Different brands, different shoes. One original shot. Simple cost.

We can adapt the same image from an outfit perspective, so we can produce content that will live for longer. Weather trends. All of a sudden it starts shooting down in Manchester, which is basically every day. And therefore now I need them. Now I need assets to push outdoor clothing, weather protection, clothing doesn’t make sense to do it in the shorts that she was wearing in the original shot. And now we can swap it out. We can put leggings on her and move forward. And again, these are single obviously options. We have a thousand of these running at every different time to automate this process and make sure that we’re on trend, on brand and resonating with the right consumer in the right market.

And then the real power play for us was really amplifying everything that we did from an advertising platform perspective. So taking that one campaign shot, distilling it down into something that we could easily retouch and reuse for any ad platform, whether that be lower funnel, middle funnel or upper funnel, and even taking the static shots that we used to have and turning them into motion. So they just have a little bit more depth when it comes to whether it’s TikTok or Instagram Reels, for example. So my takeaway for you guys really today is utilize AI for what your teams already do really well. We create content really well. The teams are super proud of that. We didn’t want to give that DNA and heritage and knowledge and depth that the guys have got over to something and just to run away, so we fine tune it with the right recipe, the right ingredients, and make sure that that can be then amplified out across AI so much quicker, so much more streamlined and into the channels that my advertising teams and marketing teams needed to be able to push that message out.

I’m around for the rest of the day if anyone wants to chat about this stuff, I’m pretty excited to talk about it. But for now, thank you and enjoy the rest of your day.

So please join me in welcoming Chris Annese, Senior Vice President of Sales and Partnerships at Amplience, with Jez Wilson, Group Ecommerce Director at Midwich. Good morning everyone. Hope everybody’s having a great day so far. So I’m really thrilled to introduce this next session, this thought leadership session with Jez from Midwich. Jez, why don’t you start by telling us a little bit about yourself, your role at Midwich and your background That’d really be helpful to start out.

So, morning, everyone. Yeah. My name is Jez. I’m the group ecommerce director at Midwich. I’ve only been there two years, which makes me a newbie in the company, I think. Average is ten years, 20-odd years. Fantastic company. Over the last 20 years, they’ve grown through acquisitions. So it started off as one of the biggest distribution companies in the UK and has slowly acquired companies around the globe. We’re now 1.2, 1.3 billion, see how the year ends up. When I joined, it was very fragmented. So even though we were a group, we act more as a federation. So all the acquired businesses operate independently.

And one of the approaches we want to take is trying to centralize ecommerce to roll out new channels for sales. So that’s what we’ve been doing over the last two years, putting a new stack. And we’ve rolled out ten sites live in about a year, so we’re making progress. Cool. That’s awesome. Great to hear. So, you run content across a lot of operating companies, right? And brands in different markets. When you first looked at kind of bringing this all under one roof, what did good actually look like for you and the team? Yeah. I think we’re still asking what good actually looks like. I mean, we’ve got we’ve got 1,800-ish brands that we work with globally. So there’s a lot of content that needs to be created around those. And then of course, each market has a different go-to-market strategy. So we kind of serve two purposes here. We need to satisfy what the group needs in terms of group global brand relationships. And then local go-to-market. So we needed to enable a very much kind of like a group ownership of what our sites and our structure looks like, whereas

the same time giving flexibility to the individual businesses to launch and trade their product. So we ended up in a hybrid operating model. So it’s a very new world and we’re still pushing it through. Yeah. Interesting. So just to give a bit more context, so you work with major manufacturers like Sony, LG, Samsung and each of these manufacturers have their own brand identity and expectations. How do you keep that intact while still running everything through one platform? Is that like, how do you guys do that? I think we’re learning as we go at the moment. We, you know, the brands themselves are kind of quite interesting. There’s very few that actually have a global approach. You do have certain brands like DJI drones. DJI drones are very strict on their roll out. Everything needs to be signed off, which is good because that means we can create content at a group level and then push it down to local level. And then of course you have the local markets. So Samsung for example, they’ve got lots of different kind of marketing businesses just within Europe. So yeah, it’s a it’s a hard approach. Luckily, our head of group marketing should be able

to tell you about the kind of chaos that we live with. I bet. Yeah. It’s a it’s an interesting time because, you know, you’re not you’re not juggling your own brand and how you want to be represented. You’ve got to respect everything that the individual businesses want. Yeah. Yeah. Great. So each company runs its you know each operating company runs its own content but stays part of one connected system. How do you give them, like, enough independence so that they can move fast locally, while still getting the benefits of operating as a business? Yeah. I mean, I think this is where Amplience really kind of comes into its own in the fact that we are able to create global content and then use the AI tool to push it down to the local businesses, and they can then repurpose that content and, you know, localize it, translate it, which increases the go-to-market. But at the same time, the local marketing teams have full autonomy of control to create their own content. So you can end up with, you know, really nice global brand pages that are unified across. But then, you know,

Dubai can have completely different content to Canada. So I think it you need to retain that flexibility if you’re in kind of a particular business that we’re in. You know, you have a product drop that’s going to drop six months ahead of one market to another market. And that’s great because you can learn from that content and then share it down. Yeah. Makes sense. So I’d like to pick through the localization piece for a little bit. So you guys create content in Dutch translated to English and vice versa. What does localization actually look like on a day-to-day basis for the people that are actually doing the work versus how it kind of sounds in theory?

Yeah. So we made I mean, if we take product content as an example, we made the decision that we were going to originate all product content in English. The main reason for that is because there’s just exponentially more content available in English than there are, perhaps in the local languages or even provided by the brands. So we do a lot of our origination in English, put it into either our PIM or our CMS tool, Amplience. And then we do the we run the AI tool to kind of do the base translation and then hand it to the local teams to actually check because, you know, AI translation is great, right? But only to a point. It’s there’s a big difference between local language and machine translations.

Yeah. So they’re looking at dialects and things like that. Correct. At the street level. Yeah. Correct. I mean, you know, take Italy for example, how somebody would translate something in north of Italy to the south of Italy could be radically different. Yeah. I say that from experience, having lived there for five years, it’s very different. Yeah. And it’s the same with Dutch. You know, you can’t. Language evolves and changes over time. And as good as the LLMs are, it’s not native language. So. But the thing is, you can speed up your content a lot quicker if you give somebody a base translation rather than say, hey, translate from English to your native language.

Yeah, great. So, you know, Craig talked a lot about, you know, B2C content, delivering B2C content. You know, your organization is a B2B organization. Right. So I guess, you know, for the audience here today, like help us understand what the difference is. Right. Is there a difference? I mean, fundamentally I don’t think there’s a difference. Like, what is the job of content is to recruit and engage customers. So you’ve got your whole acquisition pipeline, you need content for acquisition and then you need content for retention. So at the top level, content needs to derive the same strategic job. And that is growth of sales within the company.

