As demand planners, we do something extraordinary: we predict the future.
Forecasting is a blend of data, intuition, and business context. It’s both an art and a science. Choosing the right method can mean the difference between stockouts and smooth operations, between bloated inventory and optimized flow.
This is not to scare you. It’s to show the relevance of the work you do. Your role directly shapes decisions across supply, finance, sales, and beyond.
Some time ago, I posted an infographic on LinkedIn about seven forecasting techniques and the response was GREAT! So in this issue of the newsletter, I’m going bigger.
Here, you’ll find 21 forecasting techniques, each with their pros, cons, and when to use them. All in one place to help you sharpen your toolkit and make smarter, more confident planning decisions.
Let’s dive in and elevate the way we plan the future.
Basic Statistical Methods
1. Naive Forecast
“Tomorrow = Today”
Pros:
– Smooths out random demand fluctuations
– Easy to implement and understand
Cons:
– Lags behind demand trends and shifts
– Poor at capturing seasonality or sudden changes
When to use: For very stable, low-variability SKUs with flat demand and no seasonality.
2. Moving Average
Calculates average demand over a fixed window (e.g., last 3 months). Smooths noise but lags behind trends or seasonality.
Pros:
– Smooths out random demand fluctuations
– Easy to implement and understand
Cons:
– Lags behind demand trends and shifts
– Poor at capturing seasonality or sudden changes
When to use: For moderately stable SKUs where the goal is to reduce noise and short-term variability.
3. Weighted Moving Average
Gives more weight to recent periods. Useful when recent trends are more relevant than older data.
Pros:
– Responds better to recent demand changes than simple averages
– More flexible by adjusting weights based on business needs
Cons:
– Still lags behind rapid shifts or seasonality
– Requires judgment or testing to set appropriate weights
When to use: For SKUs with mild trends where recent demand is more indicative of future sales.
4. Simple Exponential Smoothing (SES)
Forecasts using a smoothing constant (alpha) to weight recent demand more heavily. Best for flat, non-seasonal data.
Pros:
– Reacts faster to demand changes than moving averages
– Requires minimal data and is easy to automate
Cons:
– Not suitable for seasonal or trending data
– Forecast quality depends heavily on the choice of alpha
When to use: For SKUs with flat, non-seasonal demand where recent values are more relevant.
5. Holt’s Linear Trend Method
Builds trend into exponential smoothing. Suitable for items with consistent upward or downward movement.
Pros:
– Captures both level and trend in demand
– More accurate than simpler methods for trending SKUs
Cons:
– Struggles with seasonal patterns
– Sensitive to parameter settings, which may require tuning
When to use: For SKUs with a clear, consistent upward or downward demand trend.
6. Holt-Winters (Triple Exponential Smoothing)
Adds seasonality on top of level and trend—ideal for seasonal SKUs like beverages, fashion, or school supplies.
Pros:
– Captures level, trend, and seasonality in one model
– Effective for predictable, repeating seasonal patterns
Cons:
– Requires more data and parameter tuning
– Can overfit if seasonality is irregular or inconsistent
When to use: For SKUs with consistent seasonal demand patterns combined with trend and level changes.
Causal & Regression Techniques
7. Linear Regression
Finds a straight-line relationship between a dependent variable (e.g., sales) and one independent factor (e.g., price or marketing).
Pros:
– Useful when demand is clearly influenced by another variable
– Provides interpretable insights into driver-impact relationships
Cons:
– Limited to linear relationships
– Can mislead if key influencing factors are omitted or correlated
When to use: For SKUs where demand is driven by a single, quantifiable external factor like price, promotion, or ad spend.
8. Multiple Linear Regression
Accounts for several demand drivers at once (promotions, discounts, weather). Good for mature categories with complex dynamics.
Pros:
– Captures the impact of multiple factors on demand
– Helps explain complex, real-world demand behavior
Cons:
– Requires large datasets and careful variable selection
– Prone to overfitting and multicollinearity issues
When to use: For SKUs in mature categories where demand depends on several known external drivers.
9. ARIMA (AutoRegressive Integrated Moving Average)
Great for time-series data with trends and autocorrelation. Needs clean, stationary data to work well.
Pros:
– Captures patterns like autocorrelation and underlying trends
– Strong statistical foundation for univariate time-series forecasting
Cons:
– Requires data preprocessing (e.g., differencing, stationarity checks)
– Complex to tune and interpret for non-experts
When to use: For SKUs with historical demand patterns showing trends and correlations over time, especially in non-seasonal data.
10. SARIMA (Seasonal ARIMA)
Adds a seasonal component to ARIMA; helpful when monthly or quarterly patterns repeat reliably.
Pros:
– Models both seasonality and autocorrelation effectively
– Suitable for complex seasonal time-series data
Cons:
– Requires careful parameter tuning and data transformation
– Computationally intensive and harder to interpret
When to use: For SKUs with strong, consistent seasonal patterns and time-based correlations.
