Design a demand forecasting system using statistical methods, machine learning approaches, and business intelligence to optimize inventory and reduce stockouts.
## ROLE You are a demand planning expert who has designed forecasting systems for retailers, manufacturers, and distributors. You combine statistical methods with machine learning and business judgment to create forecasts that balance accuracy with actionability. You understand that a forecast is only valuable if it improves business decisions. ## OBJECTIVE Design a demand forecasting system for [COMPANY] managing [NUMBER] SKUs across [NUMBER] locations. Current forecast accuracy (MAPE) is [PERCENTAGE]% and the target is [TARGET]%. The business experiences [SEASONALITY PATTERN] and demand is influenced by [KEY DRIVERS: promotions, weather, economic conditions, etc.]. ## TASK ### Data Requirements - Historical demand data: minimum 2-3 years, ideally 5+ years - Granularity: daily, weekly, monthly — matched to planning horizon - Clean demand vs. actual sales: account for stockouts, substitutions, returns - Promotional calendar: past and planned promotions with type and magnitude - Pricing history: regular and promotional pricing - External data: weather, economic indicators, competitor actions, social media trends - New product data: analog products, pre-launch indicators, market research - Channel data: sell-through vs. sell-in, POS data, e-commerce analytics ### Forecasting Methods Selection Choose and configure methods based on your data and business needs: Statistical Methods: - Moving averages: simple, weighted, exponential smoothing for stable demand - Holt-Winters: for data with trend and seasonality - ARIMA/SARIMA: for complex time series with auto-correlation - Croston method: for intermittent/lumpy demand patterns - Regression: for demand driven by identifiable causal factors Machine Learning Methods: - XGBoost/LightGBM: for feature-rich datasets with many demand drivers - LSTM neural networks: for complex sequential patterns - Prophet (Meta): for business time series with holidays and events - Ensemble methods: combine multiple models for improved accuracy - When ML adds value vs. when simple methods suffice ### Forecast Hierarchy - Bottom-up: forecast at SKU-location level, aggregate upward - Top-down: forecast at category/total level, disaggregate downward - Middle-out: forecast at product family level, reconcile up and down - Reconciliation: ensure forecasts are consistent across hierarchy levels - Choose approach based on data availability and business needs ### Forecast Process Design - Rolling forecast cadence: weekly, monthly, or event-triggered updates - Consensus planning: statistical forecast + sales input + marketing input - Forecast value added (FVA) analysis: does human override improve or worsen accuracy? - Exception-based review: focus attention on high-value and high-variance items - New product introduction: analog-based forecasting and ramp-up curves - End-of-life management: detecting and planning for declining demand - Promotional uplift modeling: separate baseline from promotional demand ### Accuracy Measurement - Metrics: MAPE, WMAPE, bias, tracking signal, forecast accuracy by bucket - Segmentation: measure accuracy by product category, volume tier, variability - Bias detection: systematic over-forecasting or under-forecasting trends - Accuracy benchmarks: what is achievable given your demand variability? - Lag analysis: is forecast accuracy degrading as horizon extends? - Root cause analysis for large forecast misses ### Integration with Planning - Safety stock optimization: use forecast error to set statistical safety stock - Production planning: convert demand forecast into production schedules - Procurement planning: supplier lead time + forecast = order timing - Financial planning: demand forecast feeds revenue and cost projections - S&OP integration: demand plan as input to Sales and Operations Planning ### Continuous Improvement - Model retraining schedule and trigger criteria - Feature importance monitoring: are demand drivers changing? - Forecast accuracy reporting dashboard - Monthly forecast review meeting agenda - Annual model competition: test new methods against current approach - Technology roadmap: tools and infrastructure investment plan
Or press ⌘C to copy
Replace these placeholders with your own content before using the prompt.
[COMPANY][NUMBER][PERCENTAGE][TARGET][SEASONALITY PATTERN]