Forecasting Process for Demand Planning
The first step in developing a forecasting process is necessarily to consider the main reasons for producing a forecast. Every company is different, but some of the reasons for needing an accurate demand forecast are likely to relate to S&OP, production planning, inventory management, sales planning, warehouse capacity, workforce and financial planning.
Early decisions are needed on some key components in the demand forecasting process including the historical data to be used, time periods, time horizons and levels of detail for products / customers. Forecasting methods may include time series forecasting, causal modelling and, of course, commercial judgement.
Some of the key steps in a typical demand forecasting process:
Forecast Solutions can help with all aspects in the development of a successful forecasting process.
Time Period and Time Horizons
The majority of businesses use month or week as the basis for forecasting. Monthly data is adequate for many companies and has the benefit of being sufficiently aggregated in many instances to facilitate the analysis that is needed in forecasting. There may sometimes be a need to adjust the historical data before forecasting to take account of unequal period lengths.
Weekly forecasting may be preferable when there is an active promotions programme or if reconciliation with weekly customer's forecast is needed. The expectation that a weekly forecast will be more accurate than monthly, however, is sometimes not realised due to the greater volatility in weekly data and complications in seasonal analysis.
There may be a need for daily forecasting, for example for products with a very short shelf life or for the forecasting of daily activity at call centres. With daily forecasting there is a need to cope with profiles of day within week as well as weekly seasonality and there are a number of different approaches that can be used in this somewhat difficult task.
Statistical Forecasting Methods
Statistical forecasting methods are usually valuable when a consistent demand history of two years or more is available. The most commonly used family of methods is referred to as time series forecasting. Here, the historical data is analysed to identify the seasonal pattern and trend that applies to the entity being forecast and these characteristics are extrapolated forward to create a forecast. Time series forecasting makes the assumption that overall environment will stay constant.
Another family of statistical forecasting methods is called causal modelling, where analysis is conducted of the company's demand history together with historical data on one or more suspected causal factors such as weather, price or economic indicators. Statistically valid relationships between demand for products or services and one or more causal factors may be found. These may alone be the basis for a forecasting solution or can sometimes be combined with a time series forecasting method.
Application of Commercial Judgement
Whether or not statistical forecasting methods are used for forecasting it is always valid to check the forecast with commercial judgement. If adjustments are made, it is good to have clarity as to which individuals or groups have access to suggest or make changes and ideally the changes should be documented. With forecast amendments the emphasis should be on important and significant differences, rather than trying to review everything.
Sometimes there may be different points of view on certain aspects of the forecast, or sometimes there may be a complete different version of the forecast. Such forecasts may have been produced by a sales team or possibly by a major customer. A reconciliation process may be needed, whether it be based on a mathematical approach or through a regular meeting of individuals. If there is a S&OP process in place the reconciliation may well be covered as part of that.
Forecast Accuracy Measurement
A major part of any demand forecasting process is the achievement of the best possible levels of accuracy, therefore forecast accuracy should be measured at the start of every forecasting cycle. All businesses are different in terms of the levels forecast accuracy that are possible and targets should be challenging, but achievable.