Demand forecasting: definition, methods and data needs
**Demand forecasting is the estimate of how much of a product, spare part or service will be needed in a future period. It combines past sales or consumption with known drivers such as seasonality, open orders, prices and promotions. Companies use it to plan purchasing, production and stock levels.**
What is demand forecasting? The definition in detail
The term is deliberately broad. Demand forecasting always answers the same question, namely which quantity is needed and when, and differs only in the target variable. In the past it was usually done in a spreadsheet. Today forecasting models do the calculation. They process seasonal patterns, changes in the product range or machine conditions as explanatory variables and adapt when new data arrives. In German the method is called Bedarfsprognose.
How it differs from related terms
| Term | What is predicted | Typical data | What the result is used for |
|---|---|---|---|
| Demand forecasting | quantity of a good in a future period | varies with the target variable | umbrella term for planning in purchasing, inventory and production |
| Sales forecasting | sales volume per product or product line | order history, framework contracts, seasonal patterns | production and purchasing planning |
| Material requirements planning | material required per order and date | bills of materials, open and expected orders, replenishment lead times | procurement and scheduling |
| Predictive maintenance | time of a failure or of wear | sensor data, control data, failure history | planned maintenance instead of downtime |
In practice, four terms are often mixed up. The table sorts them out.
Sales forecasting therefore often provides the input for material requirements planning. Predictive maintenance, by contrast, does not predict a demand but an event. The two meet in spare parts demand: if you know when wear is likely, you can also order the right part in time.
Forecasting vs demand forecasting
Forecasting is the general term for any estimate of a future value: revenue, energy load, machine wear or website traffic. Demand forecasting is the part of forecasting that deals with demand, meaning the quantity customers or internal consumers will need. In supply chain and production planning the two words are often used as synonyms, because demand is the value most plans start from.
Demand forecasting methods
There is no single best method. The right one depends on how regular your demand is and which drivers you know in advance.
Simple baselines: last year's value, moving averages. Every better method has to beat them, so we always calculate them first.
Exponential smoothing: weights recent periods more than older ones and handles trend and seasonality. Robust and easy to explain.
Statistical models such as ARIMA: model the structure of a time series. Good for stable series with enough history.
Machine learning with drivers: gradient boosting and similar models use prices, promotions, weather or order backlog as inputs. Strong when demand depends on known factors.
Deep learning for time series: models such as PatchTST learn patterns across many related series at once, for example thousands of items. We use this approach in Orpheon.
In practice we test several methods on the past months of your data and keep the one that beats the baseline most clearly for each group of items.
Spare parts demand forecasting
Spare parts behave differently from finished goods. Many parts are needed rarely and irregularly, with long periods of zero demand in between. Standard methods overestimate or underestimate such intermittent demand. Specialised methods such as Croston's method forecast the size of an order and the time between orders separately. Even better results come from adding the installed base: how many machines of a type are in the field, how old they are and when maintenance is planned. This turns a guess into a plan that purchasing and service can work with.
What the industry report shows for planning and supply chain
In our industry report, demand forecasting is the core of use case field C, planning and supply chain. The typical effect is inventory reduced by 10 to 50 %, and maturity is rated as medium. Individual cases show the range: Schneider Electric reports inventory reduced by 10 % on a value of around EUR 100 million (company statement, 2026), and Unilever in Hefei reports forecast accuracy improved by 39 % (WEF Global Lighthouse Network 2026). Both are documented cases from industry, not projects of ours.
The prerequisite stands out. No other field is as unforgiving of poor ERP master data. You need order and sales history covering several years including outliers, realistic replenishment lead times instead of well kept wishful values, and exactly one target variable, such as service level or days of inventory. How Orpheon covers these forecasting tasks in production is shown on the page on AI demand forecasting for manufacturing.
Practical example: three scenarios with Orpheon
With Orpheon we develop and operate our own forecasting platform. It covers three recurring tasks: sales forecasting from order and sales data, energy demand forecasting from meter data and production plans, and spare parts demand forecasting from maintenance history, machine data and bills of materials. In multi-plant operations, Orpheon calculates separately for each plant or line because patterns differ between sites. A first pilot with real data is ready in 72 hours, and productive operation follows in 4 to 10 weeks. How the platform is structured is described on our page on demand forecasting with AI. For manufacturers with machine data and bills of materials there is a dedicated page on AI demand forecasting for manufacturing.
Frequently asked questions
Sources
- WEF Global Lighthouse Network 2026, Unilever Hefei
- Schneider Electric company statement, 2026
- NexPatch AI and Alic.ai, industry report AI in Manufacturing, chapters 02 and 05, as of 08/2026