How to Forecast Demand Despite COVID?

The article below is a summary of one of my LinkedIn posts. If you are interested in such debates, let’s connect!

I would like to thanks the following people for their insightful remarks in the original discussion: Valery Manokhin, Nick Cronshaw, Robert van Dijk, Thomas Meersseman, Wassim Tabbara, Archit Patel, Chris Davies, Joris De Smet, Aleksandra Barteczek, Paul Balcaen, Karl-Eric Devaux, and Rohit Anand.

? COVID shook supply chains in 2020. How should you forecast future demand when everything is changing and you lack relevant data?

??Flag Outliers (Simple Solution)

The most straightforward response to an unusual demand-period is to flag it as an outlier. As the demand over the COVID months was exceptional, we can assume that it is not representative of future demand (in a post-COVID world). A safe bet would be to flag those months as outliers in your forecasting engine. Often, overwriting the demand in periods with outliers by the latest previous forecast will do the trick (see my article on outlier detection for more info).

? Pay attention to seasonality H1 2020 was heavily impacted by COVID. Even if you manually tweak H2 2020 forecasts, H1 2020 might impact early H1 2021 as your model will learn and apply new seasonal parameters. Cleaning H1 2020 is, therefore, tremendously important.

??Event Forecasting (Better Solution)

Demand outliers are often due to exceptional events: strikes, massive promotions, business shutdowns, etc. This means that if your forecast engine can learn the impact of those events, it will understand the past and provide better predictions for the future. If your forecast model allows it, you can flag the COVID lockdown as an external event. Your model will then take care of computing its impact.

? Pay attention that estimating the impact on your sales of an event that only took place once is mathematically risky. Learning statistical relationships from an event with only one observation is as good as guessing.

??Use External Drivers (Even Better Solution)

Many external drivers impact demand: promotions, marketing, events, pricing, number of stores open (or m2), number of open clients/plants for business, machine running time/consumption (for spare parts), GDP, etc. This means that if your forecast engine can learn the impact of those external drivers, it will be able to understand the past and provide better predictions for the future. Often, machine learning models do a better job of forecasting with external drivers than statistical models (see this article for an introduction on how to use machine learning for demand forecasting).

In the case of COVID, your sales were likely to be impacted by the number of shops open or the type of enforced lockdown. If you can feed that information to your forecast model, it will understand what drives your sales. And be able to properly plan for the post-COVID recovery (or any new lockdown/restrictions).

??Forecast the New Normal

COVID impacted every product in every supply chain. Some faced a decrease in demand, while others enjoyed an increase. Both cases can appear within the same supply chain with new best sellers rising while old products fade away. It means that our forecast models will have to learn, recognize, and apply those new demand levels. Unfortunately, the amount of data available to estimate these new demand levels is (very) limited. Using mathematical models to forecast this new demand is therefore risky. In such a case of changing demand and limited data, relying on judgmental forecasts might be your best bet.

??Final Advises

• Know Your Industry. Different industries will react differently to COVID, lockdown(s), and business reopening(s). Clients might have been stocking up or waiting to buy all the lost sales. Many industries will likely evolve after COVID. Buying patterns will change, and they will evolve differently for different products and services within the same industry.
• Know Your Limits. When making previsions, it is crucial to know the confidence you have in your own forecasts and inform your fellow planners about the expected ranges rather than just a middle value.
• Scenario Planning. Once you know your industry and your own forecasting limits, you can start to work on different scenarios. Those need to be aligned with various stakeholders within an S&OP process.
• Use New Models. In order to predict post-COVID demand, you could even use forecasting models specialized for new product launches.
• More Changes to Come. The only thing that we can sure about is that the uncertainty over the following is exceptionally high. Be ready to face more changes and challenges.

Nicolas Vandeput is a supply chain data scientist specialized in demand forecasting and inventory optimization. He founded his consultancy company SupChains in 2016 and co-founded SKU Science—a fast, simple, and affordable demand forecasting platform—in 2018. He enjoys discussing new quantitative models and how to apply them to business reality. Passionate about education, Nicolas is both an avid learner and enjoys teaching at universities: he has taught forecasting and inventory optimization to master students since 2014 in Brussels, Belgium. He published Data Science for Supply Chain Forecasting in 2018 and Inventory Optimization: Models and Simulations in 2020.