A Very Efficient Inequality Machine
WORDS BY Mark Bage, NOT Founder
AI, women, and the work fashion runs on.
It started as a question across the table. Nickii Gray, one of our non-executive directors, put it plainly: AI might be bad for women. Not bad for women in the abstract, bad for women's jobs, women's pay, women's careers. Worth saying who was asking. Nickii has thrived as a woman in business and says herself she has never felt the obstacle; this was not a familiar case made from bitter experience. She wasnât describing her own career. She was describing what she could see coming for other peopleâs. I disagreed by instinct and went away to do the reading, fully expecting to come back with the numbers that proved her wrong. I didn't find them.
I should say where I stand before we get into it. Sarah Coggles, the business I grew up in, was founded by my mother, Victoria Bage, in 1974. She built it and she ran it, and from the 90s I ran it with her; most of what I know about this industry I learnt inside her business. My sister worked there too. I grew up under strong female leadership and it never occurred to me that this was unusual. So none of what follows is a question about what women can do. It is a question about where a particular technology lands, and on whose desk.
AI does not target women. The labour market is not a conspiracy and neither is the technology. But generative AI happens to be very good at a particular basket of tasks: drafting, summarising, scheduling, reporting, customer responses, basic analysis. Those tasks sit unevenly across the workforce, and in fashion they sit mostly with women. You can hold any view you like about why the distribution looks the way it does, and reasonable people hold different ones. The exposure is arithmetic either way. The wave lands where the tasks are.
"Whether her role gets quietly shrunk is a decision someone above her will make, probably in a meeting she is not in."
The Numbers
The International Labour Organisation put a number on this in May 2025: in high-income countries, 9.6% of jobs held by women sit in the highest exposure category for generative AI, against 3.5% of jobs held by men. Nearly three times. The ILO ran the numbers again in March 2026 and the picture held: female-dominated occupations are almost twice as likely to be exposed as male-dominated ones. I didnât need the spreadsheet to picture it. Run down the roles at Coggles in its e-commerce years, the copy desk, the customer service team, the studio coordinators, the warehouse logistics, and most of those chairs were filled by women doing exactly the work these models now do fastest.
The reason is unglamorous: exposure follows the task mix. Generative AI is a language machine. It drafts, summarises, schedules and replies, which is the daily substance of administrative and coordination work. Plenty of men do that work too; nobodyâs inbox is safe. But the concentrations differ: women are overrepresented in the language-heavy roles, while the occupations men dominate skew physical, driving, building, fixing, moving. Robots will get there eventually, on a much slower timetable. The honest exception is software engineering, male-dominated and heavily exposed, worth conceding before anyone mistakes this for a tidy story.
Exposure doesn't mean the job disappears, and the ILO is careful to say so. In most of these roles the tools will eat tasks rather than positions. But if employers reach for AI as a headcount instrument rather than a capability one, and plenty will, the cuts land unevenly, and there are early signs of exactly that. Women's share of European tech roles has slipped three points since 2023, with a lot of the redundancy falling on the operational and coordination jobs. Employment lawyers tend to notice these things before the rest of us: Lewis Silkin, one of the UKâs biggest employment firms, has already asked in print whether AI opportunity inequality becomes the next gender pay gap.
Fashion Is A Concentrated Case
Fashion runs on women. They are the majority of fashion school graduates and the majority of the workforce, and in the supply chain the concentration is stronger still: of roughly 94 million garment workers worldwide, around 60% are women, rising to 80% in some manufacturing regions. The top of the industry looks different, with the brands run by women across the biggest luxury groups still a short list: Saint Laurent, Dior, CĂ©line, Loewe, and not many more. But I am not making a discrimination argument here. Ambition, choices and career structures all play their part, and that debate is live and worth having honestly somewhere else. My point is narrower. Look at where fashion's women actually work: e-commerce trading and content operations, marketing coordination, customer service, studio production and comms, buying and merchandising admin. Write down the tasks generative AI performs best and you have just listed those departmentsâ job descriptions.
Take one desk. Among our own e-commerce clients, many of them small to mid-size brands with all-women teams, we see the same role again and again. The e-commerce manager spends their day writing and uploading product copy, resizing and tagging imagery, building email campaigns, updating the promotions calendar, answering the customer queries, and pulling the Monday trading report. Six tasks. A capable generative model already produces credible first drafts of five. She is not about to vanish, but her job description is being rewritten by a technology nobody consulted her about, and whether her role gets redesigned around the new mix or quietly shrunk is a decision someone above her will make this budget cycle, probably in a meeting she is not in.
And that is only head office. The same wave that writes product copy is learning to cut, sew, pick and pack, which means the most female tier of the entire industry, the manufacturing floor, faces an older kind of automation now accelerating. Any honest account of AI and fashion has to include the 42 million women making the clothes, not only the people marketing them. Most accounts don't.
"Bias with a dashboard is still bias, and an unaudited screening tool is a bias you have chosen not to know about."
Who Gets The Upside
Displacement is half the story. The other half is who captures the productivity gain, because AI fluency is turning into a career accelerator and adoption has not been even.
