What Should A Fashion AI Lead Actually Do?

I’ve seen a lot of AI Lead job descriptions in the past 6 months, and they all ask for the same things: a deep technical background, experience in data science or IT transformation, and fluency across multiple tech stacks. They are, in other words, technology roles.

That feels wrong to me. I held exactly this kind of role at G-Star RAW for close to three years, and when I look at those job descriptions now, I notice what’s missing more than what’s there.

I know, first-hand, that the hardest parts of leading AI in a fashion organisation have almost nothing to do with technology. In this article, I want to explain what those missing pieces are, what I think it looks like to solve them, and how you can apply some of what I learned to your own AI strategies and transformations.

Piece 1: don’t underestimate the friction built into what seems like a straightforward vision for the future.

When we launched Maia, an AI-powered back-office employee at G-Star, the original problem we’d set out to solve was straightforward, well-scoped, and it had a clear value return. Our employee support teams, covering corporate and retail across multiple offices and languages, were being overwhelmed with support tickets. Hundreds of questions, often the same requests repeated for different people, in different languages, across different time zones. It was unsustainable and it was slowing down operations.

Back in early 2025, we followed the prevailing industry advice and built a solution that brought high value to many users by addressing all of those pain points in a form people were familiar with: an AI chatbot. 

But despite this being a complicated systems and architecture task, we didn’t want to make it sound too technical – so we framed the project as introducing an “AI employee” built on Microsoft Azure, integrated into Microsoft Teams, that could initially answer back-office questions instantly, in any language, any time of day.

This seemed, on the surface, like a straightforward way to apply a novel technology to a longstanding problem, and a chance to improve people’s everyday lives in the process. What we underestimated was what those support questions represented to the people answering them. Because tickets were just the outward manifestation of something much bigger.

We started with the IT Service Desk, followed by HR and legal. Many of the employees answering those high-volume, high-velocity questions had built part of their professional identities around being the people who knew things and could fix them. They were praised for fixing problems quickly and took real pride in that. It was easy for an observer to look at their workload and see scarce resources carrying a heavy burden, but it needed a deeper and more contextual look to understand that the status quo had been the texture of people’s working lives for a long time, and being asked to give that up – and to train an AI on it – felt like giving up something else.

When we asked them to document all of the  knowledge they’d built up into a shared database so Maia could take it over, we were essentially asking them to hand over what made them valuable. While some of them were eager and jumped on the tasks, for others, understandably, the documentation didn’t become a top priority.

The test runs weren’t much easier. Getting people into the room to help verify that the AI was doing a good job, or to provide feedback on what to fix, took real effort. It required constant reminders and active buy-in and championing from managers. When we launched the project, we invested heavily in internal marketing, newsletters, town hall announcements, just to keep the project visible and moving.

The shift came through word of mouth. As people started using Maia and finding her genuinely useful, a phrase started circulating around the coffee machine: “Did you ask Maia first?” That’s the moment we started seeing adoption move from something you push to something people pull toward themselves, but that adoption, realistically, had little to do with the technology layer underneath.

And it was only with that earned support that we got the runway to extend Maia beyond back-office tasks to include business-level knowledge. We next made sure she understood G-Star’s specific acronyms and internal context, so people instinctively understood that she wasn’t a generic GPT-style chatbot. We embedded the brand’s tone of voice, brand history, fashion DNA and we even trained the LLM with the brand values and code of conduct. 

For many employees, Maia was their first real interaction with AI with that kind of context and grounding. They might have played around with ChatGPT or Gemini on their personal devices, but this represented a higher-stakes moment. We needed the experience to feel safe, trustworthy, and high quality, because the impression it left would shape how open those employees would be to every AI project that came after.

For the brands reading this report, I cannot emphasise this enough: trust in AI has to be earned, no matter how clear-cut you think the value proposition is, and no matter how strong of positive impact you think it’s going to have.

Piece 2: there’s a bigger gap than you think between what AI can do and what different teams believe work actually is.

Naturally there is an assumption in most organisations that AI should be led by IT or data analytics teams. It makes sense because those teams understand the technology, they manage the infrastructure, they speak the language of systems. They even studied AI back at school, even if it had a different label at the time.

The problem is that understanding how technology works and understanding how a business works are two very different things, even if they both trade in logic. And there are very different sets of barriers and blockers that each party perceives.

I’ve sat in meetings where a business team asked about using an AI tool and were left without a path forward, because the Data or the IT team steered the conversation toward what the policy didn’t allow, or why AI wasn’t reliable – rather than investigating what might be possible. In a real sense, existing technology mindsets are not always well-suited to the level of agility and flexibility that AI strategies need.

And while security, data and legal concerns are legitimate, this is a different thing to the task of determining who should address them. In these cases, there weren’t clear responsibilities on who should figure out how to make things work, leaving mostly business teams frustrated and stuck with old ways of working, even if they saw an opportunity to do something differently.

On the opposite side, it’s common to see technology teams scope projects in very strict, traditional lanes and for them to overlook the opportunity for a more radical rethink. Another team I sat in with needed to calculate the energy costs and carbon footprint of hundreds of retail stores worldwide, drawing on data aggregated from hundreds of energy suppliers, in multiple languages, in multiple formats. The data team proposed a three-month custom coding project as a solution. When challenged to explore whether AI could handle it instead, they found a path that took a few days to build successfully.

Just as trust in AI is hard-won for end users, the same trend applies to development and implementation teams. Sometimes, it feels that those that were closer to technology, those who have deeper expertise and knowledge on how tech works, were almost the most skeptical and the hardest to convince about what AI could potentially do.

But then the same data-first way of thinking can create misalignment on the nature of work in other areas. 