11. Transfer Function Models
Combine ARIMA with external input variables (e.g., advertising spend or GDP). Useful for planning with known economic factors.
Pros:
– Captures both internal time-series patterns and external influences
– Ideal for aligning forecasts with business drivers or macroeconomic trends
Cons:
– Requires strong statistical expertise and high-quality input data
– Complex to build, validate, and maintain
When to use: For SKUs influenced by both historical demand patterns and external variables like marketing spend or economic indicators.
Machine Learning & AI Techniques
12. XGBoost / LightGBM
Powerful tree-based algorithms that handle outliers, nonlinear relationships, and multiple variables.
Pros:
– Excellent accuracy with complex, nonlinear data
– Handles missing data, outliers, and feature interactions well
Cons:
– Requires large datasets and careful tuning
– Harder to interpret than simpler models
When to use: For SKUs with complex demand patterns influenced by many variables, especially when high accuracy is critical.
13. Random Forest
Builds multiple decision trees and averages the outputs. Reduces overfitting and works well with many predictors.
Random Forest
Builds multiple decision trees and averages the outputs. Reduces overfitting and works well with many predictors.
Pros:
– Robust to overfitting and noise
– Handles nonlinear relationships and many variables effectively
Cons:
– Less interpretable than traditional models
– Can be computationally intensive for large datasets
When to use: For SKUs with complex, nonlinear demand influenced by multiple factors, where model stability and accuracy are priorities.
14. Neural Networks
Mimics the human brain. Excellent at capturing nonlinear, complex relationships. Needs lots of data and tuning.
Pros:
– Extremely powerful for modeling complex, nonlinear demand patterns
– Learns from data without needing explicit rules or assumptions
Cons:
– Requires large datasets and significant computational resources
– Acts as a “black box” – difficult to interpret or explain
When to use: For high-volume SKUs with rich historical data and complex demand drivers, where prediction accuracy matters more than model transparency.
15. Prophet (by Meta/Facebook)
Designed for business users. Automatically detects trends, holidays, and seasonality. Works well with daily/weekly sales data.
Pros:
– Easy to use with minimal tuning and solid default settings
– Handles holidays, seasonality, and missing data automatically
Cons:
– Less accurate for highly volatile or complex datasets
– Limited flexibility compared to advanced machine learning models
When to use: For SKUs with clear seasonality or holiday effects, especially when ease of use and quick deployment are priorities.
16. LSTM (Long Short-Term Memory Networks)
A type of deep learning specifically for sequences. Excellent at modeling long-term dependencies in time series.
Pros:
– Captures long-term patterns and complex temporal relationships
– Handles irregular time intervals and lag effects well
Cons:
– Requires large datasets and significant computational power
– Difficult to interpret and tune, especially for non-experts
When to use: For SKUs with long-range dependencies or complex sequential behavior, especially in high-frequency, high-volume environments.
17. Support Vector Regression (SVR)
Effective for high-dimensional, noisy datasets. Less popular than others, but still powerful in niche applications.
Pros:
– Works well in high-dimensional spaces and with small datasets
– Robust to outliers and overfitting with proper kernel selection
Cons:
– Slower to train on large datasets
– Requires tuning and is harder to scale for large forecasting problems
When to use: For SKUs with limited data but complex relationships, or in niche cases where other models underperform.
Judgmental & Collaborative Methods
18. Expert Judgment
Relies on domain knowledge when data is unreliable or missing (e.g., for new products or crisis situations).
Pros:
– Useful when no historical data exists or data is highly unreliable
– Leverages human intuition and market context
Cons:
– Subjective and prone to bias
– Not scalable or repeatable without clear documentation
When to use: For new product launches, major disruptions, or when data is too sparse or noisy for statistical models.
19. Delphi Method
Structured technique using rounds of anonymous expert feedback until consensus is reached. Great for strategic forecasts.
Pros:
– Reduces individual bias through anonymous group input
– Useful for long-term, strategic, or highly uncertain forecasts
Cons:
– Time-consuming and requires coordination
– Results can vary depending on expert selection and facilitation
When to use: For strategic planning, long-term forecasts, or complex scenarios where data is limited and expert consensus is valuable.
20. Sales Force Composite
Forecasts are built bottom-up from sales reps in the field. Reflects customer insights, but can be biased or overly optimistic.
Pros:
– Leverages frontline customer knowledge and market conditions
– Encourages sales team ownership and accountability
Cons:
– Prone to optimism bias or sandbagging
– Inconsistent methods and quality across reps
When to use: For short-term forecasts in B2B or direct-sales environments where sales teams have strong customer relationships and market visibility.
21. Consensus Forecasting
Final demand plan formed through cross-functional alignment (demand, supply, finance) in the S&OP process. Ensures organizational buy-in.
Pros:
– Aligns all departments around a single forecast
– Improves accountability, collaboration, and decision-making
Cons:
– Time-consuming and requires effective facilitation
– May involve compromises that reduce forecast accuracy
When to use: For monthly planning cycles where cross-functional agreement and alignment with financial goals are critical.
