How fashion is using generative AI in-house

This article on generative AI is part of our Vogue Business membership package. To enjoy unlimited access to our weekly Technology Edit, which contains Member-only reporting and analysis and our NFT Tracker, sign up for membership here. Virtual online shopping, in-store assistants and in-house data analysis: fashion brands are finding use cases for generative AI that could change the way they work. Generative AI — a category of the technology that can generate new images or text — has gone mainstream thanks to companies like ChatGPT that are available for everyday use. Brands and developers in response have been racing to test consumer-facing uses. In recent months, Kering has tested a chatbot that offers personalised purchase recommendations, H&M has introduced the option for customers to design their own wearable artwork and Prada Beauty has unveiled an AI-generated ad campaign. Both Amazon, with Carrera Eyewear, and Meta, with Ray-Ban, have announced glasses equipped with personal AI assistants. In the next three to five years, generative AI could add up to $275 billion to the apparel, fashion and luxury sectors’ operating profits, according to consultancy McKinsey. AI enhancements by luxury fashion brands specifically stand to boost consumer spend, according to a Vogue Business and Google study of luxury fashion consumers in the US, UK, Italy and France. Seventy-five per cent of the nearly 3,000 consumers surveyed think they will buy more from fashion brands that use AI within the next three years, with personalised deals and recommendations being particularly compelling. Behind the scenes, developers are seeing an opportunity that is just as transformative. The hope is that generative AI tools can help employees become more efficient and spend more time focusing on uniquely human tasks, while generative AI can do the heavy lifting on more menial chores. Most brands have tools in place to improve customer-facing experiences, “but the employee-facing side of productivity is generally under-appreciated”, says Nitin Mangtani, founder and CEO of point-of-sale tech company PredictSpring, which just introduced a new generative AI tool for store associates. As with most new technologies, there is a concern that it could eliminate jobs. This could be true in the short term, with some tasks becoming less dependent on humans. However, developers say that none of these news tools are positioned to totally replace the abilities of humans; rather, they still need to be validated and refined by humans. “Eventually, jobs will change because some things are much easier to do with AI than before. Brands will be reassessing jobs to be aligned with AI initiatives,” says Parham Aarabi, CEO of AI startup Pre, an AI startup that simulates how people use e-commerce sites and apps (Aarabi is also the founder of augmented reality tech company Modiface). “The jobs now are much more impactful.” Assisting in-store associates Enhanced AI assistants, such as Amazon’s Alexa and Meta AI, are designed to give consumers an easy way to access information. Now, they are being built for use by store employees. Walmart has created an AI tool called Ask Sam, which associates can use to verbally find items, store maps, prices and more. PredictSpring, used by Movado and Steve Madden, has built an AI voice assistant made specifically for the retail industry, training it on the tasks and workflows used by retail associates. In-store employees can communicate with PredictSpring’s existing tech to, for example, access customers profiles, view product availability (“show me leather jackets”), view incoming click-and-collect orders and check store analytics. They can also use it to initiate tasks that would normally require associates to manually go through multiple steps, such as “start a cycle count” (meaning counting in-store inventory) or initiating a return, says PredictSpring’s Mangtani. Because voice commands are inherently designed to feel natural, this also reduces the costs and times associated with training new employees in the stores. The time savings can ultimately benefit the customer as well, Mangtani says. Historically, store associate tasks were largely relegated to a series of keystrokes, he says, and that hasn’t changed much over time — which is why sometimes it can feel like the checkout process is slow. But, he says, “the people aren’t bad — the tech is bad. If every click takes 15 to 30 seconds, you are killing the usability.” The new PredictSpring technology was built to respond in 200 milliseconds. The tech has limitations. Retailers still need a system in place to build the data reports from which the AI can pull, and the results aren’t necessarily meant to be “fully deterministic”, Mangtani says, meaning that retailers should expect a small margin of error — but it will improve over time as brands use it and the AI learns. There will be immediate results, however, in terms of freeing up employees to perform the services that bring people into the store in the first place. Virtual shoppers to test e-commerce functionality A/B testing on websites enables brands to test how features such as promotions, pricing and product descriptions perform. The process requires considerable time, skill and budget to conduct and analyse — on top of the fact that brands must push various scenarios live, even if they aren’t ideal. Now, AI is able to create “virtual shoppers” to test the design of websites before any changes are made to the customer-facing experience. Swiss luxury brand Bally worked with Pre, which trained its technology on the brand’s own data on how Bally customers shop, combined with data on the approximately 100 other brands that have worked with Pre. The virtual shoppers predict how a person would behave, and the pages they are “shopping” are based on screenshots and mockups that are then “decomposed” into every element on the page, including the buttons, images, widgets and size selection tool. Pre