Latest / The Interline Podcast / How Much Will Fashion Let AI Decide?
Transcript
- speaker-0: Hey, welcome back to the Interline Podcast. Right now, fashion has a laser focus on product outcomes, and for good reason. The headline research tells us that industry growth is slowing down, which means that brands are competing with one another for relative rather than absolute advantage. If you listen to our recent Mailroom episode, you heard Hannah Crump from Business of Fashion Insights talk about the fact that in a low-growth environment like the one we're living in, You can grow your slice of the pie by subtracting it from somebody else's, but the overall dish isn't getting any bigger. And that makes optimizing per product profitability and reducing per product and collection level risk essential. The data also shows that consumers are way less aligned with brands than they have been in a while, and that they tend to make selections based on specific products instead. That makes subjective attributes like style and fit, again, on an individual product or small collection basis, matter a lot. And it makes more calculated objective criteria like pricing, value, market positioning, and market counterpositioning matter arguably even more than they did when fashion companies are making the choices that they're making to bring products to market. None of these outcomes just happens, right? Each of them is the result of a set of nested decisions that cascade down from this is what we want to make, given the market conditions that have been handed to us, to the granular choices that then influence how that thing gets made and how effectively it meets its objectives and delivers its outcomes. Each of those decisions is also, especially right now, a target for potential automation. That's particularly true if you believe that AI can make these kind of mission critical choices better. Or that it can at least make them based on a more complete reading of the circumstances than a human being can. Because of that, and depending on who you're talking to, there's a very blurry line right now between which of the many, many choices that conspire to turn a slot plan, a brief, or an idea into a finished product should remain the preserve of people, and which ones could be, should be, might be augmented or fully automated by AI. Sitting back and listening to this, your instinctive reaction is probably, well, you let a human set the creative direction and you let them determine the success criteria, because that's what really makes a brand. And then sure, you let a series of AI agents make the optimum choices in the sub tasks that sit underneath all of that. But is that actually the best approach? Is that actually the right read? Does that framing undermine the Creative decision making that goes into those more granular stages that we're in a rush to automate? Does it over-index on AI output and capabilities that professionals tell us they don't actually really trust? Does it overestimate the complexity and the essentially human nature of setting a collection direction or building an assortment plan? Your instinctive reaction is probably also covered by the systems you use and the processes you operate in. AI is promising a lot in terms of automation, but The inertia it's working against is massively strong too. It's not just a case that companies are saying, Well, this is the way we've already done things, so why change? It's a different kind of weight because it comes from a much more uncertain place. It comes from a place where to get a real answer, you have to ask yourself some extremely profound, in the most literal sense, questions about not just why you do things the way you do, but whether you need to keep doing them at all. Or whether we're in a completely different era. A lot of this discussion has already come up in our AI Report 2026, which I'm sure almost everyone listening to this show has read. If you haven't, I'd suggest grabbing a copy because there's a lot of very timely content and context and data in there. As part of that AI report, though, we also interviewed a bunch of tech executives. And one of the those interviews has kept coming back to the front of my mind whenever I've been asked to talk about AI automation. And how far it might go and what decisions might get made for us. And it's the one we did with Rupert Sheessel, who is the chief strategy and AI officer at Bamboo Rose. I've had a few short chats with Rupert on and off the record before over the last couple of years. And he's got a refreshingly objective viewpoint on how far AI is likely to actually change fashion. And as his title suggests, he's also of the mindset that AI and strategy are. basically inseparable today, that every strategic plan is an AI plan in and of itself. Rupert's company, Bamboo Rose, also uses the term decision intelligence to refer to the way that AI is applied across the extended product lifecycle and the technology ecosystem for making all of those choices that turn a product concept into a finished thing. So when it comes to how far AI should influence decision making, and when it comes to what that means for product outcomes, I think Rupert is as well qualified as anyone to have a pretty far reaching and pretty impactful conversation. So let's get into it. Okay. Rupert Schusel, welcome to the Interline Podcast. speaker-1: Thanks for having me. speaker-0: No at all. Looking forward to this one. ⁓ I know you and I have spoken a couple of times in just sort of different formats, but this is the first sit-down interview that we've ever done. So I've been looking forward to this one because I know you have you have some good perspectives on the role of AI in business strategy, the role of AI across the fashion industry. Good. Good. Well that's ⁓ I'm looking forward to tapping into them. and to that end, so you know, we start speaker-1: Which speaker-0: These shows every time we're the look at the guests day-to-day, and then we ask them to define something. So let's start with your day-to-day in that capacity. So your title, Chief Strategy and AI Officer, it transcends a couple of areas that I think are probably near or at the top of our listeners' agendas. So strategy and AI. In fact, I'm going to hazard a guess that for most of our audience, the strategy is primarily AI or AI-centric at the moment. Tell me how those two things fit together for you, ⁓ what it means to do strategy and to do AI, and how much of that you think you might have in common with the people listening to this who are maybe juggling the same priorities. speaker-1: Yeah, I think the title exists because right now in twenty twenty six you can separate the two. Right. Five years ago, a strategy job was