I think the difference is, I mean, particularly for us, like we’re a highly technical company. So yeah, we are a distributor and our job is to shift a lot of boxes very, very quickly. But we’re also probably one of the world’s experts on audiovisual installations. So, you know, we need to know as much about installing a studio like this. Hopefully they use us. If not, we can contact, all the way to something like Twickenham Stadium, Albert Hall, Oasis, Adele concerts. So our expertise is probably unparalleled in the industry. So therefore that is a lot of content in terms of what are you talking about, a product? Are you talking an install?

Are you talking about a Microsoft teams in an office, or are you talking about the stadium? Right. So not much difference. No, not much difference. Yeah. No. Makes sense. So, during our prep call, you know, we covered a topic around. AEO. Right. And I remember the term you used. You said it was a nightmare when I asked you about it. Right? So maybe you could elaborate on that a little bit. And kind of how AI, how Midwich is approaching the AI discoverability topic, because I think it’s something that, you know, a lot of the folks here today have on their mind. Yeah. I think the thing that keeps me awake at night is how do we actually approach this topic?

And, you know, I have to rely on my history of working online, I mean, and what I’ve been in the industry since about ‘94, and obviously you’ve seen a lot, a lot of changes in that time, whether it’s browser versions, internet speed or, you know, who’s going to win search. Is it going to be Yahoo? Is it going to be Bing? Is it going to be Google. And now we have AEO, which is a joy. I mean, I was just looking the other day, there’s what 348 LLMs. Well, great. Which one am I going to target? So, you know, if you want to approach content, especially text based content, which one are you going to go for and when are you going to go for it? So I think from us it’s very much a longer term strategic view. And I know that we had the conversation this morning. It’s like you’ve got to be on it now. Well that’s great, but which one.

And is that going to be relevant in six months time? So I think for all of us in terms of how we’re approaching this, I don’t know, it’s to me, it’s sit back and relax a little bit, be mindful of it, but come up with a proper strategy. So our strategy is very much dual engine. So engine one is just, you know, your business as usual. Get your Google right, get standard search right and then build into a content strategy. There’s no point in rushing into this because if you do, we’re going to make big mistakes. It’s an interesting topic right now. So what are your top ten LLMs?

Dude, I don’t know that, like, today or tomorrow, I mean, I’m yeah. And I think the interesting thing for us, I mean, just as a side note here, we’re actually actively blocking LLMs. We’ve had our sites go down six times because of over exuberant traffic from training agents and everything else. So, you know, again, we’re talking about a strategic infrastructure approach, not just a content approach. So we’ve got people saying, oh, you need to do AEO. It’s like, well, yeah, that’s great, but we can’t allow them on the sites at the moment because they’re bringing the sites down right now, whether that’s price scraping, because obviously price scraping in a competitive distribution market is intense, like really intense. And then you have all the content that goes with it. Well, okay. I need to put in another cloud, I need to put in another ten servers just to cope with AI. Who’s funding that?

Interesting. So you mentioned beforehand that, you know, the biggest challenge in getting, you know, on Amplience and working with the organization wasn’t necessarily the software. Right. Getting everything together. Right. It was it was actually you said culture. Yes. Maybe you can unpack that a little bit for the room. I’d like you to kind of. Yeah. Peel beneath the surface there. Absolutely. I mean, for starters, I’ve got to say I don’t think we’d be where we’re at if it wasn’t for Sumil and Chi kind of holding our hands along the way. They’ve been absolutely instrumental in us being able to get off the ground. But I think it goes back to, you know, we are a group that is more like a federation and we’re culturally not used to working together. Everybody is, has their own ways of working, and especially as most of our founders are still running the local businesses, they’re very much in the mindset of my way. And it’s right because they know how to run their business. And then we stand back and go, oh yeah, but you know, we need to globalize this content and we need to centralize our approach. So I think we’re in our digital maturity. We’re getting there, but it’s going to be a long time before we get it right. Yeah. But software, the software has

been great. I mean, Amplience has definitely allowed us to achieve what we need to do. Yeah. The cultural piece is big, especially, you know, across multiple markets, you know, in that manner. So, you know, if you had some advice for the people here, for the folks that are here today and trying to scale content across multiple brands and multiple markets. What would that advice be? I think that’s hard. Everyone’s different. Our approach was very much zero to one. So we took the great old Peter Thiel thing of like, you can’t go anywhere until you’ve gone to one, and if you go from one, you can go to two. So that was our approach, rightly or wrongly. So we took the foundations and we said, okay, in order to start this journey, we have to start somewhere. And that was basically get the sites up and live, start thinking about how we share content. And then gradually build on that. And I think we’re going from, you know, step one to two to three to four quite rapidly. We’ve gone from one website live to ten websites live. We’ve started to share some content across them. So build it up slowly. But without that strategy in the first place, you’re dead in the water. Yeah. And again,

going back to how, you know, what is your strategy and how are you going to execute it? Yeah, I think Craig did a good job talking about that earlier. Yeah. You know, it starts with the customer and what our goal is. Right. What are we trying to achieve? Absolutely. Absolutely. Where are you trying to achieve it? You know, there’s different maturity all across the globe. If I’m going to start a strategy for a global company. Well where do we start. You know, do I start in Singapore or do I start in Canada? I don’t know. Yeah. It’s difficult.

It’s really hard. It’s really hard. I don’t think anybody should underestimate the job that needs to be done here. Yeah. Cool. Well, thank you so much for being here today. Really appreciate it. Appreciate you spending some time up on stage. And if anyone has any questions for Jez later today, we’re happy to spend some time together throughout the day. So thanks so much. Yeah, and we got a couple of great people from the team here as well. So they can kind of. Yeah that’s. Right. Give the full exposé on what it’s really like. Awesome. Thank you, Jez.

Welcome back. I hope you’ve been fed, caffeinated and solved at least one major commerce problem over lunch. Because this afternoon we’re getting practical. AI conversations have a habit of gravitating towards the impressive stuff, but some of the biggest opportunities are hiding in the operational work that slows teams down. So what happens when you stop chasing the big unmanageable, impossible to get sign off project and start with the problems that actually need solving? So for our next panel, please welcome to the stage Isabella Barbato, VP of demand generation at Amplience, with Jayne Bibby, senior content manager at Primark, Anjina Solanki, lead e-commerce production manager at JD Sports, and Gus Iwanaga, general manager, product user experience and engineering at commercetools.