measures likelihood of interaction, user intent and likelihood of adding to a cart or converting into a sale, and can compare a brand’s site to competitors, using only screenshots. Retailers can run simulations that are a million times faster than a human simulator and change details to test how they would change the results. Using this process, Bally discovered that the product-to-checkout experience could be improved, the load time could be faster, the site visuals could be enhanced by thumbnails and that product descriptions could be better placed. The average client can see a cumulative conversion improvement estimate of 50 to 100 per cent, Pre’s Aarabi says. Surprisingly, tools that recommend sizes can either help or hinder conversions, Pre has found, so it’s more important for brands to find the right tool than to just add something in. Bally observed significant disengagement with its size finder after 15 seconds, so it ultimately created a new size translation tool. Similar to other AI tools, the more that brands use the technology, the larger the database and the more accurate the data will be, Aarabi says. The company charges based on how often the brand uses Pre’s tech, which can be every time they are considering a change to the website. He says that in terms of jobs, the wide array of AI tools available is likely to cause company structures to change dramatically, with people being assigned to work on different areas of the business. And teams still need to decide if they ultimately want to make the recommended changes; just because a machine says, for example, a size-finder tool might help, that might not be feasible for the brand, Aarabi says. Simplifying access to internal data ChatGPT enables people to type in complex queries to get digestible results based on the information and knowledge on the internet. Inspired by this easy-to-use format, Spanish fashion group Mango created a similar internal tool for its employees and partners. Called ‘Lisa’, it is a conversational generative AI platform that was built using both private and open-source models trained specifically for the company over a period of nine months. Lisa is designed to enable improvements on processes that span from the development of collections to after-sales services; Mango’s “main information assets” cover customers, stores, stock and product. Jordi Álex, Mango’s director of technology, data, privacy and security, calls generative AI an “extended intelligence”, meaning “a technology that will act as a co-pilot for our employees and stakeholders and that will help us extend our capacities, because technology will either make us more human or will be of no use”. Mango has more than 15 AI platforms focusing on areas such as pricing and product personalisation. Other companies might soon follow. Already, Shopify provides tools for merchants that lets them use AI to ask for insights on their ecommerce sites, then change them based on natural language commands. Both Hugo Boss and Tapestry are focusing on data dashboards for internal use; these dashboards could add an AI assistant to make accessing insights even more intuitive. Hugo Boss is investing millions in building “data architecture infrastructure” and has created a dashboard to support decisions related to products, shopping behaviours, marketing spend and prices. A central team at Tapestry has created a user-friendly, intuitive platform for its brands, including Coach, Versace and Michael Kors, to make decisions in product design, communications and experiences; Coach global CMO and North America president Sandeep Seth recently told Vogue Business that adding on an AI intelligence layer would be a next intriguing step. Creative shortcuts — not solutions While practical tools are positioned to have a more immediate impact to the bottom line, creative teams are also experimenting with how AI can aid in creativity, from making it easier to visualise ideas to generating inspiration for runway collections. Mango’s Álex says that AI has helped the company visualise different design concepts, including prints and textures, and created settings for the photography studio. An AI platform of images, called ‘Inspire’, has led to more than 20 garments (introduced under the mens, kids and teen lines). Human curation, across the board, is positioned to be central to the conversation going forward. Collina Strada founder and designer Hillary Taymour fed all previous collections into an AI engine to create new looks for the Spring/Summer 2024 runway show — but still ultimately personally modified and selected the final looks. Retailer Revolve, which is slated to physically create winning looks from the recent AI Fashion Week, relied on an expert selection committee, in addition to the discretion of co-founder and co-CEO Michael Mente, to decide what would ultimately be produced. A project between retailer Zalando and Google to generate avant-garde designs using trend data, fashion experts and personality profiles creating interesting designs — but just because they are new ideas, doesn’t mean they are wearable solutions. Mango’s Álex says that inspirational design is one of the most complex applications and the one in which the human factor is the most important, because the algorithms are not specialised in fashion. “The talent, creativity, experience, instinct and passion of our designers is, without doubt, superior to that of the platform,” Álex says, adding that humans are still “an indispensable asset for the company, playing a key role by providing a response in areas the technology cannot reach”. To receive the Vogue Business newsletter, sign up here. Comments, questions or feedback? Email us at feedback@voguebusiness.com. More on this topic: Mmerch is a new fashion brand rethinking production using AI With generative AI, luxury hopes to finally crack digital clienteling How Estée Lauder is using generative AI for copywriting and customer service

Make better business decisions

Make better business decisions