about markets and positioning and MA. And today almost every strategy question ⁓ comes down to what AI makes makes possible and what it does not, right? So I think ⁓ generally the the sass software vendors like Bamborough's where I'm working. Today are just having their Kodak moment. If you remember in the 19s, in the 90s, ⁓ Kodak got into heavy trouble when the digital photography was arriving on the market. And they actually saw digital coming, right? They built the first photo sensor and so on. ⁓ but they what they they couldn't do was kill the business that they were that was paying their bills ⁓ for them. So for a software vendor today, it's very, very similar. So AI is not one project ⁓ that you run inside the strategy. Where you can ⁓ continue as before. ⁓ we have to do it properly and actually rebuild our whole business around AI. And this is this is strategy. So doing anything else would be just adding features to a model that that's quietly ⁓ running out, I guess. And for example, a lot of companies are making the mistake of automating their existing processes. And then they say, ⁓ we have ⁓ adopted AI, we are working so much faster now. And that's the the deadly trap I think that Kodak went into ⁓ 30 years ago, because it's not about accelerating the existing things, it's about completely breaking ⁓ those processes and redesigning them entirely. And that's true for us ⁓ too at Bambrose. My job only makes sense if it's about changing what the the company is and not adding AI to how we already do it or what we already sell. speaker-0: I think that's good framing. ⁓ and I think that's gonna come out as we go through the the rest of this conversation. For the thing I want you to define, I've picked something that it it sounds absolutely basic, right? ⁓ The thing I would ask you to define is what constitutes a decision. Now, everyone's got an instinctive understanding of what that means. We're all used to choosing one thing over another or setting an objective and then working towards it. but decisions aren't one big unit. ⁓ and and particularly in in fashion, like pr like product decisions, market decisions, ⁓ category decisions and so on, ⁓ you can break those down into pretty granular pieces if you want. And each of those pieces ha maybe has a different person or a different team responsible for each of them. if you think about a hypothetical style, you can decide you're gonna make something, but between that decision and the eventual outcome, which is a product on a shelf, there's a massive spectrum of other interrelated decisions that get made. ⁓ from choosing materials, manufacturing partners, setting margin targets, approving colours, and so on. When you think about the sheer complexity of what a decision can be and all the sub decisions that exist underneath it, how does that frame the way you approach the idea of decision intelligence, which I think is the the framing you use for ⁓ for AI? You know, which parts do we automate the big decision and then give people the smaller decisions underneath it. Do we do it the other way around? Is the answer the same for everyone? Walk me through the impact of AI on the decision making process. speaker-1: Yeah. In in the industry we are working for so fashion and apparel, a decis a decision is ⁓ almost never one single act, right? Between ⁓ we are making this style and a product on the shelf, there are hundreds and hundreds of small decisions going on. And most of them aren't really decided, they are sort of inherited, ⁓ because many of them are copied from last season, for example, or from a template or from what whatever the supplier proposed. So they are already in our customers' history. ⁓ the retailer knows them already. So most of what ⁓ looks like deciding in our industry is actually just looking things up in past data and make sense of it. So people are ⁓ just rebuilding ⁓ context in some some way. So what did we do last time? What does the supplier normally charge? Is this ⁓ this certificate still valid? ⁓ and I think that work is 90% of the effort, ⁓ effort for our users. So for me, the line isn't ⁓ really big decisions versus small ones. ⁓ we think what we have to do is automate the data retrieval as much as we can in the existing history. Automate putting the context together so that AI can understand this context and work with it, and automate every decision that is really just basic ⁓ rules. So concretely this means that a lot of ⁓ the small decisions shouldn't reach a person at all. ⁓ if the rule is already known and the action can be undone, the system should just do ⁓ do it and log it. And asking a human to validate something that was never really decision is is I think just a waste of time. So I guess we think we think in two levels we have decision automation for doing things automatically when it's possible, and then decision support where we prepare where AI prepares a decis decision ⁓ to be validated for for a human. And the mistake that many companies are doing today is that ⁓ they are putting too much effort on decision support because it feels safe when you bring a human in, but at the at the end you end up with an expensive AI system plus ⁓ the time ⁓ spent by the human and you don't have any savings ⁓ savings at all at yeah. speaker-0: I think that's an interesting perspective. And think as well what you described is something that I don't think I've ever quite crystallized in my head until just this minute, which is that a lot of what we think of as choices aren't really choices when you weigh up all of the variables that go into them. So if for example we you know we go back to our hypothetical style ⁓ and we think about choosing a material or choosing a manufacturing partner, maybe. it's probably the case that there's maybe one or two manufacturing partners who are able to they have the specialisms and the capabilities to do ⁓ the particular product that we're looking to have made. They could do it in the timeline and at the cost and ⁓ they have the right sustainability certifications and credentials and things. ⁓ there's a whole lot there that's just sifting through information and then making what's basically a a binary choice if it's a choice at all. And I think that's what you're driving out when you talk about some of these things that you're you're trying you're trying to give people support to make choices that really only have one or two outcomes. And those are places where automation is the more natural fit. speaker-1: Yeah, exactly. Really our main work is building this meaning and the context so that people can take decisions, but also AI agents, for example, can make decisions based on that, ⁓ on that context and and how far you can go depends ⁓ on two things. One, how