A UK study published in January 2026, titled with unusual bluntness "Women Worry, Men Adopt", found women measurably less likely to use generative AI at work, partly because they weigh the societal risks more heavily. Which, it should be said, is not an irrational position; some of the worry is well founded, and the trust numbers suggest the providers have earned it. In one survey only 18% of women using generative AI had high trust that providers would keep their data secure. For men it was 31%. I watch the same gap in meetings now. Ask a room who is using these tools every day and the first hands up are rarely the people whose jobs the ILO says are most exposed. Men, though, are definitely carrying the flag for the new technology, and so am I. Iâm a massive fan; it is truly game-changing. The productivity is astounding, up there with steam, the car and the silicon chip.
Nickiiâs sharper point is that the stakes run past salaries. These tools learn from the people who use them; every prompt, correction and preference feeds back into how they behave. If adoption skews male, the data, opinion and emotional register the models absorb skew male with it. âIf these tools are becoming a brain for the world,â as she puts it, âthen that brain could be more masculine by default, and potentially regress us from an equality perspective.â The ILOâs researchers put it more formally: underrepresentation of women in the development and adoption of AI increases the risk of gender-biased technologies. Which turns adoption from a career question into a question of participation. Sitting this technology out doesnât keep it neutral. It hands the pen to whoever stayed in the room.
The encouraging part: this gap is closing fast. Deloitte found women's adoption tripling year on year, outpacing men's growth, so it looks like a lag rather than a ceiling. But a lag at the exact moment fluency starts deciding promotions is still expensive.
Bias With A Dashboard
Then there is hiring. If companies use AI to screen CVs, rank candidates and write job ads, whatever bias sits in the model gets scaled and laundered through software. The audit literature here is genuinely messy. One 2025 audit of recruitment-oriented language models found models favouring men for higher-wage roles and echoing old stereotypes in job descriptions. Another ran 361,000 synthetic CVs through five leading models and found the opposite lean: systematic favouring of female candidates, with black male applicants disadvantaged most. Nobody should be comfortable, whichever way their preferred tool happens to lean this quarter.
A tempting objection: HR is one of the most female-dominated professions in the economy, so surely the tools its people choose would not disadvantage women. But the bias does not live in the buyer. It lives in the training data. Amazon's infamous recruiting engine taught itself to downgrade CVs containing the word "women's" because it learnt from a decade of historic hiring patterns; nobody in that room chose it, and the demographics of the team running the tool made no difference at all. Bias with a dashboard is still bias, and an unaudited screening tool is a bias you have chosen not to know about. I have spent a few decades on the buying side of software for fashion businesses, and I can tell you no one in the room ever asked how the tool had been tested. Regulators have noticed. There is no UK AI act, but there doesnât need to be one: the Equality Act already makes an employer liable for a discriminatory outcome whether a human or a tool produced it, and the ICO reported this spring that most UK employers using AI screening believe the tool merely assists when it is in fact deciding. New York and Colorado have gone as far as mandating audits. Fashion HR teams buying screening software should be asking vendors for audit results now rather than after the first tribunal.
Not Destiny
Used to augment, AI strips the admin load out of exposed roles, speeds up learning, gives small teams the output of large ones, and opens technical capability to people who never learnt to code. We watch it happen weekly in our own studio and in our clients' teams. The e-commerce manager who now does in an afternoon what used to take a week is not less valuable. She is more, and any employer who cannot see that deserves the competitors it will create.
For a fashion business the to-do list is not complicated. Train the most exposed teams first, which in this industry means e-commerce, marketing, customer service and studio operations, not the leadership layer who least need it and most often get it. Redesign exposed roles around the new task mix instead of quietly shrinking them. Audit any hiring tool before it touches a CV. Treat AI literacy as core to every commercial and creative role. And the tools are being taught now, by whoever turns up to teach them. Turn up.
Used badly, AI becomes a very efficient inequality machine. Used well, it flattens barriers that training budgets and good intentions never quite managed to move.
I have seen this industry meet a technology wave before. In 2003 we put a fashion business on the internet and were told it would cheapen the brand and empty the shops. What actually happened was quieter and stranger: the work moved. Jobs appeared that had no names yet, old ones changed shape, customers were found globally, and the businesses that came out ahead were the ones that reorganised early instead of dismissing it and hiding from the online revolution.
I donât expect the same story this time. Business learnt it was too slow last time, the bigger businesses especially, and they wonât make that mistake again. They will be ruthlessly efficient, and change will come at several times the speed. Within two or three years AI fluency will be assumed in fashion the way spreadsheet, or e-commerce literacy is assumed now, invisible as a skill because everyone has it. The gender gap that matters is being set today: who got trained, whose role was redesigned rather than quietly shrunk, and who managed to hold on and learn at speed while the technology moved faster than the workforce.
Women are disadvantaged by that speed, by a tool that wasn't designed to suit men but is being allowed to. The industry has perhaps eighteen months of choices before those answers harden. That is not long. It is also plenty.
So was Nickii right? More right than I wanted her to be. The risk is real, it is measurable, and in fashion it is concentrated. But it is a risk, not a verdict, and the industry that figured out how to sell a point of view to the entire world ought to be able to manage a technology transition without leaving half its workforce behind.