Demand forecasting is the process of predicting future demand for a product using historical sales data, trend analysis, and contextual factors — producing an estimate that informs replenishment alongside usable stock, incoming supply, lead time and supplier constraints.
Quick answers
What is demand forecasting? Demand forecasting uses historical sales patterns to predict how much of each item you will sell in a future period. The forecast feeds directly into reorder point calculations, safety stock sizing, and purchase order timing. Confirmed customer orders and operational requirements can also drive buying; a statistical forecast is one input to the decision.
What methods can a small team evaluate? Start with a naïve forecast (the last observed value), a moving average and simple exponential smoothing. For seasonal demand, include a seasonal naïve baseline. Intermittent-demand methods such as SBA are candidates when demand has many zeros. Select using relevant held-out data and purchasing outcomes, not a universal method rule.
How accurate does my forecast need to be? Exact prediction is impossible. What matters is measuring the error. Evaluate on held-out periods against a simple baseline. MAPE is undefined when actual demand is zero and can be unstable near zero, so it is often unsuitable for intermittent demand. Choose a compatible error measure and assess shortages and stock alongside it. See forecast accuracy evaluation.
What about seasonality? Compare equivalent periods across repeated cycles, accounting for promotions, availability and changes in the business. A seasonal index can be useful when that pattern is stable; a few rising weeks do not establish seasonality.
The formulas
Simple moving average:
forecast = (D_1 + D_2 + ... + D_n) / n
where D_i is demand in period i and n is the chosen history window. Include zero-demand periods and distinguish them from missing records or stockouts.
Exponential smoothing:
forecast_t = α × D_(t-1) + (1 − α) × forecast_(t-1)
where α (alpha) is the smoothing factor between 0 and 1. Higher α reacts faster to recent changes; lower α produces a more stable forecast. Choose α using a documented fitting or validation procedure; the example below uses 0.3 for illustration.
Forecast accuracy (MAPE):
MAPE = (1/n) × Σ |actual − forecast| / actual × 100
Demand patterns are a starting point for evaluation
ADI and CV² describe demand spacing and variability. The conventional SBC boundaries of ADI 1.32 and CV² 0.49 help categorize a history, but they do not prove that one forecast method will work best for your future buying horizon. SBA corrects a bias in Croston’s method; it does not correct a universal bias in moving averages.
Use the demand pattern classifier to explore a history, then test candidate forecasts. Simple exponential smoothing estimates a level; trend and seasonal patterns may call for other models or explicit adjustments. Sparse data, new products and known events need additional judgment.
Worked example
A coffee shop sells cold brew concentrate. Weekly sales for the past 8 weeks: 24, 28, 32, 30, 26, 34, 38, 36.
4-week moving average: (26 + 34 + 38 + 36) / 4 = 33.5 units
Exponential smoothing (α = 0.3): Starting from week 5 forecast of 28.5:
- Week 6: 0.3 × 26 + 0.7 × 28.5 = 27.75
- Week 7: 0.3 × 34 + 0.7 × 27.75 = 29.625
- Week 8: 0.3 × 38 + 0.7 × 29.625 = 32.1375
- Week 9 forecast: 0.3 × 36 + 0.7 × 32.1375 = 33.29625 units, about 33.3
The two week-9 estimates are close. This example does not establish which method is more accurate: week-9 actual demand has not been observed. Weeks 6–8 are 34, 38 and 36, so they do not show a continuously accelerating trend. Compare a longer history and held-out outcomes before inferring seasonality or selecting a method.
Measure error over the purchasing horizon
Use only information available at the time each forecast would have been made. Evaluate forecasts over a horizon relevant to replenishment, such as supplier lead time plus the review interval. A strong one-day result may not translate into a useful four-week purchase plan.
MAPE requires nonzero actual demand in every evaluated observation. MAE keeps errors in the original units; scaled measures can help compare series, subject to their own denominator assumptions. Forecasting: Principles and Practice explains these measures and held-out evaluation.
For example, actual demand of 0, 10 and 20 against forecasts of 2, 8 and 18 gives absolute errors of 2, 2 and 2, so MAE is 2 units. MAPE cannot be calculated conventionally for the zero-demand observation. Silently excluding zeros changes the evaluation set.
Inspect bias and operating outcomes as well as average error. Sales during a stockout can understate demand; unexplained unit conversions can distort it. Record manual changes and compare their results with the original forecast.
Translate a forecast into a purchase decision
A forecast is not an order quantity. Combine expected requirements with the selected buffer, usable stock, allocated demand, scheduled receipts, supplier packs and dates. Avoid counting the same demand twice when confirmed orders are already represented in the forecast.
Forecast-error variation may inform a safety-stock policy, but it is not the only input. Lead-time uncertainty, service objectives and the chosen inventory model matter. Review safety stock and reorder points with those assumptions explicit.
Evaluate LineNow with your history
Bring a fast-moving item and an intermittent item to a LineNow inventory replenishment demonstration. Inspect the history window, zero-demand periods, sparse-data fallback, manual overrides and purchase recommendation. Follow the selected order through supplier confirmation and receiving to see how execution updates the next buying decision.
Validate the actual method and data flow with your items rather than assuming that an educational formula describes every product calculation.
Related
- Procurement Software for SMBs
- Purchase Order Software
- Supplier Management Software
- Syntetos-Boylan Approximation (SBA): Bias-Corrected Intermittent Demand Forecast
- Coefficient of Variation (CV²): Demand Volatility Explained
- Safety Stock: How to Size It Statistically
- Reorder Point (ROP) Formula: How to Calculate with Example