Hello, hello. How is everybody feeling after lunch? Good. So good afternoon. I’m Isabella Barbato. I lead marketing at Amplience and I’m very honored to be hosting this panel today. So every AI in retail panel this year will promise you the future, but we are going to do something a bit different today. So two of the people up here are actually building this right now in production for millions of live shoppers. And the third has watched many other retailers try. So here’s the pattern. The biggest AI opportunity in retail right now isn’t the futuristic shopping experience. It’s the unglamorous operational grind, the merchandising, the legacy integration, the day-to-day process.

And the retailers already ahead didn’t start with the most exciting or glamorous use case. They started with the messiest, most manual bottleneck. So today I’m joined by Jayne Bibby, senior digital content manager at Primark. Jayne leads content operations for Primark e-commerce site, including inspirational content, SEO copy, imagery, and has spent the last two years piloting AI across the pipeline. I’m also joined by Anjina Solanki, lead e-commerce production manager at JD Sports. She leads studio content production for JD Sports e-commerce estate and connects that studio output back into the wider business. And last but not least, I’m joined by Gus Iwanaga, the general manager across product, UX, and engineering at commercetools.

Gus also leads commercetools’ zero to one incubation team looking at new products and new businesses and where AI actually changes commerce. Prior to that, he spent 15 years at Google and Zalando in search and retail at scale. So I’m going to start with three adjectives. Cautious, experimental, or all in. How would you describe your organization’s appetite for AI right now? Jayne? I would say we’re optimistically cautious. Okay. And I’d say we’re a mix of everything, but optimistically experimental as well. Okay.

And you? Go big or go home. We want all of it. I like that. Okay. So let’s get into specifics. How did you actually start? So I’m going to start with you, Jayne, first. So Primark didn’t start with anything glamorous. You did a two-month pilot on product descriptions, alt text, and SEO copy. Why did you start there and how did you know it was the right decision? I mean, I think no surprises. We started with where we thought we could make big time savings for the team. We really wanted to look at where we could strip out those really admin-heavy elements of their day-to-day to allow the team to do the job they were hired to do in the first place.

And we wanted to see how we could create content at scale with the same headcount we had today. Much like what’s been said already on the stage. No one’s teams are getting any bigger. So yeah, we rolled out a two-month proof of concept working with Amplience. So we are platformed on Amplience and have been partnered with them for many years. So it made perfect sense as they were rolling out Workforce for us to trial and test that with them. So we looked at alt text, SEO articles, and product descriptions. And how we ran the process was Amplience came into our office. They did a huge training session with about 30 of us.

Some of those people had used Amplience already and were doing so day-to-day. For some people it was brand new. But we wanted to give everyone the tools they needed to do that proof of concept. And then we asked them to add it into their day-to-day to be testing this alongside what they were doing already to really see how it worked. We then did weekly face-to-face feedback sessions with Amplience so that we could explain what worked really well, where there were issues and things weren’t working quite as we were expecting. Where prompts needed to be tweaked. Where more training was needed. And it allowed us really quickly to see whether it was going to work for us or not.

Amazing. So you started small and proved it first. And you went the experimental way. Can you tell me a bit more about it? Yeah, I think we started, especially from a studio perspective, as a way of seeing what can we push, where can we go, what are the limitations that we’ve had in the past, and how can we use AI to overcome them. And I think it’s been really successful in teaching us where AI is given us limitations, where we thought it would break boundaries. And it’s also taught us where we can push different skill sets through our teams and discover different workflows within our existing teams and lean on their expertise to get the best that we can out of them. Thanks. So you are rebuilding around agentic workflows now. What does that actually look like day to day on the production side versus the everything through AI approach that you experimented with at first?

Yeah, so I think it’s been really good to start with the experimental approach in our use case. Because I think it’s made working with Amplience now really clear on building the agentic workflows and what we need specifically from those workflows and what can be most effective. Because I think in terms of where we were at the start, it’s well, what can we do? We don’t want to explore all avenues. We want to be very specific in what we’re targeting with our agentic workflows and the success that we can have with them. Okay, thank you. So Gus, you see that moment from the other side of the coin and you’ve watched a lot of retailers bolt AI into their stack. What actually separates the ones who see real value from AI versus the ones who just end up with a very expensive tech stack? Yeah, let me just start off by saying because this came up at least twice as I was paying close attention to different presentations. Garbage in garbage out starts with data and the

foundations. What I find really interesting is that a lot of executives and then different board members all of a sudden when they heard about AI or let’s inject AI all over the place, this majestic sophisticated solution will solve all of our problems. It will not period. So it starts with customers that pay close attention early on to the foundations starting off with data. Yeah, thank you. So we talked about where we all started. I wanted to get more deep into results. So coming back to you, Jayne. So you’ve got some real numbers behind this you shared with me. We talked about output going up, but the one number I actually wanted to dive into is the team one. So you told me that people left the copywriting team during this process, but you didn’t lose output. So how did that feel from the inside?

They didn’t leave because of the proof of concept, just to clarify. Yeah, we saw some really great numbers just from that kind of two month period. We saw a 70% increase in the amount of product descriptions we could generate a day, 150% increase in the amount of SEO articles we could generate per month. And then there was a nine hour saving for the team generating alt text. So I mean, that was the team working in a non integrated way on top of their day job. So those numbers were hugely exciting for us. It also showed that there were organic leavers in the team during that time, but we knew we could generate as much content as the business expected and more, which gave us confidence for the future. So whether it was due to annual leave or people leaving the business, even if we were in peak season, we knew that we could really push the content and for future, we always want to do more at scale with the same people. So we really felt we were onto something with this.

Amazing. And you have a similar story you told me, right? So you’re building agentic specifically to remove the limitations that come with scaling. What does that actually free people up to spend more time on for you? I think it’s a really exciting moment in freeing up the team’s time because these teams are experts in being creatives and they know the brand so well and they are our heroes of our brand because they know what the look and feel should be and how we can authentically connect with our consumers. So it frees them up to elevate across not just UK and European base, but we have a global brand that we can now connect with because their time is freed up to be able to connect globally across all the content that we produce rather than being stuck in the admin and the reporting. They’re able to use their expertise and their knowledge and be creative again and go back to their skill set that excites them day to day.

