General enquiries :
+44 (0)20 7602 6000

The secret to fool-proof demand forecasting

Tuesday 3 December 2019 Data Insight & Analytics

John Tansley's picture
By John Tansley

Demand forecasting is one of the most complex types of forecasting. Whether it’s forecasting calls, sales, footfall, or webchat, the best forecasts must take into account both historical data and expert knowledge. Managing this process in order to keep it both robust and transparent is key. In this post we assess a best practice approach that supports the production of trusted and accurate forecasts.

The difficulties of demand forecasting 

A multitude of different factors affect forecasts from day to day, meaning that no two weeks are ever completely alike, and unexpected events may well be quite frequent. As a result, it becomes very important to have a robust forecasting process, one that takes into account all the information available and is flexible enough to handle the unforeseen.

Compared with more traditional analytics such as churn prediction or cross-sell modelling, the major difficulty introduced by demand modelling is the fact that nothing is static. We can no longer create a single model from a single one-off dataset. The forecasting process needs to be far more fluid, with datasets, models, assumptions, and business rules changing from day to day. When things like snow days, technical issues, or competitor behaviour are introduced into the mix, forecasting can feel like a permanent race to stay on top of things.

A best practice approach to demand forecasting 

One core approach that can really help manage the demand planning process is to separate the components of the forecast into a data driven component, and a business knowledge component. We can do this through maintaining a very clear distinction between the sources of information that go into any forecast model. Forecasting, after all, is simply the process of making the best possible guess of future behaviour, through taking into account all available information. 

Typically, information comes in two distinct forms: 

  1. ‘Hard’ data in the form of known historical data or known demand curves, and 
  2. Business knowledge, which may also be backed by numbers, but is typically more of a scenario than a completely known quantity. 

Separating information in this way - data versus business knowledge - will streamline the forecasting process, by making it very clear which components of the forecast are purely data-driven, and which are more subjective and subject to change.

Why separate data driven and business knowledge components?

One major advantage of this approach is that it makes it easy to apply machine learning to the data part using models such as regression or random forests - or Excel based profiles at a pinch! These forecasts will not be perfect, but, if well designed, will provide the best possible forecasts given only historical data. 

Business knowledge can then be overlaid over the top of these purely data-driven forecasts, in order to include additional information that cannot be inferred from purely the data: things like new marketing plans, pricing changes, or customer communications.

Keeping forecasts and targets separate

This approach also helps avoid the cardinal sin of confusing and combining forecasts and targets: forecasts should be the most likely expected outcome, whereas targets are the desired outcome.

Confusing these two very separate quantities can make tracking accuracy very difficult and can in fact make planning more difficult. The initial data driven forecast should be purely based on (hopefully) robust historical data, whereas the business overlays may include a range of possible effects. In fact, different scenarios of business overlays may be applied to the same raw basic forecast in order to model a range of possible or desired outcomes.

Separating forecasts into a data driven and a business knowledge component helps to produce the best possible forecasts: those that take into account both historical data and business knowledge and are transparent and justifiable throughout.

Learn about how CACI could help with your demand forecasting or about our demand forecasting software, CACI Forecaster, by getting in touch with a member of our team.

Demand forecasting is one of the most complex types of forecasting. In this post we assess a best practice approach that supports the production of trusted and accurate forecasts.

The secret to fool-proof demand forecasting