Fashion’s next looming AI challenge has nothing to do with uptake. In the survey portion of The Interline’s AI Report 2026, we found that only around a quarter of respondents trusted AI’s output enough to base critical decisions on top of, and we identified a prevailing concern amongst practitioners that companies (and especially innovation leaders) are overestimating the capabilities of artificial intelligence and the extent of automation.
It’s clear from the data that fashion’s spending on AI is set to keep rising, but access to the technology is evidently not the same thing as establishing trust – either in its standalone capabilities, or in its place at the heart of brands’, retailers’ and makers’ technology estates.
On 15th October 2026, The Interline brought together three executives to examine that gap across product and line creation, strategic sourcing and production, and consumer-facing content. Hosted by Ben Hanson, our Editor-in-Chief, the conversation featured:
- Brian Lindauer, CEO of VibeIQ.
- Rupert Schiessl, Chief Strategy & AI Officer at Bamboo Rose.
- Vlad Mulhem, CEO and Co-Founder of Voxelo.
The speakers all had their own different conclusions about how AI is set to change the work, and the working environment, of fashion, and each of them also fielded audience questions. Now, the replay of that live session is available to watch for the entire audience, along with a full transcript below. For more from the research, editorials, and interviews behind the discussion, download The AI Report 2026.
Transcript
This transcript has been edited for clarity and length. Timestamps are included to allow easy cross-referencing with the video recording.
Opening and the trust gap
Ben Hanson — 00:00
Good afternoon, everybody, and welcome to The Interline’s webinar focused on closing the trust gap in AI. We’re joined by people from around the world, and I’ll introduce them as we get into this. This is a panel discussion, so it will focus on me asking questions of each of our speakers.
We have a lot of people registered from different time zones. It may be morning, afternoon or evening for you. Thank you for joining us live, or on demand. If you’re here live, you can type questions into the Q&A box. I’ll either address them myself or put them to our speakers. I may pick some up as they come in and save others for the end, so don’t worry if we haven’t acknowledged yours immediately.
We’ll be live for about an hour. I’m going to exert my host’s prerogative and spend about 45 minutes interviewing our speakers, followed by ten or 15 minutes for audience questions.
Before we get into the topic, why did we convene this webinar? What are some of the key findings from the AI Report, and what are we going to talk about?
If you’ve read the report, you’ll know that AI adoption among the 100-plus brands and retailers we surveyed is high. The vast majority of people across different roles, disciplines, company sizes and categories use AI. I think more than 90% use AI at work, and more than 90% use it at home in various forms.
The gap we’re here to discuss is that only around a quarter trust its output sufficiently to base mission-critical decisions on it. That’s a big difference between the majority using something and a relatively small proportion actively trusting it.
In spite of that disconnect, AI spending, expansion, outlook and attitudes are all on an upward trajectory. The majority of companies believe they’ll spend more on AI in 2027 than in 2026. They also believe more of their outward-facing content, whether for wholesale partners or consumers, will be mediated by AI: either generated or enhanced by it.
At the same time, fashion professionals tell us their foremost concern is the gap between the capabilities they perceive and what AI can actually deliver. That comes far ahead of what you might expect to be the bigger concerns. Looking at the headlines, you’d expect the ethics and environmental impact of generative AI training, or concerns about brand identity, to be high up. Instead, the tension is between what we believe AI can do and the sky-high expectation of what it should do.
I’m joined by three people with different perspectives on that gap: Brian Lindauer, CEO of VibeIQ; Rupert Schiessl, Chief Strategy & AI Officer at Bamboo Rose; and Vlad Mulhem, CEO and Co-Founder of Voxelo, an AI and 3D company.
My first question is for everybody. I want to challenge the title straight up front. I’ve explained why I think this trust gap exists, and our data tells us it does. Do you see the same thing? Is some distrust still reasonable when we’re talking about high-stakes applications in product creation, sourcing and communications? Vlad, I’ll come to you first.
Vlad Mulhem — 05:44
We’re on the content side, usually facing the customer, and we see mistrust of AI-generated content from both consumers and the brands and retailers we work with. That’s driven largely by generative AI producing things that are plausible, but not exactly accurate.
People are excited about what AI can do because of rising demand for content and the reduction in production costs. The benefit is clear, but the outputs are difficult to control. That’s the key, and why we exist. We believe we can bridge that gap by capturing a digital twin of a product using real cameras and real product pixels, which can then act as context.