Take the line plan as an example: creating the seasonal forecast and basing a collection plan on that forecast is one of the most complex jobs in fashion, bringing together data science, commercial objectives, and creative judgment. 

A merchandiser brings together internal sales data, wholesale partner insights, historical performance, and trend insights that suggest where the market is heading. Some of that data lives in databases, and AI has already been helping teams crunch the numbers and provide insights, but that’s still not enough in this case to plug things together the way they need to be connected. 

There is also a layer of what I’d call spatial context that AI cannot access, no matter how well-integrated it is with your technology estate: the strength of a relationship with a key wholesale account, the energy in the showroom during sales appointments, the direction the CEO signalled in a boardroom conversation, the feeling from a runway show about what will actually land next season.

None of that is in a database, and none of it can be rolled up in the way we think about other variables, but nevertheless there is a recurring expectation that AI can synthesize all of these inputs and create a reliable forecast just because it can do the same with other sets of non-normalised data. 

Getting the forecast wrong can have massive implications, so it’s not an easy job for merchandisers and buyers to trust in the AI tools – not because they’re resistant to technology, but because the tool is being asked to do something it structurally cannot do, even if, to the eyes of technology teams, that all looks like information that can be centralised and analysed.

Data scientists can build models that look credible precisely because they’re grounded entirely in measurable data. But data doesn’t capture what was said in the showroom. It doesn’t account for the physical product, or the conversation with the designer, or the instinct built over years of doing this job. 

This is what makes AI different from every technology that came before it. Earlier software took over repetitive, operational tasks in a way that made sense to the people doing them (who largely did want to shed the non-value-add parts of their jobs) and also made sense to technology teams who instinctively understood the blocks they were built from. 

AI goes deeper, touching the thinking, the judgment, the creative decisions that people have spent years building their careers on. When you sell that as a way to make everyone more efficient, you’re not just missing the point, you’re guaranteeing the pushback from those who mastered the craft for so many years, whether that craft is merchandising or data science.

Piece 3: you’ll be surprised at the people who move first.

Not all adoption comes from the top down. Some of the most significant progress I’ve seen started with a single person making a decision to try something on their own, and then making a business case by walking the resulting benefits upwards.

In early 2023, a menswear designer at G-Star started experimenting with Midjourney. He first kept his work to himself. When he started sharing it (mood boards, garment variations, fabric concepts) the reactions were split. Some colleagues found it off-brand and offputting. Others thought it was taking the brand somewhere genuinely new. Some people were emotionally outraged by letting AI take on design. 

I know that same spread of reactions has played out within other brands. Luckily, in this case, there was support to keep on exploring, but that same support isn’t guaranteed unless – as I’ve said above – you work to secure it.

Within a year, this person was hosting workshops, recommending tools, and helping other designers get started with AI – learning how to prompt. Most importantly, he made it feel safe for others to try. Not because of a business mandate, but because he was a designer, someone who understood the craft and had earned the trust of the team. When he said AI could amplify their work rather than replace it, people believed him.

There’s a common assumption in management that anyone can use these tools with a simple prompt, and that this means they can dictate that everyone move in unison to use them. The creatives who invest time in learning how to direct and apply them are developing a real skill, and it’s not something that can be instantly transferred to anyone else – either in spirit or as a practical way of working. A designer who understands both the craftsmanship of design and how to work with AI, and is moving early of their own volition, is not more replaceable but more valuable.

Piece 4: what the AI lead role actually requires. 

The bulk of your work, as an AI lead, will be solving people problems, not technology problems.

Many employees, even the ones reading reports like this one, are quietly uneasy. Not in a way they’ll say in a town hall, but in the sense that they can’t grasp how AI can actually be relevant for their type of work, but they feel like they’ll be forced to use it anyway. 

For visual-related work it feels obvious what AI can do. It’s impossible to ignore the generated lookbooks, campaigns, PDP content and so on that we’re surrounded with. But people that work heavily with PLM systems, Excel files, dashboards, etc (especially seasoned professionals that have been with the organisations for years) feel threatened by the thought that they might lack the technical knowledge they need to work with AI, or they might find prompting too complicated. That subtle, maybe involuntary, resistance is the real obstacle between a fashion brand and meaningful AI adoption, and it’s going to feel different to both teams and leaders to any other technology initiative they’ve been through before.

What moves this all forwards, in my experience, is empathy. As a leader, you need a clear message about what’s actually in it for each person when you’re proposing a transformation that makes people uncomfortable, for a range of different reasons. You need to make a genuine effort to separate the positive message that “your most arduous and non-rewarding tasks will change” from what people hear, which is “you are being replaced.” 

And close collaboration with management and HR teams will be mandatory to make sure an organisational push for efficiency doesn’t end up falling on individuals as existential pressure, and instead it winds up feeling like an initiative that will benefit everyone.

None of this appears in the AI Lead job descriptions. What’s written on paper is a set of asks for Python skills and familiarity with large language models. The actual role demands: the ability to present clearly to a skeptical audience; to sit down with a merchandiser or a designer and understand their process before suggesting where AI might fit; and to hold a meeting between an IT team and a business team and translate in both directions.

The best AI Leads, then, are connectors. They translate business needs into possible solutions. They find where new AI tools or existing solutions create shared value across departments. They give technical teams the business context they’re missing and give business teams enough fluency to ask better questions. They are empathetic and have patience to see both sides of the arguments and bring people together to collaborate. 

That profile rarely comes from a pure data science background. Some of the people best suited to this work come from HR or marketing, even some senior project managers or operations leaders. These are disciplines where the core skill is understanding people, communicating across differences and keeping programs moving forward.

The question fashion brands should be asking, as a result, isn’t who understands AI the best. It’s who understands the business well enough to connect what this pioneering technology can do with a clear brief of what we actually need it to do.