If AI saves your design team days of work, where do those days show up in the launch date? The time saved is real. Briefs become visuals in minutes, specifications write themselves and translation errors disappear. A large share of the back-and-forth that used to fill a development cycle simply goes away. The launch date rarely moves by the same amount. The time saved upstream runs into a constraint that no design tool can touch.
Testing the promise of AI in product development
A new version of the same promise appears every few months: AI will cut product development from months to days. It is worth testing against how a product actually gets made, and against the data that exists. The promise holds for some products and fails for others, and the reason why says a lot about how this industry works.
Start with a simple question. When a first sample takes several weeks, what happens during those weeks?
Following one sample
Take a garment from brief to first sample. The designer explains the idea. Someone turns it into a tech pack. A pattern maker reads it, often in another language, and makes their own interpretation of words like relaxed or structured. Fabric is requested from a mill. The sample room fits the job between other jobs. The sample comes back wrong in two places, and the loop starts again.
Put a stopwatch on the people actually working on it and the hours are modest. Most of the calendar is waiting: for the slot, for the fabric, for the reply, for the approval. Anyone who has set up supplier networks or run a sample room will recognise this.
Very little public data separates working time from waiting time, and few businesses track the difference.
So product development contains three kinds of time: understanding, physical work and waiting. Some of the waiting is unavoidable. A new fabric needs testing, a sample room has limited capacity, a factory has other orders. The rest comes from poor coordination. The real question is which of these AI can change.
What AI changes
Mainly understanding. Misreadings happen because design intent travels as words and flat drawings, and everyone along the chain translates them into their own references. Image generation and 3D simulation turn a sketch or a brief into a visual everyone can see, with its proportion, volume and drape, so a word like relaxed stops being open to interpretation. AI-assisted tech packs and translation then turn that picture into measurements and construction notes in the factory’s own language, and flag inconsistencies along the way.
People who know how the garment is made still have to check the result, but designer, developer and factory now start from the same picture.
Vendors of these tools report one or two sample rounds per style instead of three or four. These figures come from companies selling production AI, not from industry averages, but the direction makes sense. Fewer misunderstandings, fewer wrong samples.
For some products the effect is dramatic. Think of merchandise: an established pattern, a fabric already on the shelf, a supplier who has made the same T-shirt hundreds of times. Nothing is left to discover. Development comes down to artwork, placement, colour and a size ratio, and AI can take it from brief to production-ready specification in hours.
There is one condition, and it matters later in this story. A virtual sample is only as accurate as the patterns and material data behind it.
Where it stops working
Now change one variable. Same garment, new material. Suddenly there is a lot to discover. The fabric has to be sourced, usually with a minimum order and a mill lead time. It has to be tested for shrinkage, and other parameters. It has to behave in construction the way the designer imagined, and on the first attempt it often does not.
AI can estimate how a material will behave, but only from data someone has already measured. Until a new fabric has been tested, its digital version is a guess.
The same goes for a new pattern or an unfamiliar construction. The uncertainty gets resolved in the sample room, with scissors and a machine, not on a screen.
The pattern is consistent: the gain from AI shrinks as novelty grows. The more a product reuses what is already known, the faster AI makes it. The more genuinely new it is, the more it depends on physical testing, skilled hands and a free slot in the sample room.
Who runs the queue
To understand the queue, look at who runs it. According to EURATEX, 99.7% of textile and clothing companies in the EU are micro, small or medium-sized enterprises. These are the workshops behind many of the products the industry admires most.
A business of that size has one sample room, a pattern maker who also solves production problems, and an owner who quotes, sells and chases payments at the same time. It works for many brands at once. Bulk orders pay the bills and samples rarely do, so samples wait their turn. As simple as that.
This is where the real effect of AI on development shows. The bottleneck was always there, hidden behind slow emails and misread tech packs. Clear away the communication friction and the sample room’s capacity becomes the limit much sooner. Faster briefs arrive at the same door.
These businesses are under pressure too. EURATEX recorded a third consecutive year of falling production, turnover and employment in 2025, and exhibitors at Mare di Moda in October talked about yarn prices close to double. The brands sending them work are hardly models of speed either. A 2025 survey by Bamboo Rose and The Interline found that 66% of brands and retailers still rely on spreadsheets, email and messaging apps. Slow replies and approvals that wait for next week’s meeting are part of the queue as well.
Why the tools do not reach them
For AI to shorten this part of the process, it needs something to work with: structured material data, live capacity, digital patterns, systems that talk to each other, and people to maintain them. In many workshops none of this exists. Fabric availability lives in a phone call. Capacity lives in the owner’s head. The tech pack arrives as a PDF and gets printed.
The usual answer is that suppliers should digitise. The logic is understandable, but look at it from the other side of the table. A workshop of eight people serving a dozen brands would have to learn a dozen portals and formats, pay for it from thin margins, in a shrinking market, with no promise of more orders in return. Declining that is a sound business decision.
At this point the issue stops looking like a technology gap and starts looking like an investment problem. The value of AI is capped by the least digital link in the chain, and the cost of improving that link falls on the businesses least able to carry it.
What it means for productivity
The promise is half true. AI has made the brief far better and simple, repeatable products genuinely fast. For anything new, the calendar is still set by a workshop with limited capacity, thin margins and no reason to invest on its own.
That is also why the usual fear about productivity looks different here. Elsewhere, a jump in productivity raises fears of job losses.
Fashion manufacturing has long suffered from too little digitalisation and too few skilled people.
EURATEX reports persistent skills shortages alongside falling employment, and Eurofound research shows that small businesses often digitalise precisely because they cannot find staff. A workshop short of pattern makers needs the people it has to stop chasing emails and retyping specifications. Here, productivity growth protects jobs, makes European manufacturing more competitive and strengthens the case for keeping production close to the brands it serves.
The open question is who builds the infrastructure. A workshop of eight people cannot hire a data team, and IT talent is scarce across the whole sector. Shared platforms between brands and suppliers, such as Wonder Industries, can carry that digital load for many small businesses at once: one environment instead of a dozen brand portals, with the cost spread across everyone in it.
They will not shorten the physical work of developing new materials and patterns. They can remove much of the waiting caused by poor information, the part of the queue the industry has tolerated longest.
For anyone responsible for a range, that changes the questions worth asking. Which part of the range is configured and which part is new, and should both be expected to move at the same speed? How much of the lead time is necessary work and how much is avoidable waiting, and whose waiting is it? Is AI being used to send fewer, better requests to suppliers, or simply more of them? And is the business prepared to share the cost of the infrastructure it expects suppliers to have, through common formats, volume commitments or paid sample slots?
Product development has always been a relationship between people with ideas and people with machines and limits. AI has strengthened the first side and made the conversation between them far more efficient. The second side will move when brands and suppliers invest in the infrastructure that connects them, and the brands that understand this will get far more from their AI budgets than those waiting for software to do the job.