Many of you will have used LLMs and know their accuracy improves when you give them context: knowledge of what you’re asking them to do, and background on you and your company. It’s the same with images. The more real background you give a generative AI product, the better or more accurate the output.
The trust gap is there, and it’s now also backed by legal requirements, particularly in the EU, to label all generated images with AI-generated labels. We’re definitely seeing that gap.
Ben Hanson — 07:27
Rupert, you have a different vantage point because you’re embedded in decision-making rather than all the way downstream. When you talk to brand customers and contacts, do you see this gap between what AI is capable of and what people are willing to entrust it with?
Rupert Schiessl — 07:50
Totally. Distrust is reasonable because, as with humans, trust has to be earned. People, including our customers, often associate explainability with trust, but explainability is not trust. AI doesn’t take responsibility for anything.
At Bamboo Rose, for example, we handle a lot of compliance files, documents and test reports. If you sign off a compliance file, you have to own it. You can’t own something you cannot explain. When people use AI every day but don’t trust it for important decisions, that isn’t resistance to the technology. It’s people applying their normal standards.
We could ask how to make people trust AI more, but I think the real question is how to make checking cheaper. If verifying an answer takes longer than doing the job yourself, which is often the case, the feature is dead in advance, whatever its accuracy. If verification is fast, trust stops being a problem. We have to give people the opportunity to build that trust in the tools, and we’re working on that.
Ben Hanson — 09:28
What does building trust look like in practice? Is it exposure and experience, or institutional training and knowledge? If we think about an AI deployment, which most people have some familiarity with, what’s the gap between having it put in front of you and that trust being earned?
Rupert Schiessl — 09:56
It’s a matter of experience. In a human relationship, the more often you have a good experience with someone, the more you trust them. It’s the same with an AI feature. The more often you get an accurate, interesting, fast answer that helps you, the more you start to trust the tool.
There are many things behind that, but a key one is the data and its quality. The better the data, the faster you build trust in your tools.
Vlad Mulhem — 10:36
Back to your previous point, Rupert: verifiability across stages. AI systems are black boxes, but the more you can verify their output quickly, rather than checking every output, the more trust we can build. With imagery, there are ways to do that automatically or semi-automatically. By gaining experience and flexing that muscle, people will trust the process more.
Ben Hanson — 11:10
Brian, now we’ve stopped imprisoning you backstage, the same question to you. Do you agree there’s a trust gap between AI as deployed and AI as institutionally adopted and used every day? Where does it come from, and what can overcome it?
Brian Lindauer — 11:31
First, I sometimes trust AI more than I trust myself to figure out Zoom! As Rupert said, trust is earned, and it isn’t a single topic. I trust my kids to do some things very well, and I don’t trust them to do others.
When we talk about trust in a corporate process, decision-making or content generation, there isn’t one answer. The more you use AI, the more you trust it to do certain things, while still not trusting it to do others. This isn’t a black-and-white issue. It’s very circumstantial to the task at hand.
As with human trust, you earn it through trying things. Experience and iteration build it up. My observation, across industries, is that there’s a gap in awareness of what AI can do. You don’t become aware of what it can do unless you try to use it for something. If you don’t know it can do something, you’ll never trust it to do that thing without trying it. We have to find a way to try and iterate.
Every month, with every new model, people will find it’s capable and competent at certain things. That bar is rising every day. But if you aren’t trying it, you won’t know what it can do, and you won’t establish trust. It is cyclical. You have to try it and expect that it might fail. Increasingly, it will succeed, and you’ll build trust.
Connecting AI to the business
Ben Hanson — 13:51
I’ve had that experience myself. I saw a post from Andrej Karpathy, one of OpenAI’s co-founders, I believe, reiterating that most people’s personal exposure to AI is to basic chatbot features in the free tiers of ChatGPT, Claude and so on. That doesn’t give you the frontier learning, immersion and experience we’ve just discussed. You don’t get to try advanced capabilities, multi-hour tool calls or computer use because they’re currently behind a paywall.
Brian, I want to stay with you on the company side. We found fashion companies deploying AI off to the side of their existing technology estates. A relatively small proportion, I think around a quarter, told us it was fully and seamlessly integrated.
The cycle is that people don’t trust the answers, so they don’t connect AI to the systems holding the sources of truth or data that could generate better answers. Do you see that playing out?
