AI in fashion buying is genuinely useful in five specific places and genuinely risky in three. I am not a technologist, and I do understand the buying process well enough to be clear about which is which.
Jump to:
AI in Fashion Sits in a Much Longer Pattern
This is not the first tool that was going to change everything.
Across my career I have gone from handwriting orders and communicating by fax, to ordering online, to tracking every part of the workflow in real time. AI adds another layer, and it is not a magic wand.
Trend research is the clearest example. Building a board once meant physically collecting material: magazines, tear sheets, cutting images, gathering anything that helped you visualise a direction. The internet made that dramatically faster, and something became obvious almost immediately. You could always tell when a mood board had been assembled at the last minute, because the images had no depth and formed no coherent story. The ones that worked were where someone had invested judgement in the selection and kept refining it until the board told the story without explanation.
AI now lets you generate images rather than only find them, which genuinely helps when visualising a direction that does not yet exist. The principle underneath has not moved. What you put in shapes what you get out.
Podcast Thrive in Fashion Buying & Merchandising
Episode 058, What AI Is Actually Doing to the Fashion Buyer’s Role (And How to Stay Ahead of It).
Every one of these tools works best in the hands of someone who can tell when an output does not look right, and that judgement comes from understanding the buying process rather than the software. The Thrive in Fashion Buying and Merchandising course builds that foundation.
Trend and Demand Forecasting
A buyer traditionally builds a plan from historical sales, trade shows, shop floor feedback and her own reading of the market.
All of it is valuable and all of it is limited. Historical data only tells you what you sold. It cannot tell you about the customer who came in, did not find what she wanted, and left. It cannot see what is building on social media three months before it reaches the high street.
That is where these tools are strong. They pull together ecommerce data, catalogue coverage, search behaviour and physical retail data at once, and find patterns you would not have time to. WGSN, which most buyers know as a trend platform, has built a buying tool combining social, search, catwalk and sentiment data into projections across key items, silhouettes and colours, with a calculator for the financial case behind a buy.
What this means practically is not that the tool decides for you. It means you walk into a range review with a broader and more current picture than you could have built from internal data alone, provided you bring enough understanding of your category, customer and supply base to know what to do with it.
Range Planning and Option Modelling
This is the most immediately practical application for anyone managing a category, and it applies to wholesale and private label equally.
Instead of rebuilding a spreadsheet every time you want to test a different approach, you can model the scenarios quickly and see the implications. What if I buy a hundred options instead of a hundred and twenty, but increase average quantity from five hundred to six hundred and fifty per style? What does that do to margin? What if I restructure the price points, and how does that affect my open to buy?
You arrive at a range that has genuinely been interrogated rather than a slight iteration of last season. The value is not that the tool decides better than you would. It is that you ask far more questions before you commit, and that means fewer surprises once you are trading.
Design and Product Development
Concept development is my favourite part of the process and one of the most time consuming, because traditionally every variation depends on a designer’s time. A different neckline, a different sleeve length, a different colourway all require manual work.
These tools can now generate multiple variations from a single brief in minutes. A team can explore twenty versions of a jacket silhouette, test ten print directions, or present several colourways across a range before a single sample is ordered.
This helps buyers who find it hard to visualise a finished product from a sketch. That is more common in buying offices than people admit and it does not make someone a poor buyer. Seeing twenty versions before the first sample exists improves the directional decision you make early, which is where those decisions matter most.
The output still depends on the quality of the brief that produced it, and on the judgement of whoever is reviewing what comes back.
3D Sampling and Tech Packs
For private label buyers this is where the practical impact is strongest, because sampling is the costliest part of development, the most prone to delay, and the part that damages the critical path most when it goes wrong.
It is also where the waste sits. Products get sampled repeatedly and some never reach the final range, costing time, money and supplier goodwill.
Virtual sampling lets a designer build a garment digitally and simulate how it drapes, fits and reads across colourways. Reducing approval rounds from five to two can cut costs by around a third. Where every round costs money and time, that is a significant saving, and it is a sustainability argument as much as a commercial one.
