Sign up to Interline Insiders to receive our weekly analysis in your inbox every Friday.

Key Takeaways

  • The USA remains a primary force in fashion technology development and deployment, but a digital capability gap is emerging in the domestic apparel manufacturing industry.
  • Traditional manufacturing businesses in the US are resistant to adopting modern practices and solutions due to financial and cultural reasons, which is likely to translate into a lack of both capacity and specialisation – something that microfactories and digital production hardware aren’t yet ready to compensate for at scale.
  • Despite this, the promise of being able to transform sourcing and production to meet both efficiency and sustainability targets is attracting a new generation of disruptors – people driven by the idea of reshaping both onshore and offshore manufacturing, rather than stepping into defined roles.
  • Crowd-sourced tests of cloud-based, community AI models in stylist and design roles are demonstrating audience’s willingness to engage with AI for deeply personal purposes. But privacy pushback is likely as on-device models mature, and as consumers begin to test the waters of more comprehensive product and outfit recommendations catalysed by the upcoming overhaul of search.
  • At the same time, another trending test demonstrates that the creative powers of AI models (rather than just their ability to remix) are potentially being overestimated.

The forces pulling talent away from, and towards, the supply chain.

It’s no exaggeration to say that the USA has led – and continues to lead – the technology conversation in quite a few areas. And while Europe and Asia are now home to equally active fashion technology scenes – not just established players, but a new raft of disrupters –  the US continues to be thought of as one of the more mature markets for fashion technology development and deployment.

Or at least that’s the reputation the country has from the brand and retail perspective, but a first-hand account of the decline of America’s apparel manufacturing industry – and the digital capability gap this is creating – caught our eye this week.

The account is worth reading in full, but the essence is this: while manufacturing businesses overseas have been relatively successful in modernising (at least those of a certain scale) there’s a large cohort of domestic, American, manufacturers who are so entrenched in their historic ways of working that, as fashion as a whole progresses further down the road of digital transformation, these producers will be left behind.

There is, as the article describes, an element of choice in this. Manufacturers who refuse to adopt modern practices and solutions are doing so at least partially because of simple inertia. But the problem is more multi-pronged than that, and it has more in the way of implications for fashion’s relationships with its supply chain partners than simple stubbornness would account for.

For one, manufacturing facilities are often family-run companies and those who are on the way out are resistant to embracing new technologies from a financial perspective as well as a cultural perspective. On the former point: wages, insurance, and taxes are a significant cost to manufacturers of a certain size, which make quite narrow margins on orders as a result. With a large portion of manufacturer’s dollars being absorbed by those costs, it leaves little room for purchasing any kind of new technology in the form of machinery or software that might streamline certain processes and contribute to modernisation.

An added complication is the fact that while the digitisation of patternmaking is already proven and feasible at relatively low cost, the same cannot be said for the assembly stages of apparel production. While a lot of investment has been expended on digital means of manufacturing (plotting, spreading, cutting, printing) these remain expensive capital investments. And for the same reason that long-running manufacturing businesses aren’t buying software, they also aren’t buying connected hardware.

All of which means that the efficiency and cost savings we talk about so often in the fashion technology space are essentially outside the grasp of these businesses. And over time this will translate into not just a reduction in the competitiveness of domestic producers, but their actual capabilities; it won’t just be the case that fashion cannot find affordable Tier 1 capacity at home… it will also struggle to find specialisation.

The Interline has recently covered the reality of the “microfactory” model, looking critically at its ability to sustain and scale. And while that article concludes that the potential for a distributed, regionalised, digital, on-demand production network is real – that vision is not ready to be realised at scale. So it’s unlikely that an entirely new model will step in in time to stop the last vestiges of production expertise drying up.

(We should note that these forces apply less in luxury and other sectors where craft and tradition remain currencies in their own right. Although many of these businesses are also much further ahead in digitisation than we might think.)

But just because traditional manufacturing as an industry could be on a downward trajectory, does that mean that people are migrating away from manufacturing-adjacent roles? That picture is a bit more complicated.

In a direct sense it’s likely true: with the lack of investment into new technologies from traditional producers, those who are being primed for taking over the business (these are often multi-generational concerns) have reduced interest in doing so. And this has not been helped by the sentiment that working in the manufacturing side of the fashion industry is less glamorous than working in the creative side – even if manufacturing expertise is arguably more essential now than ever.