If you’re at the frontier, you’ve probably also used AI as a brute-force integrator. Big brands are accustomed to siloed applications requiring bespoke integrations or extensive development to use their APIs. You can sit down with Claude and ask it to pull information from systems X, Y and Z and create insights from it. Tell me what you see in terms of AI being disconnected from the rest of the technology estate, and what you think of that kind of integration.
Brian Lindauer — 15:58
That’s a simple question! I’ll do my best. There’s nothing simple about this. Building a product line is one of the hardest things on the planet when you consider how many people, systems, decisions and data points are involved.
Suddenly, there’s a superpower coming into the mix: a gigantic change. It’s like going back to 1890 or whenever cars were invented. The cities are full of horses and carriages, with horse-and-carriage companies and vendors, and an automobile drops into the middle. The city isn’t ready for it.
Here we are with a power that hasn’t been fully tested or understood and is growing every day. The biggest challenge is change management, process engineering and system integration. Integration isn’t new; people have been trying to do it for a long time. But the value of putting AI into your environment, where it can access systems, understand and synthesise information and make decisions, is higher than ever. There’s still a lot of work needed to tap into that power.
I believe AI will transform how companies work, but getting there isn’t just a matter of turning on Claude Desktop or ChatGPT. AI needs to navigate your ecosystem. You need to roll out capabilities and give users prescribed ways to use them. It’s a massive change management challenge, or opportunity. You can’t just flip a switch.
Integrating systems and giving AI somewhere to read accurate information is essential to accurate, trustworthy decisions. But that isn’t straightforward. You don’t throw Claude into SharePoint and suddenly have it understand everything needed to build a product line.
That’s the world VibeIQ works in. You need somewhere your product line is defined, where you can say, “That is my trustworthy definition of the product line.” It might be several systems. Then you can put AI on top and do powerful things.
Ben Hanson — 19:02
The car analogy is a good one. It might not resonate as much if you live in an American city with a grid layout, but it will if you’re used to driving down Europe’s tighter streets, which weren’t designed for cars.
The other part of that analogy is that you can look at a car and want to drive it, but the other driver of technology adoption is that it can get you somewhere you need to go.
Vlad, sometimes people look at AI and think, “Everyone tells me this is the future and will transform my work, so I need to implement it.” Then there’s deploying it in response to an actual pressure point. Content creation seems to be one of the places fashion feels that most acutely.
There’s tremendous demand for visual content, and an unmet need to put more written and visual representations of products in front of people so they can make informed buying decisions. There are surveys suggesting that broader coverage of a product on a product detail page or campaign leads to higher conversion and engagement.
Historically, scaling production through photography or traditional 3D modelling, meshes and texturing meant scaling overheads too. That creates tension between what the industry needs and what brands can produce. How can brands stay on top of demand with AI without scaling overheads or breaking the trust promise you mentioned?
Vlad Mulhem — 21:15
We understand generative AI’s ability to produce content at scale, but the trust gap remains because it looks plausible. Something plausible can create an uncanny valley, and you can miss that it isn’t accurate. That’s where the reluctance to use it in day-to-day production comes from.
We start with a verifiable input. We reconstruct from photography, helping people use existing pipelines and familiar tools to capture a 3D digital twin that’s accurate to the source. The more true context you give AI, the closer the output will be to that source. It’s also verifiable against the source across the chain, because the output starts from something true to life but digital.
It isn’t just one image; it’s multiple angles of the same product. We can extract metadata that’s true to the object. The creative output can then scale with the confidence that it started from your real product. AI knows that product and can generate something you can trust.
You can become imaginative with the uses of that because you’ve given a digital tool an input rich in information. Replacing a background while keeping an object’s identity is a simple task, but we still don’t trust it because AI may put the thing somewhere we don’t want. Can we control that process? Can we preserve the identity of people we use as models, making that traceable, remunerable and ethical?
Many of those questions can be answered with tools that exist today, but they aren’t readily available to everyone. As Brian said, it’s like not having the network, petrol stations, mechanics and ecosystem around this new power. We have AI’s raw power, but we’re still building the workflows and tools. That requires hands-on work: rolling up our sleeves and solving individual problems.
Ben Hanson — 24:23
Our survey gives some validation for that. People perceive AI as most mature at the beginning and end of the product journey: in market, trend and competitive analysis, in concept generation, and downstream in visual content creation.
The mechanics of making a product seem to be where it’s perceived as least mature. The closer AI gets to patternmaking, costing, sourcing and so on, the less mature people think it is.
