How can AI set the right initial price for a fashion product?

KEY TAKEAWAYS Summary by the editors

1. AI sets initial fashion prices by estimating price elasticity, meaning how demand changes as price changes, from comparable past products, attributes and controlled price tests.
2. For new products without their own history, elasticity is borrowed from similar items, and the estimate improves once early sales are observed.
3. Research with the online retailer Rue La La found that demand for an item depended on the prices of other items in its category, so prices had to be optimised across the category rather than one by one.
4. A January 2014 field experiment at Rue La La raised revenue on the experimental products by about 9.7% with minimal impact on units sold, according to Harvard Business School.
5. Initial price models need clean sales data corrected for stockouts and promotions, and their recommendations should sit inside brand price architecture rules.

AI sets an initial price by estimating how demand for a product will respond to different price points, using comparable past products, product attributes and, where possible, controlled price tests. The model recommends the price that best meets a goal, such as full-price margin or sell-through, within the brand's price architecture. A good initial price reduces the markdowns needed later.

Why does the initial price matter so much in fashion?

In a seasonal business the initial price sets the ceiling for everything that follows. Price too high and the product stalls, waiting for its first markdown; price too low and the brand gives away margin on items that would have sold anyway. Most AI pricing attention goes to markdowns, but Zara's own clearance pricing research shows that even optimised markdowns only recover part of the value: in Caro and Gallien's 2008 field experiment the optimised process increased clearance revenue by approximately 6%. Getting the first price right means less to recover.

What is price elasticity and how does AI estimate it?

Price elasticity measures the percentage change in demand for a percentage change in price. A product with high elasticity loses many sales when its price rises; one with low elasticity barely reacts. Traditional analysis estimates elasticity from an item's own price history, but new fashion products have none. AI models therefore estimate it from patterns across many past products: how products with similar attributes, price bands and channels responded to price differences and changes.

MIT News describes the approach developed by David Simchi-Levi and colleagues: match new products to similar items, use machine learning to predict the relationship between price and demand, test prices against real sales and adjust the curve, then optimise prices across many products. A key finding at Rue La La, reported by Harvard Business School, was that demand for an item depended on the prices of other items in its category. Pricing one product in isolation ignores that customers compare.

Read also

AI and pricing: markdown optimisation explained

What data does an initial price model need?

Two corrections matter most. Sales on sold-out items understate demand, so the Rue La La researchers used machine learning to estimate lost sales before modelling. And promotional periods must be flagged, otherwise the model learns that every product sells at its discounted price.

How do you set an initial price with AI, step by step?

1. Define the objective. Decide whether the price should maximise full-price margin, sell-through or revenue, and how to weigh them.
2. Fix the price architecture. Give the model the allowed price points and the ladder between entry, core and premium lines.
3. Prepare the history. Clean two or more seasons of sales for stockouts, promotions and distribution breadth.
4. Estimate elasticity by segment. Group products by category, attribute and channel, and estimate price response for each group.
5. Simulate scenarios. For each new product, compare forecast units, revenue and margin at several candidate prices, including the effect on neighbouring products.
6. Decide with merchandisers. Treat the output as a recommendation and record when and why it is overridden.
7. Learn from early sales. Compare actual response in the first weeks with the forecast and update the elasticity estimates for the next drop or season.

Can you test prices without hurting the brand?

Price testing is the most direct way to learn elasticity, but in fashion it carries risk. Showing different prices to different customers for the same item can create fairness complaints, and in the EU traders must disclose when a price was personalised through automated decision-making, under consumer law amended by Directive (EU) 2019/2161. Safer test designs vary price across comparable products, markets or time periods rather than across individual customers, and keep the tested range inside the brand's price architecture.

What results can a brand realistically expect?

Published evidence is encouraging but specific. Harvard Business School reports that Rue La La's January 2014 field experiment raised revenue on the experimental products by about 9.7% with minimal impact on units sold. MIT News also reported a test at Groupon, where revenue on deals rose by about 21% overall. Both are online, price-led businesses with many short-lived items, which makes experimentation easy. A full-price fashion brand should expect smaller and slower gains, and should measure them on its own data against the current pricing method.

Read also

Using AI in assortment planning without losing the brand

What are the limits of AI initial pricing?

• Elasticity learned from past products may not hold for genuinely new categories or collaborations.
• Models capture what customers paid, not what the brand wants to stand for; price positioning remains a strategic choice.
• Price interacts with buy depth: a low price on a shallow buy simply sells out faster, so pricing and quantity decisions should be made together.
• Wholesale customers and marketplaces may react to price points that differ from recommended retail prices, so channel consistency needs a human check.

Used with these limits in mind, an elasticity model gives merchandisers a structured view of the trade-off between margin and volume at each price point, which is a more defensible basis for an initial price than the price of last season's nearest equivalent.

Frequently asked questions

What is price elasticity in fashion?

Price elasticity in fashion measures how much demand for a product changes when its price changes. Products with high elasticity lose many sales when prices rise, while products with strong brand appeal or few substitutes react less. AI estimates elasticity for new items from comparable past products.

How do you price a new clothing product with no sales history?

You borrow evidence from similar past products, matched on attributes, category, price band and channel, and estimate how their demand responded to price. AI models automate this matching and simulate demand at several price points, and early sales are then used to refine the estimate.

Can AI replace merchandisers in setting prices?

No. AI can estimate demand at different price points and recommend options, but brand positioning, price architecture and channel strategy remain human decisions. The most effective setups let the model propose and merchandisers decide, with overrides recorded and reviewed.

Is price testing allowed in the EU?

Price testing is generally possible, but EU consumer law requires traders to tell consumers when a price has been personalised on the basis of automated decision-making. Testing across markets, time periods or comparable products avoids individual personalisation, and legal review is advisable before any test.

GuideThe complete guide to AI in fashion merchandising and buyingRead the complete guide