How to build AI-generated assortment proposals for retail accounts
A pre-filled order proposal per account can save buyers and reps hours, if it rests on clean sell-in history, clear rules and a rep who checks it. A step-by-step guide.
KEY TAKEAWAYS Summary by the editors
- An AI-generated assortment proposal is a suggested opening order for a specific retail account, built from that account's history, its peers and the new collection, which a rep and buyer then edit.
- The most important inputs are clean sell-in history per account, sell-out or reorder data where available, store profiles, budgets and the brand's own distribution rules.
- Proposals work best as a starting point in the appointment, not as an automatic order: Gartner found in 2025 survey data that 69% of B2B buyers prefer to check AI-generated insights with sales reps.
- New styles have no history, so proposals must map them to comparable past styles by attributes such as category, price band, fabric and fit.
- Brands should measure proposal quality by acceptance rate, edits made, and the season-end sell-through of accepted lines, not by how many proposals were generated.
An AI-generated assortment proposal is a suggested opening order for one retail account: which styles, colours, sizes and quantities the brand believes fit that store, given its history and the new collection. Building one requires reliable account-level sales data, a way to map new styles to past ones, explicit business rules and a rep who reviews the result with the buyer. Done well, it shortens appointments and makes assortments more consistent; done badly, it simply repeats last season.
What is an AI assortment proposal in wholesale?
In wholesale, buyers usually start a season with a blank order or a copy of last season's. An assortment proposal replaces that blank page with a pre-filled draft generated per account. Wholesale platforms already market such features: JOOR's December 2025 wholesale trends paper lists assortment recommendations among the AI functions it uses, alongside analytics and retailer connection suggestions. Many brands also build proposals in-house from their ERP and sales app data.
The proposal is a recommendation, not a decision. Gartner reported in May 2026, from a survey of 645 B2B buyers, that 69% prefer to check AI-generated insights with sales reps. In fashion, where buyers defend their own taste and margin targets, the rep's role in explaining and adjusting the proposal is central.
Which data do you need before you start?
| Input | Why it matters | Typical quality issue |
|---|---|---|
| Sell-in history per account and style | Shows what the account buys and at what depth | Style codes change between seasons |
| Sell-out or reorder data | Shows what actually sold at the store | Only available for some accounts |
| Store profile (size, location, clientele, doors) | Explains differences between similar accounts | Kept in reps' heads, not systems |
| Open-to-buy or budget | Caps the total proposal value | Shared late or not at all |
| Product attributes of the new line | Maps new styles to past performance | Incomplete or inconsistent tagging |
| Distribution rules | Exclusives, minimums, protected styles | Rules exist only in emails |
How do you build the proposal step by step?
- Clean and connect account history. Link at least two to three seasons of order lines to a stable account ID and to style attributes. Without this, the model learns noise.
- Describe every new style by attributes. Category, price band, fabric, fit, colour family and theme let the system find comparable past styles, since new styles have no sales of their own.
- Segment accounts. Group stores with similar profiles and buying patterns, so a small account with thin history can borrow information from its peers.
- Generate a draft per account. Score each new style for each account, select the best fitting lines within the budget and propose quantities and size curves from history.
- Apply business rules. Enforce minimums, exclusives, distribution limits and must-have hero styles after the model has scored, so commercial policy is never left to the algorithm.
- Explain each line. Show the rep a short reason, for example "similar style sold through well in this store last autumn", so the proposal can be defended in front of the buyer.
- Review in the appointment and capture edits. Every accepted, changed or rejected line is training data for the next season.
How should reps use proposals in the buying appointment?
Reps should present the proposal as a draft that reflects the buyer's own history, then invite changes. A useful pattern is to agree the total budget and category split first, then walk through lines where the model is least confident. Gartner reported in May 2026 that sales organisations giving sellers AI-enabled next best actions were 2.6 times more likely to achieve commercial growth; in a buying appointment, the equivalent is a short list of suggested adds and cuts, not a hundred-line spreadsheet.
Reps also hold knowledge the system lacks: a store refit, a new buyer, a competitor opening nearby. A simple field in the sales app to record such context, and a way to override the proposal with a reason, improves both the current order and the next model.
Timing matters as well. Sending the proposal to the buyer a few days before the appointment lets them compare it with their own open-to-buy and arrive with questions, which usually makes the meeting shorter and more focused. For accounts that order without an appointment, the proposal can be shared directly in the ordering portal as an editable draft, with the rep available for questions. In both cases the brand should record which version the buyer saw, so that later comparisons between proposal and final order are meaningful and the model learns from real decisions rather than from guesses.
What are the main risks of AI assortment proposals?
- Homogenisation: similar accounts receive near-identical proposals, eroding the differentiation retailers value.
- Bias towards data-rich accounts: large accounts with long histories get better proposals than small independents.
- False precision: quantities to the unit suggest certainty that the data does not support.
- Ownership confusion: if a proposal underperforms, it must be clear that the buyer and brand made the final decision.
- Data access: sell-out data shared by retailers must be used only within the agreed purpose.
Market conditions add pressure. McKinsey and The Business of Fashion reported in November 2025 that 46% of executives expected industry conditions to worsen in 2026. In a cautious market, a proposal that overestimates depth damages trust quickly, so conservative quantities with clear reorder options are usually wiser.
How do you measure whether proposals work?
Track a small set of measures each season: the share of proposal lines accepted unchanged, the size of edits by category, the time per appointment, and the season-end sell-through of accepted proposal lines compared with lines added manually. Compare against a control group of accounts or a simple rule-based proposal. If AI proposals are accepted but sell no better than the buyer's own choices, the benefit is convenience rather than performance, which may still be worth having but should be described honestly.
Frequently asked questions
What is an assortment proposal in fashion wholesale?
It is a suggested opening order for a specific retail account, listing styles, colours, sizes and quantities. It is built from the account's history, comparable stores and the new collection. The buyer and rep then edit it during the appointment.
How does AI handle new styles with no sales history?
It maps new styles to comparable past styles using attributes such as category, price band, fabric, fit and colour family. The quality of these attributes largely determines the quality of the proposal. Truly novel styles still need human judgement.
Do buyers accept AI-generated order proposals?
Acceptance varies by brand and account. Survey data from Gartner shows most B2B buyers want to validate AI-generated insights with a sales rep, so proposals work best as an editable starting point. Measuring acceptance and edits each season shows where the model needs improvement.
What data do I need to build assortment proposals?
At minimum, two to three seasons of clean order history per account, consistent product attributes, account profiles and distribution rules. Sell-out or reorder data and budgets improve results considerably. Without stable style and account identifiers, proposals will be unreliable.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
SOURCES
- JOOR (via Across Magazine): Top 5 Trends in Wholesale for 2026
- Gartner: Survey finds 69% of B2B buyers turn to sales reps to validate AI-generated insights
- Gartner: Sales organizations that provide AI-enabled next best actions are 2.6x more likely to achieve commercial growth
- McKinsey & Company: The State of Fashion 2026: When the rules change
