Markdown Optimization
The three common mistakes that leave 12% of margin on clearance racks — a guide for Merchandising Managers, Buying Directors, and Allocation teams in fashion, apparel, and lifestyle retail.
DATAQUANT RESEARCH TEAM · MERCHANDISING & MARKDOWN · 11 MIN READ
Markdown is the most expensive line of margin sacrifice in retail merchandising. For a typical apparel or lifestyle retailer, end-of-season markdowns destroy 14–20% of season-start gross margin. For weaker seasons or weather-affected categories, the figure is materially higher. The aggregate annual impact is usually the largest discretionary margin loss in the P&L.
Despite the magnitude, markdown decisions at most retailers are still made through a combination of merchant intuition, inventory pressure, and end-of-season scheduling rules. Analytical markdown optimisation has been a topic in retail circles for over a decade; in our experience, fewer than 30% of mid-market apparel retailers have deployed it operationally. The reason is not that the math is hard. The reason is that the implementation requires merchant adoption, and merchants — with reasonable cause — distrust models that override their judgement on the categories they have spent careers learning to read.
This piece is for the Merchandising Manager, Buying Director, or Director of Allocation who is either evaluating a markdown optimisation programme or trying to figure out why their existing programme is not producing the margin recovery it promised. It walks through the three structural mistakes we see most often and what working implementations look like.
Why markdown decisions are structurally hard
A markdown decision is not a single decision. It is a sequence of decisions across a season, with each decision affecting the optimal subsequent decisions. The merchant deciding whether to take a 25% markdown in week 8 is implicitly deciding what their week-12 markdown will need to be — because the week-8 decision affects sell-through, which affects remaining inventory, which affects pricing pressure later. The right framing is sequential dynamic programming, not a one-shot optimisation.
The decision is also affected by externalities the merchant cannot directly observe. Demand for a given style is shaped by weather, competitor markdown timing, store traffic, social-media trend cycles, and broader consumer sentiment — most of which the merchant has fragmentary visibility into. The intuition-driven decision is operating with much less information than the situation actually warrants.
These two structural difficulties — sequential dependency and missing externalities — are why analytical markdown optimisation produces measurable lift over intuition. They are also why most analytical implementations fail to land: the merchant correctly perceives that the model is also working with imperfect information, and rejects its recommendations on the cases where the merchant has clearly superior local knowledge. The implementation that works is the one that surfaces analytical recommendations while preserving merchant override authority.
Mistake 1: Markdowns triggered by calendar, not by sell-through
The most common pattern in apparel retail: markdowns are scheduled by calendar week relative to season start. Week 6 = first markdown 20%, Week 9 = 30%, Week 12 = 40%, Week 14 = 50%. The schedule is uniform across the assortment and is largely independent of how each style is actually selling.
The financial cost of calendar-based markdown is structural: styles that are selling well get marked down on the calendar even though the demand at full price would have continued; styles that are selling poorly get marked down too late because the calendar trigger is too patient. The first error is direct margin loss; the second is opportunity cost — a poorly-selling style that could have moved at week 6 with a 20% markdown sits at full price until week 9, by which point a 30% markdown is barely sufficient and a 40% is increasingly likely.
The diagnostic For your last full season, what percentage of styles received their first markdown more than 4 weeks after their sell-through trajectory predicted they would need one? In our audits, this figure is typically above 35%. Each of those styles represents a markdown that was deeper than it needed to be, by virtue of being later than it should have been. |
The fix is to trigger markdowns on sell-through performance against a forward-looking projection, not on calendar position. A style that is on track to sell through cleanly does not need a calendar markdown; a style that is materially below trajectory needs an early intervention. The decision rule shifts from “is it week 9?” to “is this style projected to require a markdown deeper than 30% if I do nothing?”
Mistake 2: Pan-store markdowns instead of regional differentiation
A markdown decision typically applies uniformly across the entire store base. The 25% markdown on a winter coat in week 8 is applied to every store, regardless of regional weather, regional sell-through, or regional inventory position. The merchant making the decision sees aggregate sell-through and decides on aggregate response.
The structural waste in this approach is meaningful. A winter coat selling well in northern stores (where temperatures dropped early) does not need the same markdown as the same coat selling poorly in southern stores (where temperatures stayed warm). The pan-store markdown leaves margin on the table in the strong stores and is insufficient to clear inventory in the weak stores.
