1) Meaning

Demand Forecasting is the process of predicting future customer demand for a product or service using historical data, market insights, and statistical/ML models.
It helps businesses align supply with demand, reducing stockouts, overstock, and lost sales.

In supply chain management, it’s one of the most critical planning tools because almost every decision (procurement, production, staffing, logistics) depends on how much demand is expected.

2) Types of Demand Forecasting

(a) By time horizon

  • Short-term (days–weeks): inventory control, staffing, logistics.
  • Medium-term (months): production planning, budgeting.
  • Long-term (years): strategic planning, capacity expansion.

(b) By method

  1. Qualitative methods (used when data is limited):
    • Expert judgment
    • Market research
    • Delphi method
  2. Quantitative methods (data-driven):
    • Time-series models (ARIMA, SARIMA, Prophet)
    • Causal models (Regression, ARIMAX)
    • Machine Learning/Deep Learning (Random Forests, XGBoost, LSTM, Transformers)

3) Common Techniques

  1. Moving Average (Simple/Weighted): smooths out noise, good for stable demand.
  2. Exponential Smoothing (SES, Holt-Winters): gives more weight to recent demand, good for trends/seasonality.
  3. ARIMA/SARIMA: statistical models for autocorrelation and seasonality.
  4. Prophet (Meta): additive model (trend + seasonality + holidays).
  5. LSTM/Deep Learning: captures nonlinear and long-term dependencies in sequential data.
  6. Causal ML models: consider external drivers (price, marketing, weather, economic indicators).

4) Evaluation Metrics

To measure forecast accuracy:

  • MAE (Mean Absolute Error): average absolute difference between forecast and actual.
  • MSE (Mean Squared Error): penalizes larger errors more.
  • RMSE (Root MSE): interpretable in original units.
  • MAPE (Mean Absolute Percentage Error): % error relative to demand.
  • sMAPE / WAPE: variations for skewed demand data.

5) Example

  • Historical daily sales: [100, 120, 130, 110, 140]
  • Use 3-day moving average forecast:
    • Forecast for next day = (120 + 130 + 110) / 3 = 120
  • Actual sales next day = 135
  • Forecast error = 135 – 120 = +15

6) Why it matters

  • Reduces stockouts & overstock: keeps inventory balanced.
  • Improves cash flow: avoids tying up capital in excess inventory.
  • Optimizes workforce & logistics: right staff and vehicles in place.
  • Supports strategic planning: capacity, expansion, and investment decisions.
  • Drives customer satisfaction: products available when customers want them.

7) Challenges

  • Data quality issues: missing, inconsistent, or noisy data.
  • Demand volatility: promotions, seasonality, and unexpected events (e.g., COVID-19).
  • Bullwhip effect: small errors amplify upstream in supply chain.
  • New products (no history): rely more on qualitative or analog forecasting.

Bottom line:
Demand Forecasting predicts future demand so supply chains can prepare efficiently. It can be done using qualitative insights, statistical models, or machine learning, and its success is measured by forecast accuracy metrics.