A Demand Forecasting system lives and dies by how well the system implementation is handled. As the trite saying goes, “the devil is in the details,” which most certainly applies to a forecasting project. Plus, we mustn’t forget that forecasting is an inexact science and requires a degree of mental interpolation (and a small amount of faith!) in the results.
A capable demand forecasting solution should meet key criteria such as integration with in-house software systems, ease of use, automation features and ability to scale. Once selected, the task at hand is to make the solution fit the needs of your business. Unfortunately, many projects start off the wrong way by trying to fit the business into the framework of the forecasting solution itself.
Once you’re past that hurdle, the real job is to insure a successful implementation. In most cases it’s advisable to employ an outside resource who knows the software and understands how to adapt it to your company’s requirements. Then, you should look at these key factors to help you successfully implement your automated forecasting system.
Expectations, Data and Preparation
First and foremost, you need to have appropriate EXPECTATIONS about the results of the implementation. It is important to understand the tradeoffs of a very detailed forecast vs. the expected accuracy. I find it can be difficult to manage the expectations of those people who “consume” forecasts relative to accuracy and detail (attributes). While the expectation that results will be more accurate and more detailed than in the past is generally true once the system has been implemented, the headwind that’s created after a major forecasting system investment often leads line managers to expect too much in terms of actual results.
The second factor is DATA, which represents true demand for your products — not a mixture of demand and supply chain efficiency. Simply stated, that means customer requirements (orders) without intercompany transfers. It does not mean shipments, which are sometimes a poor reaction to customer orders due to supply chain problems. Equally important is the level of detail (depth or attributes) in the data required for use by line managers. More attributes (ship-to, warehouse, etc.) means less accuracy. These are key tradeoffs that require knowledge and experience with your forecasting system design.
PREPARATION of the data used as input to the demand forecasting system is also critical. Once the data’s been selected, you need to consider special handling of lumpy and intermittent demand products, along with the number of attributes (dimensions). Of course, the upfront handing and removal of outliers which can skew the resulting forecast needs to be taken care of, as well. Careful consideration must also be given to the amount of historical demand history used as input into the forecasting engine. Too little can eliminate the ability to determine long-range trends. Too much can provide added influence during unrealistic periods of history. Once again, knowledge of the software combined with a good understanding of the business is key.
In the end and if handled correctly, the implementation of a demand forecasting system should focus on consumer (management) expectations, data selection and data cleansing. The goal in mind should be to improve both forecast accuracy and consistency and to remove human error through automation. Just be patient during the startup process and don’t expect magic from your historical data!
This post was written by John Hughes