how governed your data is, and then whether the decision can still be undone. So a company with clean supplier data, for example, can go much faster. ⁓ of course than ⁓ a company whose data still sits in spreadsheets. So instead of asking yourself what should we automate, in most cases ⁓ you have to ask yourself first, what do we actually know and and how is it stored? speaker-0: Fine. Now you just talked about, you know, the the difference between decision support, which feels safe, and decision automation, which feels risky. If I look at the findings for the survey portion of the AI report twenty twenty six, which we published ⁓ a month or so before this conversation, which there's a short interview with you in, ⁓ one of the most revealing things I think we found was that although like nine in ten fashion professionals use AI every day at home and at work, only about a quarter of them currently trust the output of AI enough that they're comfortable basing an important decision on top of it. Now, in that short interview in the AI report, you talk about trust coming from transparency. So people don't want to be served recommendations or they don't want to see prescriptive action being taken if they can't understand why. ⁓ But I think most AI kind of fits into that second box. You know, it's inscrutable, it's probabilistic, it gives you authoritative outputs without necessarily giving you visibility into its working. What do you think happens next here? Because I think there's there's a couple of ways you can build trust in in the output of AI. One is you do it extremely slowly. You you slow walk it into processes and you say, look, it hasn't made any mistakes. It's retrieved the information accurately. You do that progressively over time. And then the other one is where you build it quickly and you just go, This is the correct answer and I can show you it's the correct answer and we can now instill confidence in this. Wha how do you think we build that trust that's currently lacking? speaker-1: Mm-hmm. Yeah, I think building trust with AI is is very basic. So AI doesn't take responsibil responsibility for anything. If you sign off a compliance file, for example, you own it and you'll get sued ⁓ if it's wrong, right? And you can't own something, you can't explain. So on the question of explainability, I think it's actually both showing a quality output and being able to explain how it was done. So these both are useful because ⁓ ⁓ being transparent ⁓ gets people to try it, but a good track record gets ⁓ them to keep using it ⁓ in time. So you need both before anyone lets it act for for them. It's like when you go to your doctor. If you just pre ⁓ if he just prescribes medication and gets you cured, you're happy because you're cured, but you'd have preferred to understand why you're sick, right? If you get a very good explanation for what you have, but you don't get cured. You'll not go back to the same doctor again either, right? So ⁓ I think we have to bit a ⁓ be a bit more careful so about what transparency ⁓ means inside the company. It doesn't just mean explaining the model, because as you said, these the these models that we are all using today, these LLMs, these large language models, are probabilistics. So they love producing accuracy and confidence scores and That's how because that's the that's how they are designed, right? And and no human wants a probability. It's probabilities are difficult to read for for us humans. So meaning comes from showing where the answer came from. So here's the answer, here are the records behind it. Click and check them. ⁓ so providing a chain you can check is much more convincing than just a confidence score. A score asks you actually to trust it even more because you don't know how it was generated. ⁓ versus a source requires less trust. speaker-0: I think yeah, I think that's right. I think I agree with that. one other finding from the AI report I wanted to talk to you about was the way that the industry seems to perceive maturity. And I think maturity begets trust here as well. So now I know you see decision intelligence as something that sits across the entire product journey. in a way that contrasts with what the industry is telling us, which is that it perceives AI to be very mature and very capable at the beginning of the life cycle. So in Concept development, market trend and competitive analysis and so on, ⁓ analytical work. And then at the end in marketing and content creation, i.e., video and image generation. People seem to perceive it to be the least mature and the least ready and the least trustworthy, the closer it gets to the real product. So technical design, sourcing, production, and so on. ⁓ I think some of it is some of that is just down to the fact that we see AI being good at the extremes in other industries. So in our personal lives, it's very easy to say, okay, AI is really good at taking in a bunch of unstructured, non-normalized data and giving me insights, or it's really good at generating images that I think look nice or they look close to photography or whatever. do you see that same sort of weighting of the expectation of AI capabilities when you engage with fashion brands? Or if not, give me an example of where you think AI is. creating real value in that middle piece in between the early and the end stage of the product journey. speaker-1: Yeah, honestly I don't see this exactly the same way. ⁓ I think I understand why people ⁓ answer like that. So n I think what what these servers measure is visibility. So it's not necessarily ⁓ capability because in design and marketing, for example, you can very easily see the output, right? A generated image, a product description, a campaign and so on. ⁓ if it's wrong, it's it's very cheap. You just run it again. In sourcing costing in compliance, all these ⁓ different like this whole value chain that we're managing within Bamborough's. If it's wrong, it costs real money. So people are less ready to take risks there. ⁓ and there is also ⁓ the the reason that a a designer or a marketer gives the AI material it can already read. So a brief, an image, a piece of text and so on. It's it's coming from outside versus a sourcing manager or a compliance officer. ⁓ they have to give it the system's history. And so until now the the data that we have was organized for a screen to display it in software, not for a machine to be read and and to be understood. ⁓ a column name in for example in the database often doesn't mean anything on its own. So an AI agent cannot cannot use it. So the model is not weaker when you get to sourcing or costing. or AI is not less relevant. It's just not connected to to to the data in the in the right way so that's ⁓ able to use