Sales Forecasting Service for Demand Planning
At Forecast Solutions we can provide various types of sales forecasting service. This can be a one-off analysis carried out in order to give guidance or a second opinion, or it could be a forecasting outsource service set up on an ongoing basis. We can work remotely, through visits to your site, or as a mixture of the two.
The price for our sales forecasting service is based on the amount of time that will be needed, so we will discuss with you the nature of what you are looking for, the availability of data to support the analysis and the scope / scale of the requirement.
Dealing with Seasonality
It is difficult to over estimate the importance, for best forecast accuracy, of dealing with seasonality in the best possible way. In some companies it is obvious that sales are seasonal and that then that gives a good starting point. If logic suggests that sales are non-seasonal, for example with everyday hair care products, it is still worth testing for the existence of seasonality, as it can often found in surprising situations.
The historical data can be tested for the presence of a seasonal pattern and, if the seasonality is statistically valid, a set of seasonal indices will be calculated for use in forecasting. If the sales history is weekly, as opposed to monthly, it may be beneficial to smooth the indices to reduce the jagged characteristic that weekly indices often display due to the residual volatility at that level of detail.
Sometimes it is beneficial to analyse seasonality on aggregated data such as at a product group level, then deploy the resulting seasonal pattern to every item within the group.
Time Series Forecasting
Time series forecasting methods provide the easiest and most common ways of producing a forecast. The only historical data needed is the monthly, weekly or daily sales figures over a good period of time, normally extending over two full years or more. Then the aim is to analyse the history to identify any seasonality and / or trend that applies, then project forward in order to create a forecast. Because time series methods only use a single type of data for both history and forecast, they are also known as 'univariate' methods.
Time series methods include straight line and curve fitting, moving averages, exponential smoothing and arima models. For most practical purposes straight line fitting and one of the exponential smoothing models will produce good and reliable results. Simple methods often prove more accurate over time as compared to very complicated solutions.
However simple or complicated the forecasting model, the question as to whether or not to employ an upward or downward trend in the forecast, can often be difficult to resolve. A lot of automated solutions routinely apply a trend in the forecast in instances where a flat projection might be better.
An alternative approach to forecasting is to analyse historical data on potential factors that may drive company sales, together with historical sales data, in order to identify a mathematical relationship between sales and the causal factor(s). Factors that impact on sales might include weather, prices or economic indices. This type of analysis may be referred to as causal analysis, cause and effect analysis or regression analysis (the mathematical method that is often used).
If it is possible to produce credible forecasts for the causal factors over a sufficiently long time period to be useful in forecasting, then it may well be useful to use the causal models as the basis for a forecasting process. Except for very short term daily forecasting it is not usually sensible to use weather forecasts in the forecasting process, but it is often useful to adjust the historical data for weather effects, prior to carrying out a time series forecasting process.
We have carried out many causal analysis projects and have sometimes combined to good effect the time series and causal approaches.
Please click here for a fuller list of Forecast Solutions consulting services.