Demand Forecasting Methods

Why This Matters

Demand forecasting sits at the heart of every supply chain decision you'll encounter in this course. Whether you're calculating safety stock levels, planning production schedules, or negotiating supplier contracts, the accuracy of your demand predictions determines success or failure. You're being tested on your ability to select the right method for the right situation—understanding when historical patterns are reliable versus when you need expert judgment or external data to make informed predictions.

The methods in this guide fall into distinct categories based on their underlying logic: some rely purely on past demand patterns, others incorporate causal relationships with external factors, and still others tap into human expertise when data alone isn't enough. Don't just memorize method names—know what type of data each method requires, what assumptions it makes, and when it performs best. This conceptual understanding is what separates strong exam answers from mediocre ones.

---

Quantitative Methods: Mining Historical Patterns

These methods assume that past demand behavior contains useful signals about future demand. They work best when historical data is plentiful and the underlying demand drivers remain relatively stable.

Time Series Analysis

• Foundation of quantitative forecasting—examines data points collected at regular intervals to detect patterns, trends, and seasonal cycles
• Assumes pattern continuity, meaning historical relationships will persist into the forecast period
• Best for stable demand environments where external disruptions are minimal and past behavior is a reliable predictor

Moving Average

• Smoothing technique that averages a fixed number of recent periods (e.g., 3-month or 12-month) to filter out random noise
• Equal weighting gives all included periods the same importance, which can lag behind sudden demand shifts
• Ideal for stable, short-term forecasts where demand fluctuates around a consistent mean without strong trends

Exponential Smoothing

• Weighted averaging that assigns exponentially decreasing weights to older observations—recent data matters more
• Smoothing constant ($\alpha$) controls responsiveness: higher values react faster to changes, lower values provide more stability
• Computationally efficient and adaptable, making it popular for real-time inventory systems and automated replenishment

Trend Projection

• Extrapolates historical growth or decline into future periods using linear or nonlinear models
• Long-term planning tool that works when underlying demand drivers (population growth, market expansion) remain consistent
• Vulnerable to structural breaks—economic shocks or market disruptions can invalidate trend assumptions

---

Causal Methods: Connecting Demand to Its Drivers

These methods go beyond historical patterns to identify why demand changes. They require additional data on explanatory variables but provide deeper insight into demand behavior.

Regression Analysis

• Statistical relationship modeling that quantifies how independent variables (price, advertising spend, GDP) affect demand
• Multiple regression incorporates several factors simultaneously, revealing which drivers have the strongest impact
• Predictive power depends on variable selection—including irrelevant factors or omitting key drivers weakens forecast accuracy

Causal Methods (General Framework)

• Cause-and-effect focus identifies specific external factors—promotions, competitor actions, weather, economic indicators—that drive demand changes
• Requires robust data collection on both demand outcomes and potential causal variables, increasing complexity and cost
• Superior explanatory power helps managers understand why forecasts change, not just what the forecast is

---

Qualitative Methods: Leveraging Human Judgment

When historical data is scarce, unreliable, or facing unprecedented disruption, these methods tap into expert knowledge and market intelligence to generate forecasts.

Delphi Method

• Structured expert consensus process that collects independent forecasts from multiple experts, then shares anonymized results for iterative refinement
• Reduces bias and groupthink by preventing dominant personalities from influencing others before opinions are formed
• Best for new products, emerging markets, or disruptive scenarios where historical data provides little guidance

Market Research

• Primary data collection through surveys, focus groups, and customer interviews to understand preferences and purchase intentions
• Qualitative insights reveal the "why" behind consumer behavior that quantitative data alone cannot capture
• Complements quantitative methods by validating statistical forecasts against actual customer sentiment and market conditions

---

Collaborative and Technology-Driven Methods

These approaches leverage partnerships and advanced analytics to improve forecast accuracy beyond what any single method or organization can achieve alone.

Collaborative Planning, Forecasting, and Replenishment (CPFR)

• Cross-organizational information sharing where retailers, manufacturers, and suppliers jointly develop forecasts using combined data
• Reduces bullwhip effect by replacing independent, often conflicting forecasts with a single shared demand signal
• Requires trust and technology infrastructure—partners must share sensitive data and align on common planning processes

Artificial Intelligence and Machine Learning Techniques

• Pattern recognition at scale using algorithms that detect complex, nonlinear relationships across massive datasets
• Continuous learning capability allows models to improve automatically as new data arrives, adapting to changing conditions
• Enhances traditional methods by incorporating real-time signals (social media sentiment, web traffic, weather) that conventional approaches miss

---

Quick Reference Table

---

Self-Check Questions

1. Which two forecasting methods both use historical demand data but differ in how they weight past observations? Explain when you'd choose one over the other.

2. A company is launching an entirely new product category with no historical sales data. Which forecasting methods would be most appropriate, and why?

3. Compare and contrast regression analysis and time series analysis. Under what demand conditions would regression provide significantly better forecasts?

4. If an FRQ describes a supply chain suffering from the bullwhip effect due to misaligned forecasts between a retailer and manufacturer, which forecasting approach directly addresses this problem?

5. A forecasting team wants to incorporate real-time social media sentiment and weather data into their predictions. Which method category enables this capability, and what advantage does it offer over traditional quantitative methods?