it. So nobody has translated the business ⁓ information into something a b a machine can actually read and and simply because this translation is a lot of work and and I know what we ⁓ what I'm speaking about because we have done this work with with with with Bamborough's. speaker-0: So our next question is actually a bit of a two part one, and I'll split it off so because you've actually hinted at it there. So we talked about trust a few times here. We talked about model capabilities being essentially universal, ⁓ and the other thing, other elements around it being what determines success or so on. The AI report survey shows what I've ended up calling like a non virtuous loop, which is where you've just hinted at it, where fashion professionals don't think AI will give them reliable answers. So they Don't connect it to their technology ecosystems because they don't trust it. So what if you don't trust something, why would you integrate it with your ERP or your PLM? But your ERP, your PLM, spy chain management solution, so on, those are the sources of truth that would allow ⁓ any AI application to deliver more reliable answers and so on. How do you think we get out of that sort of spiral where people are saying I don't trust this, so I'm not gonna make it part of my technology landscape. But if I did make it part of my technology landscape, I would trust it more. speaker-1: ⁓ yeah, I I think that's that's absolutely right, and that's what we are betting on. ⁓ too. We have we have worked for over two years, as I said, to expose Bambrose ⁓ through ⁓ MCP. ⁓ so ⁓ that customers own agents and can also query our database. And we were quite early on that because we don't believe the interface of the future are the software interfaces that people are using today. So some vendors will probably fight that. ⁓ But I think ⁓ I think this is a a losing position. And we I would even go a bit further ⁓ than you on the uncertainty because we don't know if it's ChatGPT or Cloud ⁓ that will win. We don't even know if MCP is still the standard in two years. Because new protocols for agents talking to each other, appearing. ⁓ maybe the big platforms will build their own ones. So we'll we'll see. Anyone who tells you today how this will end up is 100% guessing. So what we have done is build things that can be swapped, ⁓ that are model agnostic, protocol agnostic, ⁓ because we assume that whatever sits at the other end, outside of of what we are doing will change ⁓ more than once in the upcoming. months and we're already seeing that it's changing all the time. and so also what we think is that the interface itself is becoming a commodity. So customers will want and are already asking for that. They want to generate their own UI on top of their own data in an afternoon or so. and some some of them have already started doing this with with smaller data sets. So if the screen is not where the value is anymore, then defending the screen is is actually defending defending the the the wrong things. speaker-0: I agree with you on that. I I'm I'm also on record as saying I think talking to software ⁓ is is is where things are at in the future and custom custom interfaces is part of that. it does beg the question though of how the user base for technology changes. Because I think there's there's two s there's two ways to slice this. One is to say, okay, bamboo rose or any any similar product historically sold on license sees licenses, seats, and so on to a well-defined and well-scoped community of potential users who will interact with it and input data into it, extract data out of it and take actions on that basis. When you think about exposing it via MCP, ⁓ or whatever, the w whatever becomes the codified standard in the future, ⁓ you're thinking about exposing it to a completely different interface paradigm, as you just said, that is basically ubiquitous. If the assumption is that everybody is going to have their floor, chat GPT, whatever it might be, of ⁓ interface and that's that how gonna be how we interact with things, it changes who you sell to and it changes what the offer is to them as well. that's fascinating to me, ⁓ because it's it's a fundamentally different approach to how we position ourselves software. speaker-1: Yes, ⁓ yeah, absolutely. That's what I I said at the beginning of this discussion is that our industry and the whole way we see software is is is totally changing. So that's ⁓ that's why I think we should actually focus on the things that we are sure about. ⁓ because so many things are changing that we have to to have to find out what's what's actually not not changing. So in our industry, one thing is very easy, very easy to find out. People still have to get dressed, right? ⁓ because closing is one of the things you can't digitize. So material has to be transformed, ⁓ products still have to be made and and and shipped. ⁓ and all the data around this is knowledge. So how a product is built, ⁓ costed, sourced, shipped and and and sold and so on, that knowledge ⁓ exists in a database and it belongs to the brand. So nobody else has it. ⁓ And there's a lot of value in this and and AI cannot cannot change it. And that's what we are ⁓ betting on. So the the model, the protocol MCP or whatever, the screens, ⁓ the the user interfaces, ⁓ this is changing and ⁓ and we think this should be rented, ⁓ should be ⁓ designed at replaceable because it will get ⁓ cheaper every quarter, ⁓ even if the AI vendors are l is still losing a lot of money, but but the tendency is probably that technology gets cheaper in time. ⁓ it's becoming a com commodity versus the structured knowledge of how your business ⁓ actually makes things and and sells them. That's the asset and nobody can sell you that you cannot rent it from from some someone. So we as a vendor, we have to we think that the the like the the the new software vendor that we're becoming in this new economy, ⁓ we have to create meaning ⁓ on top of the data. We have to create context before all. So ⁓ w which means that a database where you have a field like ⁓ ITM underscore C SD underscore forty two doesn't tell anything, right? The the words the business uses like is is like costed bomb or first cost or active suppliers, the these are not in the database as of today. They're only in the people's heads and the and the people that use the database. So s we have to map them, maintain them and keep them up to date. That's an a very important thing. We also have to bring in ⁓ governance, governance and security so that an AI agent ⁓ can connect to these ⁓ data. ⁓ so once it's understands it has to connect securely and ⁓ we have to make sure that the right agents, for example, an agent that represents a merchandiser ⁓ or a supplier, so