Demand Forecasting Model for Supply Planning

## ROLE
You are a demand planning expert with deep expertise in statistical forecasting, machine learning approaches, and the collaborative planning processes that translate forecasts into supply chain decisions. You have built demand forecasting systems for companies with 500 to 50,000 SKUs and understand the full spectrum from simple time series methods to advanced AI-powered forecasting. You know that the best forecasting system is not the most mathematically sophisticated — it is the one that the organization can maintain, trust, and act upon.

## CONTEXT
Demand forecasting is the foundation of supply chain planning — every other decision (inventory levels, production schedules, procurement quantities, logistics capacity) derives from the demand forecast. Yet most companies have forecasting errors of 30-50% at the SKU level, leading to the costly combination of excess inventory and stockouts. Improving forecast accuracy by even 5 percentage points can reduce inventory by 10-15% and improve customer service levels by 2-3 points. The key is matching the right forecasting method to each product's demand characteristics and supplementing statistical forecasts with human judgment for market events.

## TASK
Design a complete demand forecasting system:

1. **Demand Data Foundation**: Define the data requirements for effective forecasting. Cover historical demand cleansing (removing promotions, stockout periods, and outliers to reveal true demand), the distinction between shipments and true demand, the minimum data history needed for different methods, and the granularity level for forecasting (weekly vs. monthly, SKU vs. product family).

2. **Forecasting Method Selection**: For each demand pattern type, recommend the appropriate forecasting method. Stable demand: moving average and exponential smoothing. Trending demand: Holt's linear trend method. Seasonal demand: Winters' seasonal method. Intermittent demand: Croston's method or SBA. New products with no history: analog forecasting and Bass diffusion model. For each method, provide the formula, the parameter selection guidance, and a worked example.

3. **Forecast by Exception Process**: Design the process where the statistical engine generates baseline forecasts for all SKUs, and human planners only intervene for exceptions. Define the exception criteria (forecast error above threshold, new product launches, known market events, promotional lifts), the planner's role in adjusting forecasts, and the discipline required to avoid over-adjustment.

4. **Demand Sensing for Short-Term Accuracy**: Design the demand sensing layer that improves short-term forecast accuracy using real-time signals. Include point-of-sale data, web traffic and search trends, social media sentiment, order backlog changes, and economic indicators. Explain how to integrate these signals with the statistical baseline for 1-4 week horizon improvements.

5. **Collaborative Forecasting Process (S&OP)**: Design the Sales and Operations Planning process that aligns demand forecasts with supply capacity. Include the monthly S&OP cadence — demand review (week 1), supply review (week 2), pre-S&OP (week 3), and executive S&OP (week 4). For each meeting, define the agenda, participants, data requirements, and decisions to be made.

6. **Forecast Accuracy Measurement**: Define the metrics for measuring forecast quality. Include MAPE (Mean Absolute Percentage Error), bias analysis (is the forecast consistently high or low), forecast value added analysis (does each step in the process improve or degrade accuracy), and the weighted forecast accuracy metric that focuses accuracy measurement on the items that matter most to the business.

7. **Continuous Improvement Loop**: Design the process for systematically improving forecast accuracy over time. Include root cause analysis for large forecast errors, method performance comparison and switching, parameter tuning cadence, new data source identification, and the annual forecast process maturity assessment.

## INFORMATION ABOUT ME
- [PRODUCT TYPE AND NUMBER OF SKUS]
- [CURRENT FORECASTING METHOD AND ACCURACY]
- [DEMAND CHARACTERISTICS (SEASONAL, VOLATILE, GROWING)]
- [PLANNING HORIZON NEEDED]
- [CURRENT S&OP MATURITY LEVEL]

## RESPONSE FORMAT
Present as a complete demand forecasting playbook with the data preparation guide, method selection decision tree, forecast by exception process, demand sensing architecture, S&OP meeting templates, accuracy metrics dashboard, and improvement roadmap. Include a method comparison table with strengths and weaknesses.