The Coronavirus disease (COVID-19) pandemic has created multiple waves of disruption for retail and manufacturing supply chains because of repeated cycles of lockdown and reopening. The rollout of COVID-19 vaccines, which marks the beginning of the biggest vaccination campaign in history, means a new series of supply chain challenges and adds more uncertainty on top of the lockdown-and-reopen disruptions, which are partially understood and controlled by this time. Here are some techniques to stabilize supply chain operations during the pandemic and how these methods can be leveraged during the vaccination campaign.
Many inventory and workforce optimization solutions are based on demand forecasting models that use historical data to forecast future demand. The shocks of the COVID-19 pandemic have invalidated many such forecasts, and many data science teams have responded to this by incorporating new data sources and making other adjustments. It is highly likely that the same approach and similar signals can help during the vaccine rollout period.
Correction using macroeconomic and epidemiological signals
First, let’s consider a technique developed for a leading clothing company that operates more than 900 stores worldwide. The company faced massive store closures in March 2020 as lockdown measures were introduced. Management quickly realized that the demand and operational patterns after lockdowns ended would differ greatly from the regular patterns and that rapid accommodation to the new environment would be critical for survival. The company deployed a data science team tasked with developing a model to optimize the post-lockdown store workforce to optimize the number of store associates needed on duty.
Workforce optimization is generally based on store traffic forecasts, and the company had traffic forecasting models. However, it was clear that these models would not work for the reopening scenario. Consequently, the data science team responsible for this track researched and implemented two extensions.
The first was a proxy model that estimated the effects of regular seasonal influenza on store traffic. Then, this model was recalibrated using the limited COVID-19 data available.
The second extension was a model that quantified the effects of macroeconomic factors. This model helped to establish a link between the patterns during the 2008 economic crisis, which was the latest major disruption, and the 2020 patterns.
Then, these two auxiliary models were combined with the regular, legacy forecasting model to correct for the distortions introduced by the pandemic.
Correction using mobility reports
The second technique used by the store was to incorporate Google Community Mobility reports, which quantify changes in consumer mobility. These reports are based on data regarding mobile device use and location, which Google updates continuously.
These reports helped to implement corrections based on deviations from normal levels of mobility. The store found that this signal was extremely useful during its reopening period, and can most likely help to navigate the vaccination campaign as well.
Correction using international data
The third technique was developed for a leading publisher of media products that used a demand forecasting model to optimize pricing and promotional decisions across more than 70 countries, multiple product lines and multiple digital and brick-and-mortar retail channels. This model was able to forecast weekly demand numbers for many months ahead with reasonable accuracy, and the system generated multiple forecasts for different levels of discounts. Pricing managers used these forecasts to understand how price elasticity would change over time and to determine optimal times for and depths of promotions.
However, when the COVID-19 outbreak began in China and the subsequent quarantine measures rapidly changed demand patterns, the company faced an unprecedented problem as actual sales numbers quickly diverged from those forecast. In the case of media products, the demand numbers surged sharply compared to the forecasts. In a few weeks, it became clear that COVID-19 had become a massive global disruption that invalidated demand forecasts for all geographies.
The company immediately updated its forecasting models. After initial analysis and experiments, its data science team decided to modify the existing solution as follows:
● A secondary demand forecasting model was introduced to estimate the forecasting error of the primary model.
● Additional signals, including changes in quarantine measures and university closures, were incorporated into the models, which helped to improve accuracy.
● Global data were used to enable learning to be transferred across countries and to leverage the experience of geographies where the pandemic had begun earlier in order to forecast the trajectories in countries affected later.
The adjustments to the model were implemented in about two weeks, which helped return the accuracy of the forecasts to acceptable levels and stabilized the process of managing revenue. This contributed greatly to the company’s resilience in this time of crisis.
Consequently, international data can be extremely useful in cases of global shock. This technique most likely can be applied during the vaccination campaign because the vaccine rollout will happen at different paces in different countries.