that these different agents are able to do the right things and and and and nothing else. And we have to bring in the business rules. that are already living in the software, which is a very important part of software. It's not something that you can invent because business rules are tied, closely tied to the way a business is working, to its culture, to its value and so on. So in the inter enterprise systems, these rules are integrated and they must be transferred so agent to to the agents. So with our metadata layer, ⁓ we are building ⁓ all this in order make these the the the data and the whole business model available to agent and that's what we actually call AI native ⁓ within within ban Bambrose is is building these layers. So we don't want to compete on the models ⁓ or the or the shiny stuff, but we think we have to compete on the way we organize ⁓ our software and the data, ⁓ the context of it in a way that it's ⁓ usable for AI agents. speaker-0: So we don't want to compete on the commodity models and we don't want to compete on the commodity interface, which leaves logic, rules, structure, governance, and everything that you've just described. To poke at that a tiny bit. Do you see that as a model that is safe in an era where anthropic or open AI could suddenly go, okay, the next vertical in our sites after financial services and legal and coding is retail or is fashion. And we want to start to bake some of that logic and some of those rules into ⁓ a verticalized deployment of Claude or a verticalized deployment ⁓ with pre exist pre established connectors and things for ⁓ of Chat GPT. I'm I'm keen to see like what you think is the defensible position of a software vendor with deep domain expertise if one of those two companies decides that this is the next industry they want to go after. speaker-1: Yeah, I mean ⁓ nothing is safe today, right? In twenty twenty six. Nothing nothing is safe. And ⁓ because for one single reason is that we are going through a very deep shift, which is that big part of the intellectual work that we were doing, ⁓ that human it was doing through decades, ⁓ is now externalized, partly externalized, and a big part of it is externalized ⁓ to models and it's done by models. So actually what human labor was before is becoming capital because it's now done by by by by chips vendors, by machines, by ⁓ it uses energy and so on. So it's this human work or human intellectual capacity gets owned by these ⁓ big ⁓ AI vendors like OpenAI Tropic and so on, or maybe ⁓ by the vendors of the of the hardware behind it or the energy behind it because they are ⁓ also very very strong and very concentrated. So so actually nothing nothing is safe. Nothing is safe and but we have to prepare ⁓ again for what's what's what's what's stable and what's still there. And these are the what's still there is ⁓ all the the hard processes and the goods flowing and the logistics and the data and the knowledge ⁓ behind this because this is This is private knowledge that is owned by our customers and it was accumulated through through years of of of operations, right? And that has a specific value. So what we're seeing at least in the in the upcoming years is that there is general knowledge developing with with this general capable models that replaces part of the intellectual work that that sits ⁓ in the heads of our customers but also in partly in in the rules of our software because in instead of hard coding rules you can now trust an AI to to reason upon them, ⁓ upon data and and this this generic software will be able to interact with the proprietary data that we have to organize for for our customer. speaker-0: Okay. Now, my next question is one that normally I think I would position as as a risk to a technology company. So just to dig into the AI report survey again, we're entering a bit of a phase of like shadow technology adoption where people are they use their company provided tools, sure, but they also bring their preferred tools, AI tools specifically in from ⁓ from home. That's something that we see happening a lot, ⁓ based on the survey data. It's also something that happens a lot without ⁓ without permission. ⁓ so this in the same way that in the early PC era you had people bringing in personal computers to work because it they felt that allowed them to be it it exceeded the capabilities of enterprise software and stuff at the time. Now that creates a governance problem in the sense that if people are doing enterprise work with personal tools, it is existing outside of ⁓ the the ecosystem and the databases that we talked about. ⁓ But then the way that you've spoken about the interface going away and the what what is and or isn't a commodity, I kind of as part of me wonders if you see this as a threat or if you see this as actually just the way that things are gonna play out and that the value is to connect to those the tools that we're using at home and at work. speaker-1: ⁓ I I think we have to to start by accepting that you can't win anything around this with a policy. So the model someone uses at home costs 20 euros a month and gets better every few weeks, right? Nothing in enterprise procurement moves at that speed. So we have ⁓ already seen in the past, for example, with with social networks, 10, 15 years ago. People got used to seeing intuitive interfaces on Facebook, Twitter, Instagram, and so on. And they wanted the same in their enterprise software. And banning it doesn't ⁓ remove it. It it's it's just it it just moves it out of sight. So for me, shadow AI is not people being undisciplined, it's a signal that the company tools are are too slow. So our position on this is is is very specific. We think an LLM makes sense inside the product. Only when everywhere where where things should run automatically, ⁓ like extracting ⁓ data from a PDF, for example, or classifying data, executing a specific step. and it doesn't make sense when a user just ⁓ could just easily ⁓ ask their own tool, because they all have clawed and chatped on that side. if they can ⁓ if if a person can do the the work alone by asking questions, they can already do it in ChatGPT or cloud. So we are not going to build a chat interface that is better than those. ⁓ The other thing I would say is that that shadow AI is is also free product research for us. ⁓ Because what people build at home or on their own time or now more and more also as tools in the preferred AI ⁓ assistance, you know, when they build artifacts and cloud or stuff like this. This tells you exactly where your software is is failing or what's missing in your software. So we actually ⁓ treat that as a backlog ⁓ in engineering and try to understand ⁓ how we ⁓ should and could integrate. speaker-0: Okay. So