Rupert, you’re embedded in decision-making across that lifecycle. What needs to change? Is it giving AI more grounding data and better product and business context, as Vlad said? More fashion-specific reinforcement learning? More capable models? What would embed AI in the heart of product creation?
Rupert Schiessl — 25:38
Data and grounding are far ahead of everything else in the priorities. We’ve talked about change management, but change is needed on the technology side too. That’s our focus at Bamboo Rose. We manage supply chain data that comes from ten or 20 years of business history.
Today, we don’t have an intelligence problem. The models are very intelligent, much better than our ability to feed them. In design or marketing, it works well because you can hand AI something. It can already read a brief, an image, a design or a piece of text. In costing and sourcing, you have to provide the system and its data, which is more complicated.
Enterprise software was conceived with data organised for a screen to display to a human. It wasn’t designed for an AI agent to understand and interact with. A column name may not mean anything without a description or label. Large language models want descriptions, text and labels to understand what’s happening. Much of what we call software still lives in those interfaces without that second part.
At Bamboo Rose, we’re bringing that understanding into what we call a metadata layer. We want to structure the descriptions of business processes and the data customers have accumulated over the years, making them available for AI agents to use and get value from. Structuring and making that data available is our number-one priority.
Choosing products and keeping humans in the loop
Ben Hanson — 28:01
Brian, let’s get into some specific use cases. Your world precedes sourcing and execution: deciding what to design, develop, adopt and create. Rupert said we don’t have an intelligence problem. I don’t think fashion has an ideas problem. I don’t think anybody is particularly short of creative concepts they want to try.
It might be useful for AI to suggest colourways, but the bigger operational bottleneck is deciding which creative options to adopt and develop. Discerning a viable concept from a non-viable one is a specialist activity. What can AI do to improve those decisions, and is there a horizon where it could make some of them itself?
Brian Lindauer — 29:26
Let me set the table. Imagine a brand taking 200 CCs to market: not a huge product line, perhaps footwear and apparel. They’re at the start of the calendar, with nine or 12 months to get to market. Those 200 CCs came from a plan with parameters around pricing, breadth and depth, flow, volume, revenue and margin.
To turn that into the product line you buy and source, think about how many actions, tasks and decisions happen over that period. I don’t know the actual maths, but it’s probably tens of thousands of tasks and thousands of decisions. It’s one of the most complicated processes out there. I’d much rather build software than a physical product line. That’s why I’m in this business.
There are lots of types of decisions. We can generate designs, look at the images and ask whether that’s an accurate representation of the product, or whether the material is rendered correctly. We can ask subjective questions: will my customers buy it? Is it on brand? Does it fit my product line? We can ask about cost, margin and where to make it.
We can also ask about the assortment. I’m supposed to have a particular breadth and depth; I don’t need 27 blue T-shirts. I don’t care whether they’re accurately rendered or whether I can manufacture them. I have too many options for blue T-shirts.
That’s the world VibeIQ plays in: the product line and the products that make it up. We think through what merchandising, design and development need to consider, and how that ends up in the supply chain.
There are decisions AI can help with easily now. It can assess breadth and depth. That isn’t rocket science, but it needs somewhere to look at the current product line, the concepts we’re working on and the plan, and help us navigate between them.
What it will do less well is answer whether consumers will buy a product. That’s a loaded question, and I don’t think AI will answer it in the near term. There are efforts to create synthetic audiences and use historical analysis to predict, for example, whether a blue linen product will perform in a particular market. But it’s tough to assume past performance predicts future performance. In that subjective space, it will take longer to reach a point where you feel comfortable.
For tactical, operational, merchandising, cost and margin questions, we’ll see more progress: is this the right set of products to spend time on and take into the supply chain? Do they fit the top-down plan and the parameters for success? I feel confident there’s a lot of low-hanging fruit there. If we work through change management and get to deployable solutions, beyond people with Claude on their desktops, there’s a lot of benefit to unlock in the short term.
Ben Hanson — 33:23
Rupert, in our recent podcast interview we talked about how far decision-making automation might go, and how quickly. Thinking about the lifecycle stages that lead to a physical product, the operating framework at the moment is always, “We need to keep a human in the loop.”
I can see that for specialist disciplines or things unique to a company. But, as we discussed, there are other areas where having a human in the loop could become gatekeeping, creating friction in processes that might work better if they were fully automated.
