Markdown Optimization at Scale

How a €620M European apparel retailer recovered €7.8M of margin across 320 stores — by replacing calendar-driven markdowns with sell-through-driven optimisation.

EU APPAREL RETAIL  ·  €620M REVENUE · 320 STORES  ·  ENGAGEMENT: 16 WEEKS

Illustrative case based on engagement patterns from European apparel retail. Specific business details, names, and exact figures have been adapted to preserve client confidentiality.

Situation

A €620M European apparel retailer operating 320 stores across five countries had been running calendar-based markdowns for fifteen years. The schedule was uniform: week 6 = 20% off, week 9 = 30%, week 12 = 40%, week 14 = 50%. The schedule was applied uniformly across the assortment, across stores, across regions — with manual merchant override authority that was rarely used at scale.

End-of-season margin had been deteriorating for four years. The CFO’s seasonal P&L review consistently showed markdown destroying 17–20% of season-start gross margin. The number had become accepted as the cost of doing business in apparel retail. The merchandising team had pushed back on changes to the markdown discipline because they believed the calendar provided commercial predictability that the buying side relied on for next-season planning.

The CEO’s read on the situation: the calendar discipline had value, but the cost was higher than necessary. He commissioned a strategic review with two questions — was the markdown spend recoverable, and could the recovery happen without disrupting the planning rhythm the merchandising team needed?

Diagnosis

The diagnostic phase ran four weeks. Three findings reframed the engagement.

Finding 1 — Calendar-driven markdown was producing systematic over-markdown

A retrospective analysis of two prior seasons showed that approximately 37% of styles received their first markdown more than 4 weeks after their sell-through trajectory had clearly indicated they would need one. Those styles ended up requiring deeper markdowns at week 12–14 than they would have needed at an earlier intervention.

Conversely, approximately 24% of styles received markdowns on the calendar even though their sell-through trajectory predicted they would have cleared at full price within the season. These markdowns were direct margin loss with no commercial rationale.

Finding 2 — Pan-store markdown was masking regional variation

When sell-through was decomposed by store cluster (defined by region, customer demographic, and historical sell-through pattern), substantial intra-style variation emerged. The same winter coat that sold cleanly at full price in northern stores was sitting in southern stores at 40% sell-through by week 8. The pan-store 25% markdown applied at week 9 was insufficient to clear the southern inventory and unnecessary to clear the northern inventory.

Finding 3 — The data infrastructure was workable

The retailer’s ERP and merchandising systems contained: style-week-store sales, weekly inventory positions, regional weather, marketing-promotion calendar, and competitor markdown observations from a third-party tracking service. The data was structured, current, and — critically — already feeding into the existing markdown calendar logic. The required transformation was not a data infrastructure project; it was a decision-logic project on top of existing data.

Approach

The build phase ran from week 5 through week 16 across four workstreams.

Workstream 1 — Style-level demand projection (weeks 5–9)

Per style, per week, a forward projection of expected end-of-season sell-through under current price. The projection used recent sell-through trajectory, comparable historical styles (matched on category, fabric, season position, and price tier), weather forecast, and category-level seasonality. Output: expected sell-through with confidence interval at season end, refreshed weekly.

Workstream 2 — Markdown response curves (weeks 6–10)

Per category, an estimate of how sell-through responds to markdown depth, at the store-cluster level. Apparel categories typically show strong response curves: a 20% markdown produces 40–60% sell-through lift in the markdown week, with continuing tail effects through subsequent weeks. The response curves were estimated from historical markdown events with appropriate counterfactual controls (matched non-marked styles in adjacent categories).

Workstream 3 — Sequential optimisation engine (weeks 8–12)

Given the demand projection and response curves, an optimisation engine recommended the markdown sequence (depth and timing) that minimised total markdown spend while achieving target end-of-season inventory clearance, per style × store-cluster combination. The model evaluated multiple markdown trajectories and selected the path with lowest total margin sacrifice.

Workstream 4 — Merchant-in-the-loop UI (weeks 11–16)

A workflow tool was built into the existing merchandising-system UI. For each style requiring a decision in any given week, the merchant saw: the current sell-through trajectory, the model’s projected end-of-season outcome under no-action and under recommended-action, the recommended markdown depth and timing, the alternative options the model evaluated, and the rationale.

Merchants retained full override authority. Override decisions were logged with merchant rationale, and the data fed back into model recalibration. The expectation set with the merchandising team was that overrides would run at 20–30% in the first season, decreasing as model trust grew.

Outcome

The system went live mid-season for the spring/summer collection (a partial season) and ran fully for two complete seasonal cycles before formal evaluation. Outcomes against the prior two-season baseline:

The 4.2-point margin improvement — €7.8M annualised — came principally from two structural changes. Earlier first-markdown intervention (week 7.4 → week 6.1) reduced the depth of the eventual markdown that would have been needed otherwise. Store-cluster differentiation reduced unnecessary markdown in stores where the style was selling well at full price.

The 31% reduction in stock units reaching the 50%+ markdown tier was the more strategically important number. It indicated the model was identifying styles in trouble earlier and clearing them at shallower depths, rather than allowing them to slide into the deepest markdown band. This had a knock-on effect on next-season inventory planning — less last-season inventory carrying over meant cleaner season starts.

Lessons

1. Calendar-based markdown is structurally over-marking. The two structural waste patterns — marking what would have cleared at full price, and marking too late so the eventual markdown is deeper — together account for most of the margin recovery available from optimisation. Both waste patterns are invisible in the calendar discipline. Both are visible the moment sell-through projections are computed at style level.
2. Store-cluster differentiation produces 1–3 points of margin even alone. Even without sequential optimisation, simply differentiating markdown by store cluster captures meaningful margin. Retailers without the analytical maturity for sequential optimisation can capture much of the value from this single change.
3. Merchant adoption is the real implementation risk. The model was the easier part. The harder part was building the merchant workflow that surfaced the recommendation transparently, preserved override authority, and produced a daily rhythm the merchandising team trusted. Implementations that under-invest in the merchant UI fail at adoption regardless of model quality.
4. The value compounds across seasons. The first full season produced about 60% of steady-state value. The second season added the rest as the model accumulated more recent data and the merchant team grew comfortable accepting more recommendations without override. Plan for a two-season ramp, not a one-season payoff.
5. Inventory carry-over reduction is the second-order benefit. €7.8M of margin is the headline. The 9% reduction in season-end inventory carry-over is structurally more valuable because it compounds: cleaner season starts produce better next-season decisions, fewer mid-season clearance pressures, and over time, a more disciplined buying cadence. This benefit is not in the first-year ROI calculation but is real and material.