Every now and again I like to address the forecasting process because, as a sourcing and procurement professional, you are often negotiating contracts against a perceived volume leverage based as much off of a forecast as it is based on historical data. In Part I we reviewed judgmental and statistical forecasts and explained why you need to balance both methodologies when generating your forecast, in Part II we addressed commodities forecasting and how you need to base it on the right data and the right factors, and in Part III I directed you to Forecast Less and Get Better Results that demonstrated that the conventional wisdom that companies need to project forecasts and plans far into the future at a highly granular level is not necessarily right. Then, in Forecast with Foresight, I pointed you to a Supply & Demand Chain Executive article on a study about re-thinking demand management that noted that active/predictive demand management is necessary for good forecasting.
Part of active/predictive demand management is good demand planning. Good demand planning involves good demand planning software, so it was nice to see the Supply Chain Digest editorial staff print a short guide on how to attack the process, even if the first two steps didn’t fully address the problem.
The process, which was still quite good, that they presented was:
- Load Historical Data and Create Master Data
Identify the key data elements that need to be considered and load them.
- Clean the Historical Data
There are almost always problems with the quality and completeness of the data loaded into the system. E.g. “demand” may not be true demand, because it is taken from “sales” data, and will not include “stock-outs”.
- Generate a Statistical Forecast for Existing Products
Use demand planning software with built in statistical models to find a “best fit” that will give you a starting forecast.
- Prepare Forecasts for New Product Introductions (NPI)
Use the demand planning software to identify products with similar sales trajectories which will be used as the starting forecasts for the NPIs.
- Override Statistical Forecasts with Judgmental Input
Use data from sales channels, knowledge about changes in market conditions, and expert insight to smooth the forecasts into the most realistic forecasts possible.
- Adjust the Baseline Forecasts for Promotions
In certain industries, like consumer goods, promotions can have a huge impact on sales volume and need to be factored into the baseline forecasts.
- Manage Vendor Managed Inventory (VMI) and Collaborative Planning, Forecasting and Replenishment (CPFR) Processes
Be sure to communicate data to both customers and internal managers responsible for these programs.
- Generate a “One Number” Forecast
Integrate forecasting into a Sales and Operations Planning (S&OP) that brings together executives from key areas of the company to ultimately agree on a single forecast number and execution plan that will drive both the demand and supply sides of the enterprise.
The one change I’d make would be to replace the first two steps with the following:
- Do a Spend Analysis
A spend analysis project, performed by a spend analysis expert that uses a real spend analysis tool, will load all of your relevant data, cleanse it, normalize it, and properly classify it in multiple spend cubes. The resulting cubes will allow you to perform the analyses necessary to identify which data is relevant, which data is statistically significant, and, more importantly, which products require significant forecasting efforts and which products are relatively stable year after year. Products with relatively stable sales do not need significant forecasting efforts, because expected demand can be easily determined from the spend analysis. On the other hand, products with variable sales, especially those products with a seasonal demand that are heavily influenced both by manufacturer promotions and competitor’s promotions for similar products, require detailed forecasting efforts.
- Load the Relevant Data
Once you have identified those products that require forecasting efforts, you can load the associated data that is needed to run the statistical models, to determine the effects of planned promotions, and determine the appropriate demand forecasts.