Which styles should you drop before the sales campaign? How AI helps edit the line
Carrying too many styles into the selling campaign costs samples, minimums and margin. How AI scores styles before the line is shown to buyers, and where merchants must overrule it.
KEY TAKEAWAYS Summary by the editors
- A line edit is the decision, before the wholesale selling campaign, on which developed styles and colourways are shown to buyers and which are dropped.
- AI supports the line edit by predicting likely order intake for each style from attributes and comparable past styles, and by flagging overlap between styles that would cannibalise each other.
- Brands are already tightening ranges: JOOR data published in December 2025 shows brands developing a capsule collection of new styles rose from 19% in 2023 to 37% in 2025.
- Levi Strauss said in July 2025 it was cutting underperforming styles and colourways to focus on products more likely to sell at full price, as reported by EMARKETER.
- Predictions for new and directional styles are weak by nature, so AI should rank the core and carry-over range while merchants protect a deliberate share of newness and brand-building pieces.
AI helps decide which styles to drop before a sales campaign by estimating each style's likely order intake from its attributes and the performance of comparable past styles, and by identifying styles that overlap so closely that they split demand. The merchant team then uses this ranking to cut the weakest and most redundant options before samples are made and the line is shown to buyers. The model informs the edit; it does not replace the creative and commercial judgement of the line review.
What is a line edit, and why does it matter now?
In a wholesale business, design and development usually produce more styles and colourways than the brand intends to sell. The line edit, or line review, reduces this range to the collection that will be sampled and presented to retail buyers. Every style that survives costs money: samples, photography, showroom space, production minimums and, if it sells thinly, markdowns or excess stock.
Several pressures push brands towards tighter ranges. JOOR's December 2025 wholesale trends paper reported that the share of brands producing a full new collection fell from 86% in 2023 to 83% in 2025, while the share developing a capsule collection of new styles rose from 19% to 37%, and evergreen styles grew from 37% of platform sales value in 2019 to 49% in 2024. The same paper notes that EU rules under the Ecodesign for Sustainable Products Regulation will restrict the destruction of unsold products, which raises the cost of overproduction.
How does AI score styles before buyers see them?
The usual approach treats each new style as a bundle of attributes, such as category, silhouette, fabric, colour family, price band, size range and theme. A model trained on several seasons of order history learns how these attributes, and combinations of them, relate to wholesale intake by market and account type. It then predicts intake for each proposed style and compares it with production minimums and margin targets.
Further analyses add depth. Similarity scoring shows where two styles are near duplicates, so the weaker can be dropped without losing demand. Price ladder checks show gaps or crowding within a category. Where available, sell-out data from retailers helps distinguish styles that buyers order from styles that consumers actually buy, which matters for reorder potential.
| Analysis | Question answered | Reliability |
|---|---|---|
| Intake prediction for carry-over and updates | Will this style reach minimums? | Relatively high with good history |
| Intake prediction for new styles | How might a new idea perform? | Low, use as a weak signal |
| Similarity and cannibalisation | Which styles compete for the same order? | Moderate, depends on attribute quality |
| Price ladder analysis | Are price points crowded or missing? | High, largely rule-based |
| Margin at predicted volume | Is the style worth producing? | Depends on costing accuracy |
Which styles are typical candidates to drop?
- Styles predicted to fall well below production minimums across all markets.
- Colourways that add complexity without adding predicted intake, especially in small sizes of the range.
- Near duplicates of stronger styles in the same price band.
- Styles whose margin at predicted volume is below target after sampling and minimum costs.
- Items with weak sell-out history in comparable past styles, even if buyers ordered them.
Real companies frame range reduction in commercial rather than technical terms. EMARKETER reported in July 2025 that Levi Strauss was cutting its number of SKUs, removing underperforming styles and colourways to make room for products more likely to sell at full price, and expected this to reduce manufacturing expenses and markdowns. Nike's chief financial officer said in December 2024, as reported by Retail Dive, that the company was targeting a significant reduction in the supply of its classic footwear franchises over the following seasons. Neither report attributes these decisions to AI; they show the type of decision that line edit analytics supports.
Where should merchants overrule the model?
Models trained on past orders reward what sold before. They tend to undervalue new silhouettes, emerging colours and pieces that complete a look in the showroom even if they sell modestly. They also cannot judge brand positioning, a designer's intent or a strategic retailer's request. Merchants should treat low scores as a question (why is this style here?) rather than an automatic cut, and record the reason whenever they keep a low-scoring style, so the decision can be reviewed after the campaign.
Timing is another constraint. The line edit typically happens weeks before the selling campaign, when costing may still be provisional and fabric minimums are not final. Scores should therefore be refreshed as costs and minimums are confirmed, and the merchant team should know which inputs are estimates. A style that looks unprofitable on a provisional cost may become viable once a fabric is shared with other styles, a link the model will only see if the bill of materials data is connected to the analysis.
How should a brand introduce AI into the line review?
- Clean the attribute data for at least three past seasons, since the model depends on consistent tagging.
- Back-test: score a past season's full development range and compare predictions with the actual order book.
- Start with carry-over and update styles, where predictions are most reliable.
- Present scores in the line review alongside margin and minimums, with a short explanation for each.
- Track the outcome of kept and dropped styles, including whether dropped ideas were picked up by competitors.
Expectations should be modest. McKinsey and The Business of Fashion reported in November 2025 that technology-driven productivity is becoming necessary because scale and low-cost sourcing no longer sustain margins. A sharper line edit is one practical route to that productivity, but its benefit comes from fewer wasted samples and less excess stock, not from a model that knows what buyers will love.
Frequently asked questions
What is a line review in fashion?
It is the meeting or process in which design, merchandising and sales decide which developed styles and colourways go into the final collection shown to buyers. Styles are cut for weak commercial potential, overlap or margin reasons. It usually happens before sampling for the selling campaign.
Can AI predict which new styles will sell?
Only partly. Models can estimate performance from attributes and comparable past styles, which works reasonably for carry-over and updates. For genuinely new designs, predictions are weak and should be one input among several.
How many styles should a wholesale collection have?
There is no universal number. It depends on the brand's accounts, categories, production minimums and positioning. Industry data shows many brands moving towards tighter collections and capsules, but the right size should come from margin and sell-through analysis of past seasons.
What data does AI need to edit a fashion line?
Consistent product attributes, several seasons of order history by style and account, costing and minimums, and ideally sell-out data from retailers. Attribute quality is usually the biggest weakness. Without it, similarity and prediction models are unreliable.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
