A forecast built only on your sales history assumes the future will resemble it. Often that is true, until it matters most.
What is demand forecasting?
Demand forecasting turns past data and known drivers into an estimate of future demand. It feeds purchasing, production, staffing and marketing budgets.
It is not the same as planning. A forecast states what is likely. A plan states what you will do about it. Mixing the two turns a forecast into a target, and it stops telling the truth.
For midsize consumer goods brands, manufacturers and retailers, a forecast supports a concrete choice. It helps decide where to focus people, budget and capacity next.
How does a forecast differ from realised sales?
A forecast states how demand is expected to develop. Realised sales state what was actually bought. Sales also reflect stock, price, promotions and distribution, so they are not demand itself.
| Measure | What it says | When you know it |
|---|---|---|
| Orientation | What people look into before they choose | Before the sale |
| Demand development | How that orientation changes over time | Before the sale |
| Demand forecast | How demand is expected to develop | Ahead of the period |
| Realised sales | What was actually bought | After the sale |
Keep the four apart in every report. A rising demand forecast is a reason to prepare, not a booked result.
How do you build a forecast?
Forecasting: Principles and Practice names five steps. Define the problem. Gather data and expertise. Explore the data. Choose and fit models. Use the forecast and evaluate it.
The textbook says to always start by graphing the data. Look for trends, seasonality, outliers and relationships between variables.
It also says to compare two or three models, not to pick one upfront. And you can only properly judge a forecast once the actual data has arrived.
Which forecasting methods should you know?
| Method | How it works | Use it for |
|---|---|---|
| Naive / seasonal naive | Repeats the last value or last season | The benchmark to beat |
| Exponential smoothing | Weights recent data more heavily | Stable series with trend and season |
| ARIMA | Models patterns in the series itself | Single series with enough history |
| Regression with drivers | Adds price, promotions and external signals | When you know what moves demand |
| Machine learning | Learns across many series at once | Large assortments with rich data |
Start simple. A complex model that barely beats a seasonal naive forecast is not earning its cost.
How do you measure forecast accuracy?
Most teams report MAPE, the mean absolute percentage error. It is easy to explain and easy to mislead with. Use it alongside a measure that compares you with a benchmark.
Hyndman and Athanasopoulos list the flaws. Percentage errors become infinite or undefined when actual demand is zero. They produce extreme values near zero, which is common for slow movers.
They also penalise negative errors more heavily than positive ones. MASE is the proposed alternative. It scales your error against a naive benchmark. A value below 1 means you beat that benchmark on the training data.
Also separate level from direction. A forecast can miss the exact volume and still call the turn correctly. For a marketing or portfolio decision, direction often matters more.
A single accuracy percentage means little without a definition. Ask if it covers volume or direction, over what horizon and against which benchmark.
Also ask which level the forecast covers. Trendata's demand forecasts run up to 12 months ahead, at Category, Type and Specification level. They describe expected market demand, not sales or stock per SKU.
Why do external signals matter?
Sales data records the end of a decision. When a market turns, your history shows it last. External signals sit earlier in the chain.
The research supports this, with limits. Choi and Varian showed search data can improve estimates of current activity, from car sales to retail and travel. Les Binet found share of search led market share by up to a year in cars.
Choi and Varian frame their work as predicting the present. They explicitly do not claim that search predicts the future.
Lead times also differ by category. The IPA reports up to a year for cars, and other categories run shorter. Test the lead in your own category before you rely on it.
Sales history tells you the shape of the last wave. It cannot tell you the next one has started.
Operations teams add these signals to forecasts rather than replacing them. Keep what works and test if the extra signal improves it.
Frequently asked questions
Is a demand forecast the same as a sales forecast?
No. A demand forecast describes how market demand is expected to develop. A sales forecast also reflects your own stock, price and distribution.
What is the difference between demand forecasting and demand planning?
A forecast estimates what demand will be. Demand planning decides how the company responds, including stock, capacity and promotions.
What is demand sensing?
Demand sensing uses very recent data to adjust short-term forecasts, usually days to weeks ahead. Forecasting covers longer horizons.
How far ahead can you forecast demand?
It depends on the category and the use. Operational forecasts often cover weeks. Strategic forecasts cover a year or more, with lower precision.
Sources
- The basic steps in a forecasting task (Forecasting: Principles and Practice, 3rd ed.)
- Evaluating point forecast accuracy (Hyndman and Athanasopoulos, Forecasting: Principles and Practice, 3rd ed.)
- Predicting the Present with Google Trends (Choi and Varian, Google Research)
- Binet presents fast, cheap, predictive Share of Search metric (IPA, 2020)