AI and pricing: markdown optimisation explained
Markdown optimisation uses demand models to decide when and how deeply to discount. How it works, what it needs, and why brand and wholesale relationships must set its limits.
KEY TAKEAWAYS Summary by the editors
- Markdown optimisation models estimate how demand responds to price over time and recommend the timing and depth of discounts to meet a stock or margin objective.
- The objective must be defined explicitly, for example maximising margin while clearing stock by a set date, because different goals produce different recommendations.
- Reliable price response estimates require clean history of past price changes, promotions and stock levels.
- Brand positioning and wholesale partner agreements should act as hard constraints on what the model may recommend.
- The biggest gains usually come from earlier, more targeted markdowns rather than deeper or more frequent ones.
Every season ends with stock that did not sell at full price. The decisions that follow, when to reduce prices, by how much and on which items, have a direct effect on margin. Traditionally they are made with fixed calendars and rules of thumb: a first markdown at a set point, a deeper cut later, clearance at the end. Markdown optimisation replaces some of that routine with demand models. It can improve results, but only if the objectives and boundaries are set by people who understand the brand.
How does markdown optimisation work?
The core of any markdown model is an estimate of price elasticity: how much demand for a product changes when its price changes. The model learns this from past price changes and their effect on sales, adjusted for factors such as season, remaining stock, product age and promotions.
With that estimate, the model simulates scenarios for each product or product group: discount now or later, by a small or large amount, and calculates expected units sold, revenue and margin by the end of the selling period. It then recommends the path that best meets the defined objective.
An illustrative example: a style has a fixed number of units left with several weeks of season remaining. Keeping full price may sell some units but leave a large remainder for deep clearance. A moderate markdown now may sell more units at a reasonable margin and avoid the clearance. The model quantifies these trade-offs for many styles at once, which is impractical by hand.
Better models also work at the right level of aggregation. Individual styles often have too little data for a stable estimate, so models pool information across similar products, for example the same category and price tier, while still allowing for differences in remaining stock and selling pace. Planners should know at which level the elasticity was estimated, because it affects how much trust a single recommendation deserves.
What objective should the model optimise?
This is a management decision, not a technical one. Common objectives include:
- Maximise gross margin over the remaining season
- Clear a defined share of stock by a specific date to free space for new deliveries
- Minimise leftover stock that must go to outlets or off-price channels
- Protect a minimum price level for brand reasons
Each objective produces different recommendations. A model told only to maximise revenue may recommend aggressive discounts that harm margin and brand perception. Being explicit about the objective, and the constraints around it, is the most important step.
How does AI compare with rule-based markdowns?
| Aspect | Rule-based | Model-based |
|---|---|---|
| Timing | Fixed calendar for all products | Product-specific, based on sales and stock |
| Depth | Standard steps such as fixed discount tiers | Calculated per product or group |
| Data use | Limited, mostly aggregate sell-through | Elasticity, stock, product age and seasonality |
| Transparency | Easy to understand | Requires explanation of drivers |
| Typical risk | Discounting too late or too uniformly | Poor estimates where history is thin |
What are the limits and risks?
- Thin history: products that have rarely been discounted give little information about price response, so estimates rely on similar items.
- Confounded data: if past markdowns coincided with marketing campaigns or weather changes, the model may misattribute the effect.
- Brand erosion: frequent or visible discounting trains customers to wait for sales, an effect that short-term models do not capture.
- Channel conflict: markdowns in a brand's own channels can undercut wholesale partners who still hold the same products at full price.
What does a sensible implementation look like?
- Define the objective and hard constraints: minimum prices, protected products, partner agreements and channel rules.
- Clean the price and promotion history so that past markdowns, campaigns and stock-outs are clearly recorded.
- Pilot on a defined category or region and compare results with a group following the existing rules.
- Review recommendations weekly with merchandising and allow overrides with a recorded reason.
- Evaluate on margin, sell-through and leftover stock at season end, not only on units sold.
What should pricing and merchandising leaders expect?
The most common benefit is not deeper discounting but better timing. Models tend to identify slow sellers earlier and recommend moderate markdowns sooner, while leaving strong sellers at full price longer. That combination usually protects margin better than uniform calendars.
Equally important is what the model does not decide. Whether a brand discounts at all in certain channels, how it treats its wholesale partners and how much price consistency matters to its positioning are strategic choices. Markdown optimisation is a tool for executing those choices well, not a substitute for making them.
Frequently asked questions
What is price elasticity in fashion?
Price elasticity describes how strongly demand for a product changes when its price changes. In fashion it varies by category, brand position, product age and season, which is why models estimate it per product or product group rather than using a single figure.
Can markdown optimisation work for brands that sell mainly wholesale?
It is most directly applicable to a brand's own retail and e-commerce channels. For wholesale-led brands, the main value is aligning own-channel markdowns with partner agreements and using sell-through data to anticipate where partners will need support.
Does AI pricing mean more discounting?
Not necessarily. Well-configured models often recommend earlier, more targeted markdowns on slow sellers and fewer discounts on strong sellers, which can reduce overall discounting compared with uniform calendars.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
