Optimize Inventory Demand Planning

A senior demand-planner knowledge base for a multi-location retailer (40-200 stores, 300-800 active SKUs, using a demand planning suite such as Blue Yonder, Oracle Demantra, or Kinaxis alongside SAP or Oracle ERP, a WMS, POS feeds, and vendor portals) that sits between merchandising, supply chain, and finance. It covers forecasting method selection by demand pattern: moving averages for stable low-variability items (never for seasonal ones, since they lag trend changes by half the window), single, double, and triple exponential smoothing (Holt's for trending items, Holt-Winters for seasonal 52-week cycles, with alpha, beta, and gamma tuned on holdout data), STL or X-13ARIMA-SEATS seasonal decomposition to isolate trend, seasonal, and residual components before promotional-lift modeling, causal and regression models for price- and promotion-driven demand regularized against overfitting on sparse promo history, and gradient-boosting or neural-net ML for large SKU-by-history datasets, stated to beat simpler methods by 10-20% WAPE on promotional and intermittent items but needing quarterly retraining against drift. Accuracy is measured via MAPE (only reliable above 50 units/week), WMAPE (the dollar-weighted metric finance cares about), bias (healthy under plus-or-minus 5%, structural problem over 10%), and tracking signal (intervention needed past plus-or-minus 4).

Safety stock uses SS = Z x sigma_d x sqrt(LT + RP) for normal, stationary demand, with service-level Z-scores from 90% for C-items to 99% for critical A+ items, noting that moving from 95% to 99% nearly doubles safety stock, a variance-adjusted formula when vendor lead-time coefficient of variation exceeds 0.3, Croston's method plus bootstrapped distributions for lumpy or intermittent demand, and analogous-item profiling with a 20-30% buffer for new products with no history. Reorder logic covers inventory position (on-hand plus on-order minus backorders minus committed, never on-hand alone), Min/Max for stable items, ROP/EOQ (EOQ = sqrt(2DS/H), rounded to vendor case packs), periodic (R,S) review aligned to vendor delivery schedules, and tiered review frequency by vendor spend. Promotional planning covers stripping promo volume before baseline fitting, three lift-estimation methods with typical ranges (15-40% for a bare price cut, 80-200% with display and circular, 300-500%+ for doorbusters), cannibalization of 10-30% of lifted volume for close substitutes, and a post-promo dip of 30-50% of the incremental lift concentrated in the first week. ABC (value) and XYZ (predictability, by coefficient of variation) segmentation combine into a policy matrix - AX gets tight automated replenishment, AZ needs human review every cycle, CZ is a discontinuation candidate. Seasonal transition guidance covers committing 60-70% of a seasonal buy 12-20 weeks out with the remainder reserved for reorder, starting markdowns once sell-through drops below 60% of plan, and a hard season-end liquidation cutoff.

Decision-framework tables map demand pattern to primary and fallback forecast method with a review trigger, segment to target service level and Z-score, promotional lift to a four-branch decision tree, and sell-through-at-midpoint to a markdown action with expected margin recovery. Eight named edge cases are covered with concrete responses: zero-history new product launches, viral demand spikes where you should not chase since lead time exceeds the spike, overnight vendor lead-time doubling, cannibalization from an unplanned competitor promo, demand-regime changes, phantom inventory from WMS-versus-physical-count mismatches, vendor MOQ conflicts, and holiday calendar shifts breaking week-over-week comparisons. Communication guidance calibrates tone across six scenarios including routine reorders, lead-time escalations, stockout alerts, and markdown recommendations, and an escalation framework defines automatic triggers across a four-level chain from planner to VP Supply Chain, backed by tracked KPIs including WMAPE, bias, in-stock rate, weeks of supply, excess and dead stock, and PO fill rate with target and red-flag thresholds.

Use when forecasting product demand, calculating safety stock, planning replenishment cycles, estimating promotional impact, conducting ABC/XYZ segmentation, planning seasonal buys and markdowns, or investigating chronic stockouts, excess inventory, or forecast bias severe enough to need a redesigned planning process. Not for single-SKU, single-location, or ad hoc inventory questions with no real forecasting or safety-stock decision behind them - the frameworks here assume a multi-location retail operation with real vendor lead times and service-level trade-offs.

Input is demand history, lead times, service-level targets, and promotional or seasonal context for a SKU or SKU segment. Output is a selected forecasting method with parameters, a safety-stock and reorder-point calculation, a promotional lift and dip estimate, an ABC/XYZ classification with its policy, or an escalation or communication artifact matching the documented tone and structure. Deeper detail lives in three companion references: decision-frameworks.md, edge-cases.md, and communication-templates.md.

Assumes a demand-planning suite (Blue Yonder, Oracle Demantra, or Kinaxis), an ERP (SAP or Oracle), a WMS for DC-level inventory, POS data feeds, and vendor portals for purchase orders - the same system stack named in its own role description.

Demand planners, inventory analysts, and supply chain managers at multi-location retailers who need concrete formulas, thresholds, and decision tables, not just conceptual forecasting theory, for method selection, safety stock, reorder policy, and promotional or seasonal planning.