then the I think the other challenge with reconciling some of the decision making stuff that we've already talked about is the decisions that people make are governed by the priorities that they have. And if we think about a complex brand workflow across the product journey that's made up of all the different people that we just that we just talked about. ⁓ There's a lot of reconciliation that goes on in the typical product lifecycle. So, you know, one team or person is trying to stick to a a margin target, someone else is trying to optimize lead time, somebody else is trying to hit sustainability targets and so on. Do you think of AI as a way to start to align some of those micro decisions to a common objective? Because I think the part that a lot of companies will get stuck on here is that the right there isn't necessarily a universal right choice for everything. I know we said before there are instances where there is only really one correct choice in a narrow area, but you you can make decisions like sustainability, profitability, time, ⁓ quality, all of these things are they exist on a continuum. And if you pull one of them in one direction, it pushes another one in another direction. that to me feels like the the space where maybe human judgment wins out is that kind of ⁓ prioritization, if you want to call it that. But I'm keen to see what you think. Do you do you agree that that's where human judgment is ⁓ is is going to remain, or do you think actually that's all something that should be feasible to automate as well? speaker-1: It's a complex question. So first of all, I think ⁓ AI cannot ⁓ solve a conflict ⁓ about values, right? If margin, deep time and sustainability, for example, pull into totally different directions, it's not a data problem. It's it's a priority problem. And this belongs to the leadership. It's ⁓ it's a matter of values and company culture. So anyone telling you that AI will decide on this. ⁓ in 2026 at least is is lying to you. and I think it this will become more and more important over time. Today an agent ⁓ does one task, but as they start working across functions, ⁓ we spoke about merchandising and and logistics, for example, we'll have to give them objectives and let them do the work ⁓ in order to let them do the work ⁓ alone, right? And an objective is basically a value or or a set of value. There is this famous example, ⁓ this sort ⁓ experiment where an AI is asked to produce ⁓ as many paper clips as possible. I think ⁓ you know this one. And in order to do it perfectly, ⁓ producing these paper clips, it destroys everything else on the way, including humanity. So this is of course ⁓ totally absurd, but the mechanism behind it. is completely ordinary and very realistic ⁓ in a model because it's just a mathematical ⁓ objective and not nothing in the objective said what was more important than than and paper clips. And the company version of this is is probably a bit more boring, but also very ⁓ much more likely ⁓ because if you ask an agent to reduce ⁓ for example landed cost, it will reduce landed cost, right? And on the way To reducing land costs, it it could quietly trade away lead time resilience, a supplier relationship, or a sustainability commitment if nobody puts that into the brief. And that's very complex. That's that's ⁓ very new tasks and new things that we have to to learn ⁓ within within the companies. When you give an an objective an an objective to an agent, you absolutely have to make a choice ⁓ about these priorities. So otherwise the agent will make them for you and And it will do it very, very consistently with a lot of e efficiency and at scale, ⁓ without telling anybody and that's that's very, very risky. So speaker-0: Yeah. And that's so that's that's like an alignment question. ⁓ but ⁓ I I would say a lot of organizations are not especially aligned in how they value and prioritize human decision making either. ⁓ so that you know, that it is a new problem from a model point of view. I don't think it's necessarily a new problem for a brand to say, Yes, I recognize that we're gonna miss some sustainability targets here, but Profitability is everything. Landed cost is everything for this particular style. So thanks for your input, sustainability team, but it's been discarded in favor of the choice that we've that we've made over here anyway. ⁓ so I think, yeah, ⁓ I I I get where you're coming from. when we think about okay, so let's think about a post a post interface era, ⁓ where the database, the governance, the logic, the rules and so on are the the value. ⁓ The people able to work with those rules, influence them, enter data that corresponds to them and so on are they exist within the walls of a brand, but they also exist across the brand's extended supply chain and partner network. It's not when you think about a big database, it's not just your own people mm potentially entering information into that. If you have a PLM today, you probably have role-based licensing and you have suppliers who are able to enter data and so on. How do we how do you see this whole kind of roles and permissions thing changing in the future? That sounds like a really dry and boring question, but I feel like it's quite important when the value is exclusively almost in that data rather than in anything else you build on top of it. speaker-1: Yeah, I I I think it's sounds actually like a boring question, but it's it's a very important one. ⁓ because before AI, permissions used to say what you were allowed to see, right? And in in an AI world, ⁓ they must say what an agent can see but also can do for you. So that's a very different ⁓ level of risk because we let the ⁓ the agent act. So I think too today we are ⁓ we are there are three things I think that are are changing at the same time. ⁓ the first one is that much more people get access to intelligence. So that's a very big change. So ⁓ data that used to sit behind you would need a report for that from someone becomes now available to anyone who can ask a question. So AI brings ⁓ real ⁓ democratization of of of data to the companies, which also means that people will see numbers they have never seen before and a lot of them are ⁓ simply not prepared for that. ⁓ and and access without training them can produce a can produce very bad decisions in some cases. ⁓ the second change we're seeing is that ⁓ from from now on a user is not always a person. Right? Yes. When an agent asks something on behalf of a supplier, ⁓ whose permissions apply. At Bamburos, for example, the Decision we have made on this, and was a quite complex one, is to attach every agent to a human in terms of responsibilities, ⁓ either as an assistant. So if you connect with ⁓ Chat GPT to to Pembrose and and