Yeah. So basically, there’s two slightly different angles of the same kind of value story, which is the do more without growing the team. So I wanted to hand it over to you guys. Is that the value story you hear most from retailers, or is it something different entirely? That is very true, but I’ll have to unpack my answer a little bit because there are different categories of AI-powered solutions and I usually think of three different categories. The first one is self-productivity or maybe team productivity related. This is essentially just leveraging AI just to make yourself and your team 10x, 100x more productive, more efficient. This is the first category. The second one is developing internal applications because one could argue that, well, if you have a software X, if you have the engineering horsepower to build it, why not build that much faster? This is the second category. And last but not least, the one that I would agree, you got to be a bit more cautious, is customer facing AI-powered solutions.

If that’s okay, I forgot to add two things about the previous question. So my bad. Your question was, what are the retailers that are actually seeing value with AI? And there is one aspect that I tell at least our customers or clients, which is holistic thinking. It’s not only just injecting an AI bot on the homepage of your online storefront, but it’s how can AI manifest in different ways? And then for the ones that have been at least playing with AI a little bit, you can enhance the carousels using AI. You can reinvent, if not the hero surface of all times, which is PDP. I think I forgot his name, the presenter from Forrester. He was just saying, oh, if you guys are updating your PDP every six to eight months, yes, that is completely flawed. But you’ve got to think holistically and then not only copy paste what you see working, but it’s just understanding deeply the behavior of your customer. And then how can AI play a role just to improve the overall customer experience?

Yeah, amazing. Thank you. So we’ve unpacked a little bit what the return so far has been. I wanted to move on to the next theme, which is what’s next. So Jayne, what’s on the roadmap for Primark next year? So we are now an innovation partner with Amplience, which means we have much more access to their product teams. We can help inform their roadmap. And we’re also working with their Premier Success team. So that is really going to help us roll out AI for the future. On top of that, and since we’ve done the proof of concept, we’ve since rolled out agentic workflows for alt text and product descriptions, which is working really well for us. We’ve also launched a shop the look capability on our product pages. And there’s no way we could have done that without AI. So that allows the customer to shop the head to toe look that the ecom model is wearing.

We actually partnered with StoryStream to implement this so that the AI agents, they grab all the products for us from a very large product catalog that we have at Primark. And that’s fairly recently launched, which we’re really excited about. And then we’re also looking at AI for backgrounds for product imagery, which we’ve seen a bit of already today. We ran a very, very small A/B test a few months ago, which was comparing a much more, I guess, simplistic product shot for, say,, homeware with an AI produced much more elevated kind of hotel aesthetic bedroom.

And what we saw across CRM and paid were higher engagement and ultimately higher results using the AI imagery. So we know there’s something in there for us. And that’s something that we’ll explore further this year. Wow. Very, very exciting things coming up for you. And I wanted to get your take on something, which is a very hot theme in retail right now, agentic checkout. So meaning buying directly through ChatGPT or Copilot. So before that works without breaking, what has to be true on the production side?

I think it’s making sure that we’re connecting with the rest of the business. And this is why your data in is really important to your data out. Like how do we in a traditional studio workflow, connect with the rest of the business in the way that’s impactful and how can we talk to customization and localization with the data that we can get. So it’s effectively using those agentic workflows as well to connect with the wider business, to harness the information that we have and use that to push things forward.

Yeah, yeah, that’s so true. So, Gus, over to you. So you very briefly talked about giving teams autonomy to experiment with AI. So I wanted to unpack that a little bit. And so you said something about the fastest wins coming from automating the repeatable commerce operator work behind the scenes. So if a retailer gave one team full autonomy on Monday, would that be the kind of ugly process you’d actually recommend they started with? And also follow up question around data and systems, because we talked about data quality. What would have to be true about their data before that actually works? Yeah, very, very good question. I think what I wanted to get across is that when I interact with different leaders and then they address or they call AI transformation, this gigantic multi-year program that will take forever.

And then my piece of recommendation to our leaders is that, so, hey, why don’t you allocate at least a handful of people, I don’t want to call resources, but like us at commercetools, we decided just to set people aside, myself included, this notion of labs, let people actually just use AI top to bottom, end to end, not only one fragmented piece of the workflow, otherwise, you won’t be able to actually validate the entire hypothesis. And this is very critical, right? And then besides, when you have such a distributed fragmented landscape, when one team is responsible for step A, one team is responsible for step B, it’s really hard to measure.

But I can tell you this as a living proof, if you can give autonomy to a small team end to end, magic will happen, period, but just don’t get in the way, let people fail, let them experiment, give them access to all the tools. I think in Europe, we tend to be very overly cautious at times. So let this small team just prove what can be done with AI. And then I can tell you, we can do loads, I can even demo some of the cool stuff that we’ve been cooking. I use this term, let the minions do the work for you, right? So you can just deploy as many minions, aka agents, as required just to get the job done.

I love that. So this is the one. And then we talked about data, right in the context of commerce, especially as we prepare in Europe for the real, agentic shopping or autonomous shopping. And then we just touched upon checkout. And it’s really critical to understand the primitives. So your product catalog, inventory, pricing, and then everything that revolves around these primitives. What do you got to do from a data standpoint, I could spend the whole day just talking about from data perspective, all the attributes that are needed, freshness being one of them, quality, which is very subjective, and whatnot, but just making sure that every single person has a deep understanding of quality, freshness, reliability, actually looks like for these different primitives, and then going to be in a good state.

Yeah, thank you so much. So we’ve covered a lot of ground. And I think this is all the time we had. But there’s one thing I want everyone to leave with. So if there’s one thing to take off this stage is start ugly, not exciting. The retailers ahead didn’t chase the most impressive AI use case, they fixed the worst they had. So don’t leave here asking what your flashiest AI project could be. Leave asking what your ugliest one is. So thank you very much.

Oh, I like that walk-in music for us, don’t you? So we’ve heard so much today that is so relevant to what we’re about to talk about. This talk is not a why-AI talk at all, it’s a talk about how to get these projects funded in a very practical way. So quick intro from yourself, Michael, if that’s all right. And tell us what’s live today in The Very Group when it comes to agentic? Yeah, so I’m Michael Ferris, I’m a Lead Product Manager. And my remit entails improving products that drive awareness and acquisition to the website. I was given the poisoned chalice of improving our product data 12 months ago, it’s something we’ve tried to do for quite a long time. But that process led us to selecting the Workforce capability from Amplience which we’ve used to improve our copy on our PDPs. Great, so that small, I guess, “ugly” project, like Isabella was just talking about kind of kicked off this next phase for you guys. And you already have the full platform anyway from Amplience,

including CMS and DAM and Dynamic Media. Making the case for content ops, they cost retailers, as Jeremy was saying earlier, about $33 million to $70 million a year. And product descriptions alone actually cost approximately $450 per SKU which can be nearly halved with AI, this is from The CORE Report which you guys will see around the venue today. But it’s not only cost savings, you know, the quality for GEO is also really important which makes product descriptions actually quite high value and it makes sense actually that you started there. So take me back to kind of before there was a project. What did the first conversations about this actually look like internally at Very?

