Demand Forecasting in Logistics: What It Is, Methods and How to Improve Its Accuracy
Demand forecasting is one of the most strategic processes in any supply chain. When it works well, everything flows: inventory is just right, costs are controlled and customers receive what they need when they need it. When it fails, the consequences ripple across every department: surplus stock on some products, stockouts on others, last-minute transport costs and dissatisfied customers. This article explains what it is, how to calculate it, what methods exist and how to sustainably improve its accuracy.
What Is Demand Forecasting?
Demand forecasting is the process of estimating the quantity of products or services that customers will demand over a future period. Its goal is not to predict the future with perfect accuracy, but to reduce uncertainty enough to make better decisions about purchasing, production, warehousing and transport.
In logistics, demand forecasting is the starting point for virtually every operational decision. It determines how much stock to hold, when to replenish, what transport capacity to contract and how to size warehouse resources. An error in the forecast amplifies along the entire chain.
Why Is It So Important in the Supply Chain?
Demand forecasting is not just a statistical exercise. It has a direct impact on the bottom line:
- On inventory: a forecast that is too high generates excess stock, with the associated warehousing, obsolescence and tied-up capital costs. A forecast that is too low causes stockouts and lost sales.
- On logistics costs: without a reliable forecast, companies end up booking last-minute urgent transport at rates far above negotiated prices. Well-predicted demand allows shipments to be planned in advance and routes optimised.
- On production: manufacturing plants need advance notice of production volumes to organise shifts, raw material purchases and machine capacity.
- On customer satisfaction: service level depends directly on having the product available when the customer demands it. Accurate forecasting is the foundation of a high OTIF.
Types of Demand Forecasting
Not all forecasts are alike. Depending on the time horizon and purpose, three main types are distinguished:
| Type | Horizon | Main Use |
|---|---|---|
| Short-term forecast | Days to 3 months | Inventory management, replenishment, transport planning |
| Medium-term forecast | 3 months to 1 year | Production planning, logistics resource contracting, budgeting |
| Long-term forecast | 1 to 5 years | Strategic decisions, capacity expansion, new market entry |
Demand Forecasting Methods
There are two main families of methods: quantitative, based on historical data, and qualitative, based on expert judgement. In practice, the best systems combine both.
Quantitative Methods
These are the most widely used in logistics because they allow large volumes of data to be processed systematically:
- Moving average: calculates the average of the last N periods to smooth out fluctuations. Simple and useful for stable demand, but slow to react to sudden changes.
- Exponential smoothing: similar to moving average but assigns greater weight to more recent data. Reacts better to trend changes.
- Regression models: relate demand to external variables such as price, seasonality, marketing campaigns or macroeconomic indicators.
- ARIMA models: time series analysis that identifies trend, seasonality and random components. Very accurate but require sufficient historical data and statistical knowledge.
- Machine learning and artificial intelligence: algorithms capable of processing multiple variables simultaneously and detecting non-linear patterns that traditional models cannot capture. Increasingly accessible through specialised platforms.
Qualitative Methods
Useful when there is insufficient historical data or when unpredictable external factors carry significant weight:
- Delphi method: structured consultation with a panel of experts who converge on a consensus estimate.
- Executive judgement: forecasting based on the criteria of commercial or product managers with direct market knowledge.
- Customer surveys: direct collection of purchase intentions, useful for new product launches.
How to Calculate Demand Forecasting Step by Step
Although the specific method varies by sector and available data, the general process follows these steps:
- Define the objective and time horizon: what is the forecast needed for? Weekly inventory management or quarterly production planning? The horizon determines the most appropriate method.
- Collect and clean historical data: sales data, orders, returns and any relevant variable. Data quality is the single most critical factor in the entire process.
- Identify patterns: trend (is demand growing or declining?), seasonality (are there recurring peaks?) and irregular components (were there atypical events that distort the historical data?).
- Select and apply the model: based on identified patterns and available resources, choose the most appropriate method.