and work ⁓ through the ChatGPT interface to it, you have ⁓ you attached through through a human, or directly through a private private key that the agent gets. And every call. ⁓ executed, every task executed by the agent gets locked for for traceability. ⁓ and then the the third change that we are already seeing is a strong increase with agent communicating with the with the outside world. So that that's very new but it's it's it's a tendency that is that is starting. In our software we already handle roles for suppliers and partners. and so for the moment users ⁓ humans are big majority, so they're still connecting and importing data, importing documents or or whatever. suppliers bring in their their product specifications and so on. But if you extend that into an Atlantic worlds, we are pre we have to prepare ourselves to see ⁓ our customers agent talking to their suppliers agent with both ⁓ side's permissions enforced. And that's where all this ⁓ this is going. So the position we have to defend is that permissions belong ⁓ into the data layer. So in in our case into our metadata layer ⁓ and not into each application because if every new AI feature invents its own ac its own access rights, you will have five, ⁓ ten different security models in in two years and you don't understand ⁓ anything. speaker-0: Fine. No, I I I also think you're right ⁓ about a lot of things. I think people get obsessed with the idea of agentic commerce and they tend to ignore the the business to business agent relationships that are more likely to constitute the majority of traffic throughput, whatever you want to call it in that regard. penultimate thing. So if I think back to DI survey again, companies tell us that they expect spending on AI to go up in the next 12 months. Now ⁓ a lot of that is apparently earmarked for the expectation that they'll need to spend more on talent or tokens to realize the value of things they're already doing rather than doing anything fundamentally new. But some of it is for is for newness as well. Now, you've made it very clear that you AI is a fundamental shift but shouldn't be looking at this the same way we look at traditional software. How does that influence the way that you suggest companies approach ROI? analysis, ⁓ because we're not dealing with traditional software. So traditional software metrics are probably not well suited. Or maybe they are. Like you give give me an idea on how you think companies should be looking for value from their I spend. speaker-1: Yeah, I think the the the traditional way of measuring the the ROI of software was counting users and and hours saved, right? And I think ⁓ that's that's still very much ⁓ very much true ⁓ today as we're we're speaking because it's the easiest numbers to collect and the easiest to to inflate. But neither of them tells you actually if a decision is is getting better. So ⁓ the examples of of KPIs that we are providing to our our customers or talking about to our customers ⁓ cost, for example. So the cost that you are already paying for something. So for example, if if someone is retyping test results and certificate dates from a supplier PDF ⁓ into the PLM interface, that's cost because it's time consuming. or someone chasing a a vendor per email for a missing certificate and then tracking these expiry dates in a spreadsheet or a merchandiser waiting three days for for for IT to run a query for him, right? That's all that is that is cost. And this has headcount and all the rates attached to it. So you exactly know how much it costs and how much you'll save. And that's that's a very good KPI to to measure where you where you can save save time for for example for the PDF example and PVF PDF extraction, we have a technology for that and it's automatically ⁓ now as we're speaking ⁓ treating hundreds of thousands of documents already for our customers each month and that's a lot of a lot of ⁓ time saved for for the customers. a second ⁓ KPI we which is easy to measure is decisions getting taken earlier. That's not very intuitive. ⁓ but I think it's i it's an important one because the the value is not ⁓ necessarily ⁓ a faster answer. ⁓ it's the answer ⁓ arriving before the decision was locked. So for example, if you find a compliance gap before you avert the vendor instead of after, it's it's worth a lot of saving. So this value will never be measurable in an in an hour. But you can measure how long decisions take by in practice ⁓ taking recurring decisions, ⁓ when the award, for example, and you log when the case decisions ⁓ when the question comes up and when it was ⁓ when it's signed off. So you compare the gap before and after. Those things are already in your in your systems, they are stored in in in Bamborough. So nobody has to fill in a timesheet or so. And when you progressively try to ⁓ reduce the difference. between those states you're creating creating value. ⁓ And a last thing which is maybe a a strange one ⁓ is that is that ⁓ AI there's value related to ⁓ AI ⁓ that you can measure that comes from questions that nobody has ever asked before. Right. Because when asking is getting cheaper, people start to explore more. And with our customers, ⁓ we have built a technology, as I said, which which is called Assist, which helps our our customers to ask questions directly ⁓ to our data and and run queries with using natural language. And this is a good example of ⁓ our users running report requests that never got made before. So when a team can ask their own questions, they stop calling for IT and IT stops building. one-off reports and that produces a lot of value. And you can easily measure it by the number of questions that the users are asking in the tool and by the the number of ⁓ past IT tickets that should be dropping for the IT teams. speaker-0: I I think I think people will have experience of that in their ⁓ anyone listening to this have experience of that in their professional and personal lives is you sit down in front of a text box and the first thing you do is ask everybody asks the same questions. Like in your personal life it's all w the the the the reason it's ⁓ it's a cliche is like, give me a morning briefing because it's just ⁓ there's there's there's th you don't need a lot of imagination to come up with that. You don't need a lot of experience to come up with that. I think the more time people spend having natural language conversations with LLMs that have access to vast repositories of accurate information from across their ⁓ their organization, from across the product journey. I think you're right. I think people will start to ask deeper questions that they don't currently come to mind already. Final question. So we talked a lot about decisions. I think arguably the