Yeah, so this process started three years ago and it was before my time on the project. So we identified that we’d massively grown our assortment, almost exponentially grown it and we had over 150,000 new lines launching every single year. But we didn’t upscale our product copy team with that. As is often the case, it’s been a common theme through all of the presentations today. And it meant that around 60% of products were going live on the website with no copy at all which obviously wasn’t great from a customer perspective but it’s also not good from a search engine perspective either. So at that point in time we started with a tactical approach and from a non-fashion perspective we leveraged our supplier base and used supplier-provided copy. And then from a fashion and sports perspective we started using our first AI copy capability with quite early adopters in that space. And that drove our coverage from 40% to 100% and it was really effective as a first stage. But when I took over responsibility 12 months ago we wanted to pivot our focus away from coverage and to quality.

So obviously we’ve talked about how quickly the world’s evolving throughout today. And ultimately we knew that the quality of copy we were putting live was not unique enough for Google, or didn’t have sufficient detail for Google or LLM-based search. So therefore we wanted to really step change our product quality. The challenge we’ve got is we are launching 150,000 products from 2,000 brands, 900 suppliers and there’s 200 people who are involved in getting data into our systems. So as you can imagine by the time the data’s got into our PIM it’s not in great shape and that wasn’t really helping our copywriters to write great quality copy. So we had a time constraint, we’re asked to take the challenge of improving our copy on site by this time. Sorry, by now, essentially, this time last year. And that sort of points in the direction of an AI-based strategy.

Yeah, and times are changing so much recently too about less organic traffic to your website anyway. So that made it really necessary that that copy was not only there but good. So what had to happen internally, I know there was lots of lines, lots of things to do, time but what actually had to happen internally for stakeholders to kind of move this from idea into action, into procurement? I think we were quite lucky because our sort of senior stakeholders shot the website regularly as customers but also on behalf of our customers and the stakeholder I needed to sign it off, his pet peeve was our poor quality data. So it was a great starting point because we weren’t trying to convince them the need to do it. He was trying to convince us how quickly he wanted us to do it and that was almost not the biggest challenge we had. So we did our first iteration of the strategy this time last year and we thought we had quite a sensible sort of timeline to deliver it and he was like, no, it needs to be faster. So our ultimate aspiration is to fix the data that’s going into the business but we’re going to try and fund that by essentially improving our data quality,

driving more views and conversion and we’ll have that self-funding cycle to do and tackle that problem later on. Yeah, there’s nothing quite as powerful as an executive who like uses ChatGPT and can’t find you or like shops on the website and doesn’t like what’s happening there. So I guess that’s another little lesson about how can we make these things like a bit personal too? Yeah, you basically know what he’s doing next which is he’ll start asking about shorts or something like that, he’s on holiday this week Okay, so getting it into kind of project was maybe easy because you had executive sponsorship but what challenges did you face during the process? I think it’s just the sort of scale of our business. So we started with fashion and sports but obviously there’s so many nuances within fashion and sports and we now want to scale this out beyond fashion and sports to our whole estate. So essentially it’s trying to understand also the complexity that goes into every single part of our copywriting and make sure that our capability can capture that. Yeah, and I guess that varies by department too, right? Yeah, so I’m guessing you had some metrics around

deciding what was success, what success looked like. What was the reason or metrics that made you choose from, move from kind of your previous solution with the previous provider into using Amplience’s Workforce and kind of a future of that? Yeah, I think we should be honest and fair to our previous provider. They did a great job of meeting our requirement at the time. It’s just that obviously as we said all through the day, the requirements are evolving so quickly. We then needed to reevaluate. So there was three sort of key reasons we opted to move. So the first was sort of scope and speed. So we wanted to move beyond fashion and sports the first time which we hadn’t been able to do with our previous provider and we wanted to be able to do it at pace. So it took our previous provider six months to move from our first category on fashion to having all the fashion and sports covered. We got to that milestone within 36 hours of going live with Amplience. So it was a real sort of key driver of the change. And obviously when you’ve got another five divisions to roll out to you want to be able to keep that momentum going.

The second element was quality. So our previous provider did a great job of doing a basic coverage and some basic SEO capability. But we wanted to evolve that and get a real tone of voice and something that’s quite uniquely Very on the website. So that was another key driver. And I think the biggest lesson we learned from our initial implementation was that the market was evolving so quickly. So what we didn’t want was a black box, we wanted flexibility. And that was what the Amplience Workforce capabilities provided us with. Yeah. And I guess this is kind of the point where some AI content transformation projects die. When something takes too long, the quality at the end isn’t quite what you’d expected. So well done for not giving up You know, sometimes that enormous scope of like a massive AI project. I think Gus was just speaking about this too. It can collapse under its own weight. And it takes too long to show value.

So before the rollout there was like a test and proof phase. Can you tell us a bit about that and what were you trying to prove and to who? Yeah, I think ultimately obviously when you’re asking any business. And particularly business of our size for investment. There’s an expectation that there’s an ROI on that investment. And we know that in this sort of like the views and conversion space, they’re quite messy metrics. So it’s really hard to prove that A, it has increased or not. And B, what’s driven it. And obviously there’s many projects ongoing in our business at any point in time. That might be driving those metrics. And when the numbers go up, success has many fathers. And when the numbers go down, failure’s an orphan, isn’t it?

So we’re basically in a situation where we’re like we have to objectively prove. We’re improving our product quality. So we created like an objective scoring mechanism that we’ve used throughout the project. So we used it right at the start of the project to identify where our copy was. And like a baseline metric and who was doing copy well in the marketplace. And then we’ve used it throughout the project. So as we moved into testing, we used it as a way of identifying that we’ve seen the right outcomes from the capability. And then as we got into that go live phase and trying to get the stakeholders really engaged with it’s great to sit there and go. Our previous copy was scoring 18 out of 50. Our new copy is now hitting between 37 and 41 out of 50. And it makes it really, really tangible when you’re presenting it to some more senior stakeholders. Absolutely. It’ll be really interesting to see how that turns into more organic traffic to the website More AI referrals, things like that. Did you have any skeptics or any kind of hurdles?

And was there a turning point kind of moment that you could share with us? I’d like to sort of sit here and say that we had the whole company against us. And there’s one moment where we’re really charismatic and we turned the business around. But I think we’re quite lucky that essentially all the senior stakeholders are quite engaged from the start. I think what we have done a really good job of though is managing that momentum of the project through the project. Jen, who works with my team, is here today. And she says that the business is probably sick of the sight of me now. It might be like a way of hearing some sort of feedback. But essentially, we’ve been out in force constantly in the business. So any opportunity, we talk about how the project is progressing from its sort of infancy through to the point where that’s go live. And obviously, knowing any big business, there’s budgets are always flexing and priorities are shifting.