In an indirect sense, what we’re more likely to be seeing is a steady swing towards roles that are occupied with orchestrating and transforming manufacturing, rather than simply slotting into its long-standing personas.

We saw this week, for example, that companies are now placing significant emphasis on hiring for the supply chain. These are arguably among the most important jobs post-pandemic, considering just how fundamentally fashion needs to change at the foundational level to solve the challenges of overproduction, sustainability, and social responsibility. The massive scale and depth of fashion’s problems necessitates far greater quantities of skilled people wanting to solve them.

But that new generation of talent that cares about the mechanics of production enough to find ways to redesign them – not join them – is definitely a work in progress. While it’s clear that people across all age ranges care deeply enough about fashion’s heavy environmental and ethical footprint to want to reduce it, reshaping the way the industry thinks about production is not a challenge that will be addressed quickly. And by the time it is tackled – the urgency of things may have moved the goalposts even further.

Testing the limits and locations of AI.

This week, Apple introduced new software features for cognitive, speech, and vision accessibility that will be made available later this year. One of these is for people who may lose their ability to speak, and allows them to create “a synthesised voice that sounds like them.”

Fans of Apple’s on-device approach to machine learning will be pleased to see that this push into assistive voice synthesis does not look as though it will come at the expense of privacy. While personal voice models will likely be synced via iCloud – encrypted, presumably – this is not a direct analogue to the cloud-first models that characterise Midjourney, ChatGPT, Bard, and other AI solutions.

This may not sound like it has a lot to do with fashion, but consider this: the same forces that lead people to not want voice training data to be part of a community dataset are going to feel the same way about having photographs of themselves used in that capacity – which is becoming a very real prospect based on people’s apparent desire to have ChatGPT recommend outfits to them.

To be clear, at this point there are only narrow applications of large language models in product recommendations, and no universal AI stylist (Amazon did try this somewhere back in the dark days of 2020, but the idea did not take flight). But all the same we’ve seen this week that people have been requesting ChatGPT play the role of stylist – with #AIFashion racking up over 14 million views so far on TikTok.

ChatGPT is, of course, a cloud-based model – it devours compute at massive scale – but it’s not unreasonable to consider a scenario where someone might want to use a machine learning tool as a personal stylist, but want to keep information as personal as possible for both privacy but also accuracy reasons. And it’s also reasonable to see that on-device models like the one published by Facebook will run well, locally, in the near future.

Imagine that if you are able to train your on-device AI tool with images of only your fashion choices, with no external interference, the result will be hyper-personalised recommendations for what to wear for any given occasion (provided you could also include a bit more information around the nature of the event) – especially if these recommendations tap into the huge overhaul that’s happening with web search at the moment.

But while it may be true that AI is capable of delivering serviceable recommendations at the style level, and assistance at the design level, another new trend that emerged this week (asking LLMs how many times the letter “E” occurs in the word “ketchup,” oddly) that suggested there is still something ineffable about creativity that it can’t yet capture.

Just like Google’s large language models (LLMs), which were the ones being tested as part of that trend, ChatGPT is very much a language tool rather than an intelligence platform. And this trend is a reminder that, while those two things sometimes align with frightening acuity to make chatbots seem intelligence, anthropomorphising AI is likely to be giving its current applications far too much credit for their creativity.

This is part of why, in The Interline’s coverage of AI fashion week, we found it surprising that fashion professionals were rushing so quickly to extend the definition of “designer” to include someone who generated a collection of clothing using an AI tool.

Fashion designers bring a unique blend of their subjective experience, artistic vision, human intuition, cultural understanding, and trend forecasting to their work. They consider various elements such as fabric selection, garment construction, functionality, and emotional appeal. Fashion design is not solely about creating aesthetically pleasing designs but also about understanding the needs and desires of consumers, creating cohesive collections, and capturing the essence of a brand. AI is, in a nutshell, capable of faking a lot of this, but as the simple “ketchup” test demonstrates, there is a great deal of complexity inside human creativity that we simply don’t understand well enough to distinguish properly from artificial facsimiles of it – until we get a very simply and stark example like this.

That being said, AI can be a valuable tool in the hands of a skilled fashion designer, or even just a person who needs some fresh ideas on what to wear, allowing for inspiration that may not have occurred before, and remixing existing ideas to generate some form of newness. As technology continues to evolve, the intersection of AI and fashion opens up intriguing possibilities, but it’s important to remember when we start to interrogate the roles of stylists, planners, designers, and other key fashion stakeholders: there’s a lot more to them than just the output on a screen.