Demand Planning - Practicing the Best Practices (contd.)
In the last blog on this topic, I mentioned that I will share some of the best practices in some of the sub-processes in the typical demand planning cycle. In this blog I will focus on first couple of sub process of demand planning cycle - 1) Setting up demand planning objectives and metrics for different business units/customers/key items/locations 2) Setting up the frequency of the forecasting process (create/review/publish) with the time horizons.
While setting up objective of Demand Planning process, Organization needs to be cognizant of the fact that objective should not be just limited to generation of the most accurate statistical forecast by using best suited statistical algorithms. Achieving 100% forecast accuracy is next to impossible. Hence organizations need to leverage internal collaboration (for demand shaping) and external collaboration (for demand sensing) as well in defining the holistic process in order to cater to the perennial demand uncertainty or variability. Demand planning process also needs to address specific business scenarios like New Product Introductions (NPI) and Product Lifecycle Management. In case of new products, even after applying concepts like 'Like Modeling,' achieving the goal of forecast accuracy is very much a challenging task. So, till the time new product is settled in the market, the goal in case of NPI can be a continual reduction of forecast error.
Another best practice in setting the objective is to strive for a single forecast across the organization for various products. Demand planning is a sub-process within sales and operations planning and is not a compartmentalized task. Each of the functional departments within company e.g. Sales & Marketing, Manufacturing, Finance would normally come up with their own forecast and there are high chances that some sort of biasness would creep in while generating different numbers. Hence Demand Planning process' main objective should be to derive a single operational consensus forecast by having a planned cross-functional meeting every month.
Now, let's look at different forecast measurement metrics e.g. Mean Forecast Error (MFE) or Bias, Mean absolute Deviation (MAD), Mean Absolute Percentage error (MAPE), Weighted Absolute Percentage error (WMAPE).
MFE measures the average deviation of forecast from actuals. However zero MFE does not imply that forecasts are perfect as positive and negative errors cancel out here. Hence it is not recommended as a single metric to be followed.
MAD measures the average 'absolute' deviation of forecast from actuals. Positive and negative errors thus do not cancel out here similar to MFE. However since this is a number, there is no way to know, if MAD error is large or small in relation to the actual Sales data. Therefore business planners will not be able to assess the real implications of higher or lower MAD in assessing the forecast value. Hence it is also not recommended as single metric to be followed.
As against MAD, MAPE measures absolute error as a percentage of the sales. Hence, as best practices, companies use both MAPE and Bias for measuring forecast error. Based on AMR Research, a 6% forecast improvement in MAPE could improve the perfect order compliance by 10% and deliver a 10-15% reduction in inventory.
Most of the companies normally choose to generate forecast at category or family as compared to the SKU, location and customer level. The greater the degree of aggregation, the more accurate is the forecast. However the best practice is to generate forecast at the SKU, location and customer planning level. In the past, this was not possible due to higher processing requirement of the servers. But this is pretty much possible today with any of the leading supply chain planning products available in the market.
With regard to frequency of forecast generation, most of the organizations adopt monthly forecast generation for the horizon of 18-24 months which is allocated over weekly buckets by using allocation rules. As per the research conducted by Aberdeen group, almost 60% of companies generate forecast at a frequency of lesser than a month. However even a month is a pretty long duration for sensing the demand for CPG company, as on daily basis there are wide fluctuations taking place in the consumer marketplace. Hence CPG leaders like P&G, Kraft are performing short-term forecasting (by making use of Demand Sensing product) each day for the one-to-three/four week time period in addition to regular monthly forecasting for longer duration.
In next blog, I will keep on adding best practices in CPG industry for the next sub-processes of demand planning process. Till then, please enrich this series of blogs with your experiences as well...