We’ve basically made sure that our project was always front of line to the business. Yeah, it’s good when they get sick of you. That’s when you know it’s finally starting to sink in. You might be coming to this later, so I apologize. But one thing that we spoke about when we talked was about how flexible the Workforce platform is. And how that almost caused a bit of like, “Ooh, I don’t know how we’re going to manage this”. Can you talk a little bit about that? Yes, I think when we set out, we had a very sort of tight RFP that we’d set out. And it was basically to create great quality fashion copy. And with a roadmap, to be able to do that on all the divisions. And Amplience answered the RFP really well.

And then we brought people in to sort of demo the capability. And I think the Amplience Workforce capability is probably the most polarizing during the demo phase. Because there were some people who were absolute specialists. And all they did was write copy, and their capability is fantastic in that regard. And then everything we said to Amplience, “We don’t have it yet, but we’ll have it there for launch”. And I think it almost felt too good to be true. But I think also, the benefit was the fact that you instantly saw that you could use it for beyond copy. So we weren’t just buying a copy capability, we were buying a platform we could use for automation All sorts of content. And I think it took a few people a bit of time to sort of see that long-term vision. But we got them around in the end. And I think now Nick’s team is very busy with, “Hey, can we do this? Can we do that?”. So that’s really good. Okay, so how did the proof of concept, the demos and the integration part kind of help with making a decision?

Yeah, I think the Amplience guys were really patient. So I think it was like an RFP process that had a set deadline to start off with. And just kept rolling. Because we wanted to make sure that every single stakeholder who had a say in the decision was brought into it. So we had sessions with our architects, we had sessions with our sort of category teams. We had sessions with people who would be using it operationally day to day. And we just wanted to make sure that absolutely everybody was comfortable with the decision we’re making. So that we had everyone aligned to start off with. Yeah And beyond kind of the initial use case of product descriptions. Were there any other use cases that added value to your decision? So at the point we sent it out, I said we were very much focused on product descriptions. But I think the moment we start to see the opportunity and see some of the demonstrations. There’s much more opportunity here. So. Mark’s already talked to the opportunity to sort of use it for image cropping and image zooming, et cetera. We know there’s multiple places where we’re writing copy across the website. Whether that’s copy on our PLPs or sort of buying guides. We know this opportunity really maximizes this.

And we’re seeing use cases crop up, we didn’t even know existed. So we’re now in a situation where there are too many use cases. And we’re having to sort of ruthlessly prioritize what we worked on next. Yeah. And how are you doing that prioritization? So I think we need to look at the sort of impact and effort assessments. So I think the massive part of what’s delivered the benefit here is that we said, okay, we know what it is we’re going to try and achieve. We know what the benefit is and we know the time scales we can deliver it in. And then if you can sort of get those more bite-sized bits of benefit. Then it helps us to advocate for something that’s more meaningful. Yeah, yeah. Okay, so you just had a launch of this. Yeah. Can you walk me through kind of what happened? Yeah, so we’re sort of like seven working days into this launch capability now. And we haven’t got a great track record of delivering change within our retail team in the business. I think we’ve got a team that’s like so used to Excel for most of the tasks that they do. That change hasn’t always landed well. This has probably been the most seamless launch we’ve ever had. So we’re now seven days in we’ve written over 4,000 lines of copy in the capability already. And it’s at a point where people are actually

asking us to turn it on more because at the start we limited the number of lines going into each day. So sort of maintain a bit of control. And I think that’s testament actually to sort of like real rigor on both sides before we went live. So from a testing perspective before we went live, we knew that everything was working. It was flowing through the tool properly. It was flowing into and out of the tool our side and land on the website. So we test all that in advance. But even before that is the sort of design phase that we got right. So previously before any AI was taking our copywriters about 15 minutes per line to write copy. I think we wanted to really dig into why that was taking so long. And obviously they’re making considerations about tone of voice about like green claims about any sort of banned words. We can’t use from a copyright perspective all those sort of considerations we had to bring into the mix. Before we even start thinking about making it sort of SEO and GEO rich. So what we wanted to do is make sure that all of that sort of quality and care that our copywriters had previously put into the creation of copy was captured by the tool. And our workflow that you sort of see on screen in the demo so far today

is now gigantic. It’s massive, it’s like over 80 different points where the tool is either being given instruction or making a judgment. And we’re looking up to like nine CMS tables where we’re sort of capturing data that the tool can and can’t use. And we’ve got that sort of person in the loop as well to make sure that the output is great. So we knew there weren’t any surprises because we sort of designed them out of the process. Yeah, and I think that flow is just for fashion, right? And it’s like 80 steps, something like that. And you’re doing that, like you work with the people before it even goes in to figure out actually what do the humans, what do we have to do to get this copy live. And it’s different per line And even for that as well, managing every single stakeholder. So obviously we know that our sort of buying and merchandising teams own the KPIs for the lines, they don’t own the copy. But at the moment something’s not performing, it’s one of the first things they look at. So sort of like making sure they’re bought in on the journey as well has really helped with the engagement and landing this change. That’s very exciting, very good. So 15 minutes per line before, what was your target for post launch and kind of what

are the metrics that you’ve seen? So as has been a common theme, we’ve got four copywriters in the business to sort of cover 150,000 new lines a year. When we moved to the first AI model, it was just fashion and sports only. But there’s now around 500 extra lines that we’re going to tip into that team when we move beyond fashion and sports. So our target was to be able to write copy in around two minutes per line. And that’s essentially checking that the output is great. From a customer perspective, we then sort of convert that into the SEO rich copy. And just making sure there’s no sort of funnies when you read through it. We’ve also included auto navigation. So previously our on-site taxonomy was all manually done by our buying teams. They had no images and only a tiny description at the point where they navigated the product. So as you can imagine, the navigation wasn’t completely complete and it wasn’t particularly rich. And that’s now also being done by the capabilities and checked by the team. So it’s around two minutes per line and it’s obviously allowing us to leverage much better data quality across the site. Oh, okay. So this is actually your internal systems as well that have benefited not just the customer

