Implementing hierarchical disaggregation in forecasts to read granular signals & localized trends, during a period of atypical demand history.
The lower you go in analyzing base demand, the more difficult it becomes to see through the weeds. In assessing how the coronavirus outbreak has altered demand forecasting, the same edict can be applied. Planners continue to struggle through forecasting an unprecedented period of demand history, and it remains unknown how practices will permanently change until the outbreak subsides.
However, the outbreak has already exposed a long-standing chasm in demand planning practices. Despite advances in data availability, analytics, and technology, many teams continue with traditional approaches: consensus driven forecasting, relying on shipment and statistical time-series data – in truth supply signals – deployed by their ERP systems, leading to an untold number of manual overrides before the forecast can be used as an input into production planning.
When the coronavirus hit, forecasts based on historical sales were unable to capture changing demand patterns and the external factors associated with the outbreak. The discrepancy between historical vs. present data forced demand planners to adopt a reactionary focus based more on supply planning.
Though we are not out of the weeds yet, it is time to step back and assess how the outbreak has affected your demand planning process and whether it has identified stagnant forecasting practices at your organization.
How can demand planners take back the driver’s seat? Granularity.
We all know that downstream data like POS and syndicated data are the truer demand signals. Many of you already receive these datasets. Though, an inability to incorporate these datasets within your forecasts, clouds visibility from your customer to the end consumer.
This breakdown has been exacerbated during the coronavirus outbreak and has led to even more manual interventions in the demand planning process and a reliance on buffer stock to respond to fluctuations in demand and supply. Fine-grain demand forecasting, through hierarchical disaggregation and the deployment of POS and syndicated data, will allow you to regain control and build models that reflect the dynamics of localized demand.
Yes, hierarchical disaggregation is a substantial undertaking requiring changes to your data sources, import processes, and location hierarchy. That said, as chasms in recovery widen heightened localized insights will serve as a critical competitive advantage.
Hierarchical disaggregation results in a larger number of forecasts. Your forecasting process will become more resource intensive, especially if incorporating qualitative forecasting and planning techniques requiring demand planner intervention. You must consider factors which would inhibit disaggregation; such as processing and forecasting tool constraints, as well as the sufficiency of downstream data delivery cadence to inform forecasts within a useful planning horizon. When considering the current state of your demand planning process, would the additional resource requirements not be worth the gains in supply chain agility and the reduced reliance on buffer stock?
The atypical demand during the initial outbreak and subsequent recovery during the pandemic has forced many retailers and FMCG manufacturers to turn off their statistical forecasts and other sophisticated machine learning models used for demand sensing. This has required substantial intervention to diagnose events and make corrections, as change outpaces machine learning models ability to self-correct. Disaggregating to a lower hierarchy and making use of downstream demand signals may help models adapt to localized market drivers more quickly.
In our current environment, supply chains are subjected to unprecedented strain. If you were to disaggregate demand to a localized level, is there enough flexibility remaining within your supply chain to support these fine-grained opportunities? Segmenting demand involves additional costs. Are those costs justifiable for increases in supply chain velocity and product visibility in areas that are experiencing faster demand recovery? It is important to consider the granularity of the data you intend to use as demand signals. Can the format of POS data, syndicated data, or other datasets be adapted to a lower hierarchical level and incorporated into your forecasting system? Are their additional costs for obtaining the data at a more granular level? Can the data be delivered in a regular cadence that is useful for your target planning horizon? What changes will need to be made to your current disaggregation logic?
Fine-grained demand forecasting will heighten your S&OP team’s understanding of localized demand trends, highlight markets to prioritize based on speed of demand recovery, and provide an advantage over your competitors that lack similar market visibility.
However, the process of hierarchical disaggregation requires a time and resource investment from demand planners at a time when your organization is already facing an unprecedented period of disruption. Consider if this step will not only be useful in stabilizing your current forecasts, but also become a cornerstone in the maturity of your future planning processes.
About Fiddlehead Technology: Fiddlehead works with prescriptive analytics to find elegant solutions to some of the food and beverage manufacturing industry’s most complex problems. The result is more accurate demand forecasts, allowing companies to lower inventory, achieve higher levels of service, and improve their margins.