Short Term Demand Forecasting
Short term demand forecasting is carried out in support of stock planning and/or production planning, sometimes also for account planning. It is generally assumed that company resources and the prevailing competitive structure and economic background will remain constant over that short horizon. It usually needs to be carried out at a very detailed level.
The phrase 'short term' has different meanings for different types of business. Typically with short-term forecasting we are thinking somewhere between two months and 6 months, but for a pharmaceutical company it might be a lot longer, or for a FMCG manufacturer with short shelf life products it might be much shorter. Sometimes it's useful to define 'short term' based on the longest lead time for the supply of products, ingredients or components.
The requirements for short-term forecasting are likely to be different to those in the medium term forecasting that is needed for budgeting or sales and operations planning. Short term forecasting may be carried out using monthly, weekly or daily time buckets and the considerations vary somewhat in these different cases.
In short term forecasting it is often necessary to take account of seasonality, promotional effects and sometimes causal factors such as price or weather. There is a need to keep track of changes in the level of demand, but because of the short horizon, changes can be followed with simple plus or minus adjustments and/or a straight line trend. There is generally no need to fit complicated curves.
When the number of items to be forecast is small Forecast Solutions may be able to provide a forecasting system using Excel and for more complex needs we can help in a software selection process.
Monthly Demand Forecasting
Monthly forecasting is used to good effect in many companies and is often a relatively easy and straightforward way of working as compared to making use of shorter time buckets.
There is often a need to deal with seasonality. Proprietary forecasting software usually contains adequate facilities for analysing monthly seasonality and incorporating it into the forecast. If working in Excel there are a couple of simple methods that can be implemented quite easily.
If promotions are a feature of the business then effective promotional planning is vital in terms of cleansing the historical data prior to forecasting and building the future promotional plan into the forecast. Monthly time buckets may be somewhat long for promotional planning, this being a major reason for many FMCG companies to adopt weekly rather than monthly short term forecasting.
In terms of forecasting methods, simple time series techniques such as moving averages and the family of exponential smoothing methods are generally reliable. If upward or downward trends in demand are taking place, straight line projection is usually adequate.
Causal modelling (cause and effect analysis) can sometimes be useful in monthly forecasting to reflect the effects of factors such as relative prices or unseasonal weather. It is usually carried out at an aggregated level of detail, therefore may require a top-down process if used within a short term forecasting process.
Weekly Sales Forecasting
Forecasting in weekly time buckets is somewhat more difficult than with monthly data, but is sometimes necessary for a number of specific reasons. For example many FMCG companies may need to carry out promotional planning in weekly time buckets and may receive customer forecasts or EPOS data in weekly form.
Seasonal analysis of weekly sales data is more tricky than with monthly data, one reason being that there are not exactly 52 weeks per year. As a result it is more likely that bank holidays and other key annual events will fall in different week numbers in different years. This may necessitate dealing with such matters through event planning rather than through a simple seasonal analysis. Extensive historical data cleansing may be needed.
There is much greater volatility in weekly data. Even if a seasonal analysis is conducted using several years of sales history, the resulting seasonal pattern often shows a ragged pattern that is not credible for short term forecasting. It often helps to calculate seasonal profiles at an aggregated level (the 'group seasonal indices' approach) and there are a number of other methods that can be employed. See our page on seasonality in forecasting for further discussion on this.
Forecasting methods can be similar to those referred to above, but there are some recent developments in techniques that can help address the issues with seasonal analysis. Causal analysis can be useful, subject to the same qualifications mentioned above as for monthly forecasting.
Daily Demand Forecasting
With daily sales forecasting even lower volumes per time period, high volatility and intermittent sales are usually seen. There is usually a strong day within week cycle that dominates and may overwhelm other factors.
A relatively straightforward approach to daily sales forecasting is to carry out the seasonal analysis and forecast on a monthly or weekly basis, then split the forecast down to days using estimated daily splits. Alternatively, there are now some sophisticated forecasting methods that can attempt to model multiple cycles such as day within week and week within year.
Causal modelling may be possible with daily forecasting. Making use of a weather sensitivity analysis may be even more helpful than with monthly or weekly data, because the possibility of obtaining an accurate weather forecast covering a few days ahead is much greater than with a few months ahead.