Regional markdown differentiation — cluster stores by sell-through pattern, weather, and customer demographic, then make markdown decisions per cluster — typically recovers 2–4 points of season-end margin on apparel categories. The operational complexity is real (more cluster-level decisions, more pricing complexity at store level) but most modern retail systems can support it.
Mistake 3: No measurement of markdown effectiveness
A surprising share of retailers do not measure the actual incremental volume produced by their markdowns. Sell-through rises after a markdown — this is observed and counted as the markdown’s effect. But sell-through would have risen anyway as the season progressed, and competitor markdown timing affects category-level demand independently. Without isolating the genuine incremental effect, markdown decisions are made on the basis of what looks like worked rather than what actually worked.
The basic measurement requirement is to compare actual post-markdown sell-through to a counterfactual baseline that controls for season position and category dynamics. This is straightforward statistical work but is rarely operationalised in merchandising teams because the analytical resources sit in the central data team and are not embedded in the merchandising rhythm.
What working markdown optimisation looks like
A markdown optimisation capability that produces verified margin lift typically has these components:
Component 1: Style-level demand projection
Per style, per week, a forward projection of expected sell-through under the current price. The projection uses recent sell-through trajectory, comparable historical styles, weather forecast, promotional calendar position, and category-level seasonality. Output: expected sell-through with confidence interval at season end.
Component 2: Markdown response curves
Per category, an estimate of how sell-through responds to markdown depth. Apparel categories typically show strong response curves: a 20% markdown produces 40–60% sell-through lift in the markdown week and a smaller continuing lift in subsequent weeks. The response curves are estimated from historical markdown events with appropriate counterfactual controls.
Component 3: Sequential markdown optimisation
A model that, given the projection and response curves, recommends the markdown sequence (depth and timing) that minimises total markdown spend while achieving target end-of-season inventory clearance. The model is sequential — the decision in week 6 is made considering its effect on the optimal week 12 decision — not single-shot.
Component 4: Merchant-in-the-loop UI
A workflow tool that presents the merchant with the model’s recommendation, the underlying projection, the alternative options the model evaluated, and the rationale for the recommended choice. The merchant retains override authority. Override decisions are logged and feed back into model recalibration.
Component 5: Outcome measurement loop
Weekly review of model recommendations, merchant overrides, and outcome variances. The review surfaces categories where the model is consistently wrong (recalibration target), categories where merchants are consistently overriding (training opportunity), and categories where the model and merchants agree (those decisions can be increasingly automated).
What this looks like in practice
A €620M apparel retailer we worked with deployed markdown optimisation across women’s ready-to-wear and men’s tailoring — covering 320 stores. The implementation took 16 weeks, including 4 weeks of category-by-category response-curve estimation and 6 weeks of merchant-workflow design.
After two seasons of operation: end-of-season margin improved by 4.2 points across the optimised categories — €7.8M of recovered margin annualised. Markdown depth in the deepest tier (50%+) reduced by 31% of stock units, because earlier intervention at shallower depths cleared the inventory that would otherwise have ended up at 50%+. Merchant overrides ran at 22% of recommendations — high enough to indicate the merchants were genuinely engaged, low enough to indicate they trusted the model on most decisions.
The model didn’t replace the merchants. It gave them better information than they’d previously had — and they used it. |
Three pitfalls in implementation
- Treating it as a black-box optimisation. A model that produces a recommendation without surfacing the rationale will be rejected by experienced merchants. The transparency layer (showing projection, response curve, alternative scenarios) is what makes the recommendation acceptable. Without it, the model produces a deck no one operationalises.
- Building the model before securing the data layer. Style-level demand projection requires clean style-level sales data, joined to weather and category-level seasonality, refreshed weekly. Most retail data warehouses have this data but it is rarely structured for daily/weekly modelling use. Spending time on the data layer first produces a faster overall implementation.
- Failing to align finance and merchandising on the success metric. Merchandising teams optimise sell-through and end-of-season inventory; finance optimises gross margin. The model needs to optimise gross margin minus markdown spend, which sometimes requires accepting higher inventory carry-over than merchandising would otherwise prefer. Pre-aligning the success metric across functions prevents post-implementation conflict.
Closing thought
Markdown is one of the few merchandising decisions where analytical optimisation has been studied for years and has demonstrated repeatable, large-scale margin recovery. The reason most retailers have not deployed it is not that the analytics are immature — they are quite mature. It is that the implementation requires merchant adoption, and merchant adoption requires a transparency-and-override architecture that most analytical projects deprioritise. Build the merchant workflow as carefully as the model itself, and the margin recovery follows reliably.