facing. That’s amazing. So as I mentioned earlier, like this kind of operational bloat is what you’re describing. You’re really working hard to close that gap for your team. You’ve talked about a lot of numbers today, but kind of was there one that mattered the most to you? Yeah, I think it’s quality. So obviously like right at the start of the journey and we sort of talked to coverage was the initial aim. And that was great because it wasn’t very good as a business our size. Sending products live with pretty much just material content and fabric composition or previously with no content at all. We know that we over-index on PPC and don’t get enough sort of organic traffic to the site. And if you sort of search for any sort of major products, we’re generally sort of lagging on page two, three, four of Google. Which just doesn’t get enough traffic. So driving quality was the number one focus now. And as I said, we’re sort of consistently getting to sort of between 35 and 41 out of 50 when we’re scoring copy. And those sort of decisions, which sort of things we’ve made are now sort of impacting our score. So like for

instance, we haven’t opted to include things like delivery message and returns messaging in the copy So some of those sort of like considerations we’ve actually made and now harming the score. But the tool is producing great content every time. Yeah. No offense to any of the brands that supply Very but it would be really cool if Very started to go above them in the natural results. Because of this project maybe not cool for them. They should talk to us too. But yeah, tell me about what’s next like you’re rolling out this product descriptions use case division by division after that. What’s the plan? Yes, we’ve got quite a really aggressive timeline now. So every six weeks we roll out a division. So we’ve got almost like a two week phase where we’re doing a research and get our requirements. Because there’s little nuances for every single new division we send live. Then two weeks to sort of like tweak and evolve the workflow and then sort of two weeks to test the output and send it live. And we’re currently in that sort of testing phase for toys now. We want to get that copy live as quickly as possible ahead of golden quarter. And then we’ll move through the rest of the division between now and Christmas. We’re in that process now starting to capture all those additional use cases that we’ve identified. So.

Mark is obviously quite keen to get his use case off the ground and obviously hopefully we can get rid of it. So there’s sort of internal policies to get that moving and then we’ll start to capture some of those more and all the use cases across the business. Yeah, amazing. So if you were to give advice to retailers in the room who are at more of the start of their journey. What’s the one thing you’d tell them to get right early? I think I’d probably echo the advice that has been shared in the sessions now. And it’s a lot of businesses want to be seen to be doing AI and they try and tackle the biggest problem they have. And ultimately it fails. I think if you take that biggest problem and turn it into bite size like iterative chunks that you can prove you can deliver. And then scale it’s a much safer way to deliver it.

Yeah, good advice. What’s your one-sentence takeaway for the room? So I’d say don’t buy a product copy capability, buy an orchestration capability, because it’ll have multiple use cases. Yeah, we’re looking forward to what you wind up finding and how you kind of drive forward the Workforce platform from all of the ideas that you have. So for me, I think, you know standing still on this isn’t without cost. And the cost of doing content operations the old way is going to get higher and higher as time goes on. And what you’ve just heard here from Michael and also from many of the other retailers on the stage today is like is what moving looks like. And it is feasible and it is possible. So jump in. Thank you. Thank you

So please welcome to the stage, Rory Dennis, VP of Global Partnerships at Amplience with Puneet Goenka, UKI Retail Sales Leader at AWS. Great, thank you, Amy. I just in the front row, so that’s great, I guess. Thank you for coming. So everyone, wanted to kind of draw some kind of analogy here to introduce Puneet. So we know that Abbey Road Studios is famous for The Beatles’ Abbey Road, obviously. What’s probably less well known is that there was a gentleman by the name of George Martin, and George Martin produced that album for The Beatles. In fact, he produced quite a few. He was commonly known as the fifth Beatle. So as the Beatles were creating their music, he was behind the scenes orchestrating everything for them. And that’s where Amplience kind of joins with AWS. Ever since we moved to the cloud in 2012, 13 maybe, AWS has been our infrastructure partner.

And we’ve had a great relationship with them. And what’s happening now is that we’re furthering or deepening our relationship with AWS as we use AWS Bedrock as the producer that sits underneath our platform. So we brought Puneet here today. Puneet is an expert in AWS, has been a great partner to us. So we brought him on, we thought we’d drill him on a few little bits and see what he’s got in store for us into the future. So Puneet, would you like to say a few words about your background and tell everyone why you’re here? Thanks, Rory, and thanks Amplience team for having me. It’s been a great partnership, and I think I can claim today that I recorded at Abbey Road Studios now.

By the way of introduction, I help retailers transform their business and their operations, both retailers and brands. And at AWS, I’m the head of sales for retail and CPG at AWS. That’s me, 26 years in the industry. Before AWS, I was with Apple and a few other Silicon Valley companies, that’s me. Fantastic, well, thanks for joining us at Abbey Road. All right, to get into it a little bit, Chuck spoke to us earlier about adoption and what he was really talking about was, what’s it like from February ‘25 to ‘26 in terms of GPT usage? And we see it’s gone from, was it 12% to 34% queries beginning on GPT in February ‘26? So my kind of question to you on that is, if that’s true and the pace is as Chuck claims it to be, what is preventing retailers and e-commerce brands from really getting into, from moving that towards 60, 70%? Is there some blocker there or what’s kind of happening? I think that’s an interesting one. I think if you look back 12

months back, the conversation in the marketing or the content teams were about who can produce the content fastest, right? What’s the speed? How many people, how much efficiency and so on. But if you see in last 12 months, the things have fundamentally changed. You’re not creating content just for human eyes. You’re creating content for machines and more. If you look at the traffic, the search traffic is up by 200% from the agentic search or the things coming from agentic chatbots are up by 400%, right? What has historically happened is retailers and brands had a great website, great content, sitting in a beautiful system, but they were designed for human eyes. They were not designed for machine to read it, understand it, compare, trust it and do a transaction. And That’s the fundamental shift.

So more and more, retailers need to look at it from designing for the machine rather than just pure human. So we’re saying we’re just in a period where that’s about to evolve and we’re gonna move to a place where, do you think, in 12 months’ time we’re at 60, 70%? I think at this stage, it’s a very difficult to put, draw a line to be honest. I think the LLM space is evolving very fast. We at AWS think that in the fullness of time, LLMs will be a commodity. So it’s not about one single LLM, whether you are looking at an OpenAI or Anthropic or... That’s a good point actually, because I wanted to ask about that.

Again, I think if we went back 12 months, everyone was kind of zoning in on one AI that they were using and often it was OpenAI. And then maybe like halfway through this year, I don’t use that anymore. I’m all in on Claude now. Do you think that that being wedded to a single model is in the past now and should we be looking at multi-model? And is that essentially what AWS is doing for us with Bedrock? Is it combining all these models for us? Yes, I think where we are seeing the AI space evolving is historically, if you look at 12 to 24 months back, people started more with ChatGPT as a starting point.