Tech pack generation traditionally takes four to eight hours per style. These tools produce a structured first draft from a sketch, which a technical designer then reviews and refines. Moving from a blank page to correcting something is considerably faster. It remains a starting point, and it still needs someone who understands construction, fit and grading well enough to spot what is wrong. A tech pack with incorrect information will still produce a bad sample.
Making Sense of Customer Data
When I started, a buyer’s view of the customer came from sales data, store feedback and the occasional focus group. The picture was partial and slow to update.
Ecommerce changed that entirely. Suddenly there was data on what customers clicked, how long they stayed on a page, whether they added to basket or removed it, what they searched for and did not find. The volume now available is almost incomparable to ten years ago.
The volume is now the problem. Most of it sits in systems buying teams have neither the time nor the tools to analyse. You get a weekly sales report. You are unlikely to get a real-time read on search behaviour, return patterns by product type, or what your customer looked at and walked past.
This is where the technology makes the volume usable. It will not tell you which parts matter to your business and your category, because a great deal of it will not be relevant to either. Knowing what matters is your job, and it always was.
The Three Risks Worth Taking Seriously
- Treating an output as a decision. A range review is in two days, everyone is stretched, and a tool produces a plan that looks logical. The option counts read well, the phasing looks right, and nobody has time to interrogate it. The problem arrives when someone asks why a particular decision was made. If the honest answer is that the tool recommended it, the decision has no foundation. When you commit quantities to a budget you are working through a commercial strategy, not populating a spreadsheet. The test is simple: can you defend every decision in that plan in front of a room of senior stakeholders?
- Data quality. Outputs are only as good as the data behind them, and in most buying teams that data is not as clean or complete as people assume. There are gaps and anomalies, and the last several years have produced more of both than most businesses have ever had to handle, from disruptions that made sell-through look artificially high to seasons where product simply could not be obtained. The absence of sales data does not mean there was no demand. Think about it the way you think about a PLM system: genuinely useful, and only as reliable as what was entered into it.
- The career risk, which concerns me most. The two hours you would have spent working through an analysis manually are not only about reaching the answer. They are how commercial instinct gets built. Sitting with a sales report style by style, noticing anomalies and asking why week twenty seven looks the way it does, is how pattern recognition develops. Skip that early and you reach outputs faster without the foundation, so when a tool produces something that looks good and is wrong, you have nothing to catch it with.
None of this argues against using the tools, and the pressure driving the shortcut is real. When something produces in twenty minutes what took two days, the temptation is understandable. The reality check still has to happen, because you remain accountable for the decision.
Elisabeth's Final Thoughts
These tools are not going away, nor should they. Every one of them works best in the hands of someone who brings enough commercial understanding to know what to do with the information and, more importantly, to recognise when something does not look right. That recognition is not a technical skill. It is built slowly, through the unglamorous work of reading your own numbers and asking why, and it is still the thing that makes a buyer valuable.
Related reading
AI in Fashion Buying FAQs
Five areas in practice: trend and demand forecasting across online and offline signals, range planning and option modelling, design and product development through generated variations, 3D virtual sampling and tech pack drafting, and interpreting the volume of customer data that buying teams already hold but cannot process.
No, because the tools produce outputs rather than decisions. An AI range plan can look logical and still be indefensible in a review if nobody can explain the commercial reasoning behind an option count or a price point. The judgement about whether something is right for your customer, at your price, in your trading context, is still yours.
Virtual 3D sampling lets a designer build a garment digitally and simulate how it drapes, fits and reads in different colourways. Reducing approval rounds from five to two can cut costs by around a third. For a private label buyer where every round costs money and time, that is a genuine commercial saving as well as a sustainability one.
It can produce a structured first draft from a sketch, covering the flat, the bill of materials, points of measure and construction notes. A technical designer still has to review and correct it before it goes anywhere. A tech pack with incorrect information will still produce a bad sample. What has changed is how quickly you reach the first draft, not whether it needs checking.
Treating the output as a decision rather than the starting point for one. The related risk, and the more serious one long term, is that junior buyers who skip the manual analysis never build the pattern recognition that becomes commercial judgement. If you have not developed the instinct, you cannot spot when a confident-looking output is wrong.
