How AI-based size curves and allocation reduce stockouts and markdowns in footwear.
Footwear exposes size risk more than most fashion categories. A brilliant colorway or silhouette won’t sell if the right sizes aren’t on the shelf, in the right store or DC, at the right time.
Traditional size curves built from last season’s sell‑through often miss shifts in trend, region, and channel mix; they also underweight returns reasons and out‑of‑stock lost sales. The result: chronic stockouts in popular sizes (women’s US 7–8, men’s US 9–10) and piles of odd sizes that get marked down. For performance and luxury footwear, misallocation is even costlier—release windows are narrow, and missed demand pushes consumers to resale.
AI helps by turning fragmented signals into dynamic size curves and smarter allocation. Inputs should include historical sales with true availability context (to avoid reading stockouts as “low demand”), returns reasons tied to fit/last width, localized trend signals, channel behavior (DTC vs. wholesale vs. marketplace), and launch calendars.
When models correct for stockouts and incorporate trend drift, they recommend different size mixes for different geographies and channels—even for the same style. That shift alone cuts lost sales and keeps odd sizes from clogging inventory. Industry sources highlight how trend and visual signals change demand timing; see Heuritech. For a broader overview of fashion e‑commerce dynamics that intersect with allocation decisions, Shopify’s enterprise brief is a useful baseline: Shopify. Macro context on volatility and consumer shifts appears in the McKinsey State of Fashion.
A practical design starts with data hygiene and explainability. Define a footwear attribute model that merchandisers recognize: last shape and width, heel/toe type, stack height, materials and stretch, lacing/closure type, and category (performance runner, court, lifestyle, dress). Tag transactions with availability so models know when a “slow” size actually sold out. Ingest returns reasons (“too narrow,” “heel slips”) to refine size curves by last/fit block.
• Correct for stockouts (censored demand) using techniques like demand estimation under inventory constraints.
• Segment curves by channel and geography; a style that skews narrower sizes in Tokyo may skew larger in Texas.
• Incorporate pre‑order and waitlist signals; treat them as demand with confidence bands.
• Use early‑signal transfer: borrow learning from similar silhouettes and brands for new styles with little history. Buy-to-allocate loops:
• Pre‑season: create initial size curves by style/channel/region with wide confidence bands where uncertainty is high.
• In‑season: update with weekly sell‑through, return feedback, and social/trend signals; reallocate across stores/DCs.
• Launches: simulate drops with constrained inventory; prioritize allocations to stores and digital channels where the model projects sell‑through without markdown.
For context on why visual trend signals matter for demand timing, review Heuritech. For the commerce lens on fulfillment and availability, Shopify’s fashion overview remains a good reference: Shopify.
Run this like a product with a scoreboard.
• Stockout reduction on top sizes by style/channel/region.
• Markdown rate and weeks of supply for tail sizes.
• Full‑price sell‑through and gross margin return on investment (GMROI).
• Transfer efficiency: % of inventory successfully reallocated before markdown windows.
• Publish freshness SLAs (e.g., weekly curve refresh; daily for launches) and a change log when curves update.
• Keep an explainability panel: “Why this mix?” with signals such as historical adjusted demand, return‑fit feedback, and regional trend shifts.
• Wire alerts when a store’s run‑rate deviates; trigger transfers or localized marketing.
• Use progressive delivery and canary rollouts when changing allocation policies to limit risk; deployment patterns summarized by HashiCorp.
• Instrument observability; track latency and error budgets for the planning service and data pipelines. Executive‑friendly overview: Splunk.
Outcome: fewer “sold out in my size” moments, lower markdowns on odd sizes, and a better customer experience—especially critical in performance and luxury footwear where fit and availability define loyalty.