- Adjust with qualitative information: incorporate business knowledge about promotions, price changes, new competitors or any known external factor.
- Measure error and review: compare the forecast with actual demand and calculate the error to iteratively improve the model.
How to Measure Forecast Accuracy: Forecast Error Metrics
A forecast without error measurement is useless. The most widely used indicators are:
| Metric | Formula | Interpretation |
|---|---|---|
| MAE (Mean Absolute Error) | Average of |Actual – Forecast| | Easy to interpret, in the same units as demand |
| MAPE (Mean Absolute Percentage Error) | Average of |Actual – Forecast| / Actual x 100 | Enables comparison across products of different volumes |
| RMSE (Root Mean Square Error) | Square root of average (Actual – Forecast)² | Penalises large errors more, useful when extreme deviations are critical |
| Bias | Average of (Forecast – Actual) | Detects whether the forecast systematically overestimates or underestimates |
Factors That Make Accurate Forecasting Difficult
The perfect forecast does not exist, but understanding the main sources of error helps manage them:
- Uncontrolled seasonality: recurring demand peaks not correctly incorporated into the model.
- Atypical events in historical data: an exceptional promotion, a stockout or a crisis that distorts the baseline data.
- Poor data quality: incomplete, duplicate or erroneous records that contaminate the model.
- Bullwhip effect: small variations in end-customer demand amplify progressively up the supply chain, generating increasingly large oscillations in supplier orders.
- Information silos: when sales, marketing and logistics work with different data and do not share information, each department’s forecasts are incompatible.
- New or history-free demand: product launches or entry into new markets lack the historical data needed to build a statistical model.
Strategies to Improve Demand Forecasting
1. Unify Data Into a Single Source
The first step to improving forecasting is ensuring all teams work with the same data. When sales, purchasing and logistics have access to a single, real-time data source, forecast quality improves immediately.
2. Incorporate External Variables Into the Model
Models based solely on sales history have a ceiling. Incorporating variables such as seasonality, planned marketing campaigns, competitor pricing or macroeconomic indicators captures factors that historical data alone cannot reflect.
3. Segment by Demand Type
Not all products behave the same way. Applying different methods according to the demand profile of each SKU (stable, seasonal, intermittent, new) significantly improves overall accuracy.
4. Implement an S&OP Process
Sales and Operations Planning (S&OP) is a collaborative process that aligns sales forecasts with the operational capacity of production, logistics and purchasing. It transforms forecasting from an isolated technical exercise into a cross-departmental consensus.
5. Leverage Artificial Intelligence
Machine learning models can simultaneously process hundreds of variables and detect patterns that traditional statistical models cannot capture. Their greatest advantage is the ability to adapt automatically when market conditions change, without needing to reconfigure the model manually.
6. Measure and Learn From Errors
Establishing a routine of periodic forecast error review by product, family and channel identifies where the model is failing and prioritises improvements. Without measurement, there is no improvement.
Frequently Asked Questions About Demand Forecasting
What is the difference between demand forecasting and demand planning?
Demand forecasting is the process of estimating what will be sold. Demand planning is a broader process that takes that forecast as a starting point and translates it into action plans for production, purchasing and logistics. The forecast is an input to planning, not its substitute.
How often should demand forecasts be updated?
It depends on the horizon and the sector. In general, short-term operational forecasts should be reviewed weekly or even daily in high-turnover sectors. Medium-term tactical forecasts are reviewed monthly in most companies.
What is the bullwhip effect and how does it affect forecasting?
The bullwhip effect describes how small variations in end-customer demand amplify progressively up the supply chain. A 5% increase in consumer demand can translate into a 30% larger order to the supplier. Sharing real-time demand data across the entire chain is the most effective way to reduce this effect.
Is it possible to forecast accurately without historical data?
For new products or markets without historical data, qualitative methods such as expert judgement, customer surveys or analogous product analysis are the only alternative. As real data accumulates, the model can incorporate statistical components that progressively improve accuracy.