Using AI in assortment planning without losing the brand

Ask a model trained on three years of sales to build next season's range and it will probably recommend more of the black knitwear, the slim chino and the logo sweatshirt. It will be statistically right and strategically dangerous. Assortment planning is where AI's strengths and the brand's identity collide most directly, and getting the balance right is now a core merchandising skill.

What can AI do well in assortment planning?

The quantitative layer of the assortment is where machine learning earns its place. It handles large numbers of styles, stores and sizes consistently and spots patterns that are easy to miss in spreadsheets.

• Store and account clustering: grouping retail doors or wholesale accounts by actual buying and selling behaviour rather than by region or size alone.
• Depth recommendations: suggesting quantities per option and cluster based on sell-through of comparable products.
• Size curves: estimating size distributions by category and cluster, which reduces leftover fringe sizes.
• Range rationalisation: identifying options that consistently under-perform or overlap with stronger alternatives.
• Gap analysis: showing price points or categories where demand appears unmet in specific clusters.

Why can data-driven ranges dilute a brand?

A model optimises for the objective it is given, usually sales or sell-through, using the history it was trained on. That creates a built-in bias towards repetition. Products that were never offered cannot appear in the data, and pieces that build brand image but sell modestly look like poor performers.

Over several seasons, a range planned purely on these signals tends to converge on safe, average products. Wholesale accounts notice: if the collection looks like everything else on the shop floor, the brand loses its reason to be stocked. Consumers notice too, often before the data does.

Read also

The AI vocabulary every fashion buyer should know

How should roles be divided between AI and the team?

The principle is simple: the brand sets the creative architecture, AI informs the quantities, and planners take responsibility for the result. Where the boundary sits will differ by brand, but it should be written down rather than left to whoever operates the tool.

How do you protect the brand in practice?

1. Ring-fence part of the range. Define a share of options as strategic or experimental and evaluate them on different criteria, such as press, full-price sell-through at key doors or new customer acquisition.
2. Use attributes that describe the brand. Tag products with style, fit, design signature and newness level so that the model can distinguish a signature piece from a commodity item with similar sales.
3. Give newness its own forecast logic. Treat new styles as tests with small initial depth and rapid re-order potential rather than forcing them through models built for carry-over.
4. Review recommendations in a merchandising meeting. Present AI suggestions alongside the creative rationale, and record overrides and their outcomes.
5. Measure the right outcomes. Track full-price sell-through, markdown exposure and range distinctiveness alongside volume.

What changes for wholesale assortments?

In wholesale, the assortment is negotiated twice: once internally when the collection is built, and again with each account during the selling period. AI can help at the second stage by proposing account-specific selections based on what similar retailers bought and sold through. Used well, this gives buyers a stronger starting point and helps sales teams avoid accounts cherry-picking only the safest pieces.

It also creates a responsibility. If every account receives a proposal built from the same best-seller logic, the brand's presence across the market becomes uniform and diluted. Sales leadership should decide which key pieces must be represented at which tier of account, and the recommendation logic should respect those rules.

Read also

AI in fashion merchandising: what it does, what it needs and where it fails

What should buying and merchandising leaders take away?

AI makes assortment planning more precise, but precision is not the same as direction. The tools are best at answering 'how much and where', while the brand must keep answering 'what and why'. Teams that make this split explicit, protect space for newness and measure success beyond volume will gain efficiency without becoming forgettable.

The test at the end of each season is straightforward: did the range sell better, and does it still look unmistakably like the brand? If the answer to the second question is uncertain, the balance has tipped too far.