In sourcing, supply chain and product development, how much do you expect people to be doing in two years? Will they be doing a higher-order version of their job, or will there be fewer people in those roles? The hidden part is that you keep a human in the loop until someone says, “We don’t need them anymore.” A lot of people want to know when that is. I know you’re fairly optimistic that it could be sooner rather than later.
Rupert Schiessl — 35:11
Having a human in the loop can only work if that human can verify the data. Most of the time today, with AI, they can. The problem isn’t that AI is sometimes wrong; of course it is, as humans are. The problem is that you don’t know which answer is wrong, so you have to check all of them.
If you have to do that every time, you automate nothing. You’re replacing doing the work with verifying it. For me, that’s one of the main reasons so many AI projects aren’t showing the expected return.
Take the compliance example. You can have thousands of product certificates with their contents extracted by AI into your software. If the compliance officer must reopen every document to check the extraction, you aren’t saving anything. You’re transferring workload from production to verification.
On the other hand, AI can be built to justify its answers, give detailed examples from the extraction and produce quotes, as we do at Bamboo Rose. The user can quickly see what’s right, what isn’t and what needs attention. They might know to focus more on a type of document that’s wrong more often than others. Then you start saving time and building trust, and the human-in-the-loop function begins to work.
Consequently, you reduce the human time needed for those tasks by transferring them to the machine reliably. Anyone who tells you AI won’t replace jobs is lying. If the company is growing, you can grow while keeping the same people. If it isn’t, you need fewer people to do the same job.
New jobs and skills are needed: managing agents, judging exceptions and maintaining the business and technical knowledge required in a complex industry like fashion, apparel and retail. That knowledge is still necessary to build products, but a large part of the repetitive work will transfer to machines.
The economics of content creation
Ben Hanson — 38:17
I broadly agree with a lot of that. Vlad, alongside the human-in-the-loop argument, another phrase we hear is “execution is cheap”, usually in relation to software development. Historically, developing software required specialist skills and training. Now people can do it relatively cheaply, putting a higher value on ideas.
What’s your take from the content perspective? The economics of visual content in fashion have been fairly fixed for a long time: photography, studios, locations. I’m sure the costs vary; I’m not a photographer or stylist. But the costs of authoring in 3D, including software licences and specialist skills, have also been relatively static.
What does AI do to the unit economics of content creation? And what does pairing it with 3D do to unlock further iterations and extend the value of content once you have it?
Vlad Mulhem — 39:36
From our work with retailers, brands and large enterprises, their internal teams cannot keep up with the increasing demand for content. We see those people being supercharged to deliver more, rather than replaced.
The tooling isn’t there yet. We’re asking Microsoft Paint to deliver a Photoshop-level masterpiece. The tools are evolving, and we believe 3D has a big role because it grounds AI outputs in a digital twin that’s faithful to reality.
On the production side, the technology we’re developing, Gaussian splats, can democratise 3D creation by removing the laborious intermediate processes between capture and a finished model. It can make the technology available to people who haven’t had access before.
Teams of photographers and creatives, which will probably remain a similar size, can use it because we allow people to create a 3D model with photography equipment. They don’t need technical knowledge. They can get a model faithful to the original with a few minutes of input using existing skills.
What happens downstream interests us. Those creatives can adapt to new tools using familiar concepts. They have a virtual object and a virtual camera, similar to CGI or its promise, but the outputs are immediate because AI takes the object and camera and grounds the creator’s intention.
It’s about evolving creatives’ understanding of tools, as happened with Photoshop, digital retouching and digital cameras. People worried about those replacing things and skills, but demand for content grew as the supply of tools that could produce it more quickly increased.
We tend to think something will be replaced. Email didn’t make us read or write less; it made us read and write more. We fill that space with more efficiency, and the market demands it. That’s where I see things evolving. Democratising the tools while retaining creative control in the age of AI is the most important thing.
Measuring value and defining useful milestones
Ben Hanson — 42:38
Brian, I’ll try another AI cliché: AI is good in verifiable domains, where you can look at something and say it worked or it didn’t.
Your space is fuzzier when it comes to return on investment. With complex decisions based on many variables and fast-moving information, there may be no objectively right answer. Or different functions may have different right answers that need reconciling.
The value might be something that never happened: you didn’t bring styles to market that the market didn’t need or couldn’t sustain. How do you judge that less clearly defined ROI, compared with producing more content that converts, or developing software that compiles?