But if you look at it now, the model that you use for creating your image is completely different than the model you use for creating your product copy versus a model that you would use for your compliance check or content management kind of a model. So it’s more about what’s the right model for what right job and the speed of change is very fast in that space. So what you want really is not to be married to a single model, but you want to build your logic layer in a way where you can switch models without having to modify your application code. And that goes back to what John was talking about in the morning.

The users and the content creators should not be worried about what model to use. It’s for the platform like Amplience to give them that capability and for the users to say this is what they want to do. Yeah, fair, I get it. So because what John was talking about earlier was that you shouldn’t have to worry about that because that’s abstracted away and then they choose the best model for the job under the hood. So you’re not specifically having to go run this service through that model because it’s better. It just does it for you automatically. I think that our mental model is over a period of time, the users should not have to worry about what model they’re using. It’s for something like an Amplience or a Bedrock to define based on the requirement or the prompt that the users are giving. And that’s what we are doing in Bedrock. So if you look at AWS Bedrock, which, like Amplience, and many other customers use, is we are bringing in all the models, both the paid models and the open source model.

There are roughly around 130 odd models there, which the engine can decide based on what the prompt is and get the right output. Okay. And then I suppose, sticking with the Bedrock theme, then what you’re abstracting away from me as Bedrock are things like governance, guardrails, model selection? Yeah, I think as I said earlier, the content needs to be defined or designed for a machine as well as human. And it’s really changing from being a human creative to more of a engineering function in some ways.

And what we classify as there are three essential items that any enterprise needs to look at. First one being governance, around what kind of governance you have. Like is the LLM allowed to look at something? Is it doing the right thing? Is it as per your compliance policy? Second is the agents or the machine should be able to read the content and trust it. What an agent doesn’t want is to go to a five different system to look at the pricing and availability, right? So governance, compliance, and then the guardrails around that. So I think those are the three primary areas. Okay, then I have a question. You might not be allowed to say any brand names here, so I understand. But I don’t know if anyone’s been watching the news the last two days, but there’s been a bit of a spat in the market from between Amazon and Meta. Muse has launched, it has been blocked from crawling Amazon shopping.

And then of course Shopify have come out and said, well, we won’t block Meta from Amazon shopping. And of course, Elon Musk has weighed in, hasn’t he? Because that’s what he does. And he’s saying that no single platform will ever actually win, it will never outperform all the other platforms. So is Amazon being protective here? Or is there some other reason that you think they’re blocking out Meta? I don’t think we’ve been protective here. And given the scale Amazon operates, it’s really difficult to be protective in that space. What we are essentially saying, and this applies to most of the retailers and brand, is that there needs to be a guardrail and there needs to be a governance. Now what we saw some of the LLMs may not comply to the security policy or the user experience that we want to give.

So in those scenario, we want to have a capability where we can stop certain agents accessing our site and getting the customer information. All we are asking for is here’s our guardrail, here’s our governance layer. If you want to read and scrape the data from my website, you need to comply to a certain security protocols. And I mean, you took some names, but this goes beyond any particular individual LLM or any particular technology. That’s the general rule that we follow. There has to be a guardrail. Okay, probably something brands could learn a lot from as well then that they could put similar guardrails in place so that content could be protected or? Yeah, I think the guardrail and the governance is the most important part in the space.

There’s been a lot of conversation about, you know, whether you give your data to the LLMs and whether you can do the shopping journey through the LLMs. And our position is that the brands and the retailers understand their customer best and they need to define the customer experience, the customer journey they want to provide. And in that sense, it’s not for the LLM to define it’s for the retailer to control that data and decide what experience they want to give. And I think just to add to that, and again, goes back to what Jeremy was talking about in the morning, that’s why we work with the companies like yourself, because that gives a capability to the retailer, to the brands to build that kind of a guardrail. Okay, fantastic. We’ll watch on with interest anyway. We’ll still not mention any names. All right, two final bits, I suppose, before we wrap up. One, I was interested, I’m sure our audience is, can you think, I know you can, because we’ve prepared this. So, can you highlight some brands for us that

you think are doing really well, we’re producing agentic experiences right now? I think we might have some visual assistance. I think we had a slide for that, I don’t know if that’s the right one, but I’ll take an example where we work with a beer brand, Corona, and they operate across 100 plus countries and for every country, they create a content which is specific to those geos. Yeah, this is a beer brand, Corona? Yeah, a beer brand, Corona. And what we did was we brought in an AI layer where they create a base image for the Corona on a beach, bottle of Corona beer on a beach and then when they market it across multiple geographies, let’s say Spanish beach or a Mexican beach, they can basically change the background and the layer around that. And they can just generate it from AI. And I mean, there are some stats around how well it worked out for them. Essentially what it meant was that it was not about generating it from AI, what it was about was building content and the content lifecycle in a governed way where the guardrails and the governance is in place, you can do it at a speed and roll out to multi geographies

without having to worry about your compliance and the copyrights and so on and so forth. Okay, gotcha. And that was Corona, I thought we had a visual, but that’s fine if we don’t. And then you had a fashion brand as well that you like that you see an excellent piece recently. I think the fashion brand is a very interesting one. We are seeing a lot of work in the fashion space right now on the AI ranging from, I think somebody from Very was talking about product description, image creation and so on and so forth. We’ve been working with many and we’re also working with a lot of agencies. And I think the example we had here was with Dentsu, which is an agency and if you think of it as an agency, their traditional model is to have humans to create those contents and follow workflow. Whereas what they did was they brought in an AI model, a life cycle on AWS Bedrock and automated the entire process of content creation with obviously human in the loop.

But again, building the governance layer, the logic layer and the guardrail within the platform and the productivity gains and the turnaround time for the content was a significant change from how they operate in a normal old world. And I think, I don’t know if that’s the slide or the next one. Yeah. And we’ll take your word for it. I won’t go into the numbers of that. We’ll go and check it out afterwards. Okay, final thing. If you had one piece of advice to give our friends here before we get off the stage on their agentic journey, where would we be going?

I think the race for the future is not about the retailer or the brand which generates the maximum content. It’s the one whose content can be read and acted upon by machines or as you call it agents. So future is not about just pure productivity gains. It’s also about how are you ready for the hybrid world? Don’t get married to one model. Is that right? I think that’s obvious. I’ll take that. Yeah, that’s obvious. I think if you’re married to a single model, you’re doing a bit of a disservice to your own enterprise. Sure, okay. I think we’re gonna take a few questions if there are any.

Just to open that to the floor. Anyone wanna ask AWS anything? Not related to meta, obviously. I think we were super clear. Yeah, good. I think we can get out of here then. Well, thanks everyone. Thanks everyone.