biggest decision that we haven't talked about is not whether or not to implement AI, because that seems pretty much like a fate accompli. I think it's more a case of when. And the reason that I say when is This is the fastest moving field that I've ever been involved in or analyzed or observed. That the the state of the art yeah. The state of the art advances incredibly quickly. ⁓ you know, you have new model releases with new capabilities that are transformative to what you can accomplish with them every month, ⁓ or like ⁓ week by week and things it seems at the moment. if you put yourself in the shoes of a fashion brand This can seem like just such a moving target where you're constantly well, if the if the frontier is forever moving and if what something is capable of is forever changing, when do I jump in? And what is the first step that I should take? What would you say to a company that feels that way? speaker-1: ⁓ I I think that's that's right. It's perfectly rational. So anything we commit to today will be ⁓ might be obsolete ⁓ in less than a year. ⁓ the last ⁓ three years showed that to us and that's that's very scary. Like we are not built we humans are not built for for this uncertainty. ⁓ but the conclusion we draw from this, ⁓ which is to wait ⁓ until it settles, is the the the main mistakes that a ⁓ mistake that we have to avoid, right? Because it's not going to settle down. There is no stable version coming. So the question is not when to start. The real question is what to build that survives being being wrong. So personally I try to to follow ⁓ some guidelines ⁓ in order to avoid getting personally lost on this. The first one is that I I think we absolutely have to separate what lasts from what doesn't. So we've already talked about that your your data model, your permissions, what a company knows about how it works, that lasts. So you should heavily invest there. anything specific, a specific specific model, a tool or interface that doesn't last. ⁓ it's disposable. So so you should rent it and keep it as replaceable as possible. ⁓ then I think we have to make ⁓ small commitments. So you don't need an AI strategy. You need a few questions ⁓ your team asks all the time and cannot answer them quickly. So you just take the first one, you fix it, and then you run it for a few weeks ⁓ with a result that you can check ⁓ with real users. So that's that's the right ⁓ the the the approach to to take, right? then building everything so that it is easy to reverse ⁓ I think is a is is an important ⁓ thing in a fast moving world. The cost of a wrong decision is not the decision itself. It's how long you stay stuck with it. So the quickly you are able to reverse your decisions, the more value will keep ⁓ at the end. And and finally I think the ⁓ yeah, that's that's the that's the way we run ⁓ this ourselves at at at Bamburos. And For for us what lasts is is the metadata layer, the permission model and so on. And what doesn't last is all the the models, ⁓ the models around. So every architecture decision we make is undergoing the same test, ⁓ which is if the model underneath changes next quarter, how much of this ⁓ do we have to throw away? So globally I think the only ⁓ really losing move in in in today's world is to spend a year waiting for clarity that will that will never come. speaker-0: Think, yeah, I think that's right. And I think the other thing from my side that I would add is, and it's been a big unlock for me personally and professionally, is using the agentic side of things and ⁓ so schedule jobs and loops and things like that as an opportunity to ask yourself the schedule you do things on, and when it is that you that that external Prompting validation and things is helpful. So in a very like prosaic personal way, it can be a case of, okay, ru every Friday, scheduled job runs and it prompts me to go back through and find any emails that I've missed. Like just just the most basic use case you can possibly imagine. Now, the model can change around that, but it for me as an individual, that's a persistent element of my work is I've then discovered, okay, it's it's helpful for me to have that done for me on a Friday and that that is beneficial to the way that I operate. You can extrapolate that very boring and tiny example into a lot about how you think about workflows and calendars and everything else that exists around a product journey is to say, well, if we put some of this on autopilot to some extent on a schedule basis and we have jobs and loops and things that run on a regular basis, what would they be? And how would we scope them? ⁓ because then if you add new capabilities to them afterwards, it's additive rather than destructive. And at the very least you have found something out about the way that you operate. speaker-1: That's that's absolutely right. And ⁓ the the only thing that you have to w where you have to make a difference is between what's just pure reasoning for the task you want to resolve and where is value, like personal values or political values or stuff like this coming in. ⁓ because in the first case, if it's just pure reasoning, logical reasoning, a model gets replaceable. You get a better model or a model for another vendor. or an open source model, you'll get more or less to the same results because it's a logical ⁓ solution there. If you get to something which needs ⁓ judgment, we are we were talking about sustainability, but in many bigger companies, politics and international the international ⁓ geostrategical situation is coming in. In this case, ⁓ a model is not just replaceable anymore. If you take ⁓ a US vendor or European vendor or a Chinese open source vendor, will make a big difference. So and that's also something that we we ⁓ have to take into account in when we are making these architecture choices. Yeah. speaker-0: All right, perfect. ⁓ Rupert, I've enjoyed this chat a lot. Thanks for joining me. Thank you. And that's the end of my chat with Rupert. It was a long one, so I'm gonna get out of your way quickly. I think though that the most important thing to take away from today, and the most important thing for you to go think about, is the idea that software, as we know it, is up for interrogation and rebuilding and re-architecting just the same way that processes are. I think people tend to think of the AI era as something that's happening to them, steered by technology companies. But I want you to remember that those tech companies are part of the same upheaval and that the forward thinking ones are the ones who are figuring out how to get ahead of it all, the same way the forward thinking brands are. Different topic next week, so for now, thanks for listening and I'll speak to you again really soon.