Brian Lindauer — 44:00
You keep giving me these easy questions! First, let’s be frank. We’re not at the point where AI will tell an outdoor apparel company exactly which products to take to market and everything about them. Maybe one day, but we aren’t there.
Measuring whether a decision was good is an industry problem with or without AI. How do you attribute the performance of a product line and inventory to your decisions? That hasn’t been solved. Was it the designs, the breadth and depth, or something else? Those things pile up, and ultimately you measure markdowns and sell-through. We have a way to go.
Let me distinguish automation, analysis and decision-making with an analogy. I wake up and need to commute to downtown Boston. I open Google Maps. It says one route will take 45 minutes, another 47, and another will save tolls but take 57. Which do I want? I decide to pay the tolls because I care about getting there faster.
I made the decision, but Google gave me the information I needed. It automated the extraction of that information and put it in a form I could understand. I could say, “Thank you. I’ll take that option.” I think that’s where we are now.
Brian Lindauer — 46:06
The impossible task for a merchant, designer or developer is understanding everything happening across 200, 2,000 or 20,000 CCs. AI can pull information forward, put it in the right form, and automate its collection, aggregation and synthesis. That puts people in a position to make a good decision. The human makes it.
Engineering that workflow is one of the biggest opportunities for the industry over the next 12 months. Put the people who make decisions in a position to make the best ones. You probably won’t ask AI to make those decisions for you. How you then measure whether a decision was good is another can of worms.
The opportunity now is automating collection, synthesis and the compacting of information, and giving it to me in the right place. Right information, right time, right place, so I can be the most effective human in the loop. We can build on that.
In this conversation, we need to identify reasonable milestones for a company or the industry next year. It won’t be that everything is automated or agent-driven. It could be that particular people are more effective at particular milestones in the calendar. That’s a good marker. How do I use AI to get there?
Ben Hanson — 47:49
I think that’s a good analogy because you’re presented with three routes, not 50 or 100. There may be 50 or 100 viable routes, but are they optimal against the variables you’ve chosen? Cost, perhaps, or maybe your right indicator doesn’t work and you don’t want to turn right. There’s a lot of information to weigh.
It seems a good short-term goal for AI to define a subset of the best options and let an expert choose the optimum one.
Brian Lindauer — 48:36
We could make that tangible. As a merchant, what are five examples of that Google Maps analogy? At this point in my calendar, I’d like this information so I can choose option A, B or C. That would help me.
Find five examples for merchandising, five for design and five for development, with that level of specificity and a good analogy. A company can build a plan around that. There’s something tangible, a place you can picture getting to. Then you can start the change management process and build the steps.
At the moment, it’s often, “AI: what are we doing with AI?” We need to turn that into stepping stones we can rally around, communicate and track progress towards. That’s the hardest part. A lot of people on this call have deployed technology and know the technology is often the simplest element. Rolling it out and changing processes is harder. AI amplifies that, almost by an order of magnitude. Being able to talk concretely about milestones will go a long way.
What makes a brand distinctive
Ben Hanson — 50:07
I’ll fast-track the last couple of questions so we can get to the audience. Vlad, thinking about AI as an amplifier and the democratisation you described, there’s a movement towards recreating established software. People are using AI to recreate things like Adobe Creative Cloud or Microsoft Office, which have been the de facto standards.
There’s a separate argument about whether that’s legal or whether the recreations are any good. But if everybody can access the same tools, build things and create things, what do companies need to do to stand out?
Vlad Mulhem — 51:06
The same things companies have always done: build brands, be authentic and connect deeply with customers. The content and tools are just tools. Democratising them improves brands’ capacity to achieve their strategic goals and build customer confidence.
Give customers as much information as possible to make the right buying decision. Show how a product is made, whether it will fit and whether it’s right for them. The more you do that, the more loyalty you’ll get, because customers who buy the right product are more likely to return.
Standing out in a sea of sameness has always been a challenge, ever since technology began democratising commerce globally. It’s still about the core principles. With the right tools, a brand can confidently say its visuals are authentic and true to the original, explain how they were captured and offer a better experience through 3D, augmented reality or virtual try-on.
Customers can buy online with confidence, as they might in a traditional shop, because you’ve given them those tools. That’s what democratisation is about, and it can help smaller brands and enterprises alike.
That’s how people win, rather than relying on a churn of things and always demanding more. It’s about the quality of the thing and the tools around it. Human judgement is as important as ever, if not more so.
Building around changing models
Ben Hanson — 53:21
Rupert, anyone implementing AI over the last year has faced the challenge that the core technology changes faster than they can deploy it. There’s a new model every week, even just from the major labs. The release window has compressed to an average of something like 11 days for new products.
Then there are separate classes of things that aren’t strictly LLMs, such as OpenAI’s Decisions API and Jev. Implementing AI and getting value from it is a moving target like I’ve never seen before.
How should brands and retailers approach that? Is staying at the frontier important, or do the product and brand matter more? How should people think about it?
Rupert Schiessl — 54:23
It’s important to stay at the frontier, as with any technology, but things move so quickly that doing so all the time is expensive.
At Bamboo Rose, we try to distinguish the things that move fast, such as models, from the more stable things. With LLMs, the leader changes every day. We can’t build everything around one model. Our software needs to make them replaceable.
The more stable things are private to our customers, and that’s where we consider the real value to be: their data, how we organise and describe it, and the metadata we build on top. We’re investing heavily in that. Those things have to work together with fast-moving commodities. Models are commodities.
If your assortment decisions rely only on the best foundation models, your competitors will do the same. But if your software organises data so frontier models can access your private business knowledge, accumulated experience, supplier relationships and processes, you have something exclusive. OpenAI, Gemini and the others don’t have that knowledge. It can give you an advantage in the market.
That’s our focus. There’s a lot of noise around new technologies, and I don’t think we should care too much about that.
Audience questions
Ben Hanson — 56:29
We have two minutes left and some good audience questions. Rupert, Katrina asks how companies can feel comfortable giving AI access to that IP: trade secrets, sensitive documents and the things that make a brand unique. If they want to use ChatGPT, Claude or Gemini, how should they think about that?
Rupert Schiessl — 56:58
The basic answer is that AI is another piece of software running in the cloud. We have to implement the same security and data confidentiality practices we already apply to cloud software.
We have contracts with providers and security measures to ensure data is encrypted and not shared with other customers. Ultimately, with OpenAI or whoever supplies the foundation model you provide to your customers, you have to trust the other party. The contract ensures that trust in the same way as buying Salesforce or other software hosted on Google Cloud, Azure or elsewhere.
Ben Hanson — 57:50
The second question is whether demand for consumer-facing content and differentiation will prompt brands to place more value on their product data.
Every brand and retailer has faced the challenge of wanting to put more content in front of consumers, offer personalisation and create a different experience, without having the full spectrum of core product data available to build on. Anyone can take that.
Vlad Mulhem — 58:34
When technology gives us one format, one digital twin that can carry all that data, it will be easier for retailers to access it from the supply chain, provided there’s an ecosystem around it. I think that ecosystem is emerging.
People are waking up to what a digital twin can do in fashion. The term has been used extensively in industry, where established practices allow component parts to be exchanged in a machine-readable format that can be traced and reused.
I think we’re seeing the same across consumer products and fashion. A product can have its digital passport, certificate of origin and supply chain traceability, as well as its physical manifestation in a 3D format. Because that is now so easy to create, it can start propagating through the supply chain. But it will take time.
Ben Hanson — 59:51
Brian, I’ll put this one to you and paraphrase it slightly. Paulina asks how AI might affect the way consumers discover and buy fashion.
If agentic commerce becomes as big as people suggest, could it change which products brands and retailers bring to market? A new channel gives you a new way to sense demand and understand customer data, and another lever in deciding what to make in the first place.
Brian Lindauer — 01:00:35
It’s a good question. I’ll shift it slightly and say what I hope happens. A good outcome would be brands making even better products, and consumers finding and enjoying them in new ways, whether through personalisation, better technology or better fit.
At the dinner table with my family, we have pessimists and optimists, and everyone wonders where AI will take us. I like to ask what the best-case scenario is and what we hope for.
It’s a nice thing to hope for that we create and enjoy the best possible products to make our lives better. Whether that comes through agentic commerce or agents designing things, I hope that’s the outcome. I’m excited to see how it plays out.
Closing
Ben Hanson — 01:01:42
I like the optimism. We’ll be here at The Interline to chart how these things develop next year and beyond.
Thank you for your questions and for listening. Thanks to all our speakers for fielding my questions and everybody else’s. If you haven’t already, I encourage you to read the AI Report. Each of these gentlemen is in there, and some have been interviewed separately as well. Thank you for your time, everybody.