In January 2020, the novel coronavirus (COVID-19) outbreak in Wuhan, China plunged the world into a global crisis. This highly contagious respiratory illness has infected more than 20 million people, causing over 900,000 deaths worldwide. The scale of this pandemic is unprecedented, with severe lockdown measures hugely impacting business across all sectors. The World bank estimates that the global economy will shrink by 5.2% this year, which would represent the deepest recession since the Second World War.
While all can agree that the global impact on business today is severe, the future of business and economic recovery remains blurry, especially when we observe how different sectors have been impacted. For business leaders today the level of uncertainty faced can be overwhelming, and despite an endless amount of data in the digital age (virus metrics, lockdown measures, customer data, social listening, macro indicators) our clients face difficulties in identifying what data is relevant and how to use that data to effectively plan for the future.
In this article, we share a structured framework to project the recovery of a specific business sector & market, embedded with the flexibility to adapt your data inputs as the situation develops in different parts of the world. Ultimately this framework will help you to leverage data & analytics to narrow the range of uncertainty of the future, steer your business, and hopefully gain an edge over the competition.
We break down the core elements of our framework, share some concrete forecasts, and then discuss how this framework could be replicated or adapted to your specific industry.
The first thing we recommend is to avoid the urge to model the macroeconomics (GDP, consumer confidence, etc.) of an entire country. Ultimately, your goal is to understand your specific business sector. Trying to model the entire economy of a country will simply invite more questions about the interactions of different sectors and population dynamics and will distract you from the focus required to zero-in on your specific market and industry.
To illustrate our structured recovery framework, we will take the Japanese beauty market as a use case (a market and industry where we have deep expertise).
Like other countries around the world, Japan’s economy is being severely tested with significant slowdowns in trading and tourism, including the postponement of the 2020 Summer Olympics. Here we review data collected on the Japanese beauty market which has experienced a -33% decline YoY Jan-Apr 2020. While we do see a boost to ecommerce sales, this has not nearly compensated for the steep decline in offline sales resulting from lockdown measures and department store closures across the country. (Figure 1)
Our challenge was to utilize our recovery framework to model what a recovery for Japanese beauty could look like for the remainder of 2020 by leveraging data, advanced analytics, and sectorial expertise.
Business leaders in the digital age have a potentially limitless choice of data sets to collect and sort through. Data can come from inside the company (customer data, financial statements, etc.) or from outside the company (macro-economic indicators, social media, stock market). In the COVID-19 era, we must also add virus propagation figures, lockdown measures and social restrictions to that pool of data.
We recommend that instead of taking on the seemingly endless task of trying to collect every data set available, you leverage your expertise to identify key categories of information that represent your market’s overall performance.
For the Japanese cosmetic market, the key impact data categories are consumer behaviour, tourism and macro-economic signals. Once your data categories are identified, you can then prioritize & organize the hunt for data. This framing effort will reduce the burning of time and resources on sourcing less critical data sets or data sets that will only have a small or subtle impact on your market dynamics.
Obviously, these categories are specific to the specific market & sector you are focused on. In Japan, tourist shoppers contribute significantly to the beauty sector annually and are important to include in our data collection efforts. For a different sector such as grocery, tourism may not be relevant. Instead, you might look for data on take-out and/or dine-in sales to understand those customer dynamics.
After identifying your market KPI(s), and structuring the different categories of data signals to source, you can proceed with the three core elements of our framework:
Our recommended approach is compartmental modelling. This involves segmenting the stages of infection into discrete “compartments”, and then modelling how the population moves between those compartments. A set of assumptions then governs how each member of the population moves through compartments (e.g. movement from Exposed – Infected may be based on medical research of Virus infectiousness). Using our open-source Python library (pyepidemics* by Ekimetrics) you can build models to fit real data, develop scenarios on how the virus will propagate and even simulate different lockdown scenarios.
Best practice tip: Public reporting of new virus cases is highly dependent on the level of testing that is conducted. Leveraging data on hospitalizations (critical cases) or deaths is usually a more reliable metric and is what helped us to predict a resurgence of virus cases in Japan where testing rates were low.
Linking virus propagation to consumer anxiety:
In the Japanese cosmetics example, one additional step we took was to understand how virus propagation impacts consumer anxiety, which is ultimately what drives whether consumers feel safe to venture outside and buy in physical retail. When comparing virus related search queries in Japan we can see two interesting phenomena:
Combining your virus propagation models and consumer anxiety effects will help you build the scenarios to feed into your recovery models.
Once your key impact data categories have been identified (consumer behaviour, tourism, and macro-economic signals for our Japanese beauty use case), you can gather data sources to test in your models. Some data, such as government reported consumer confidence or tourism figures, can be easily found. However, you will likely need to get creative about building your own indicators. To capture consumer behaviour in Japan, we constructed an e-commerce indicator combining data sources from all major e-commerce platforms in Japan.
In this phase, it can be useful to automate data collection by leveraging web-scraping techniques to reduce manual collection efforts for future iterations.
Finally, you can start to build your models and to test the various data sources you have collected. There are a host of different statistical models to choose from but in a situation where the number of historical data points is limited, simple regression models are a very effective start.
Once your model has stabilized, you need to start thinking about how to build forward-looking assumptions. Here, it is important to leverage a mix of your industry expertise and expert thought leadership. Expert opinions can be captured from interviews with industry experts, recorded reports or opinions, consumer surveys, economist opinions, etc.
By ingesting and synthesizing this information, you can create a set of baseline scenarios for each data signal which represent a reasonable range of uncertainty to consider.
In our most optimistic case of a quick market bounce-back we projected a -18% evolution YoY May-Dec 2020, and in a lower bound scenario a -44% evolution where we assumed a resurgence of virus cases and extended impact to the market. The situation in Japan today has obviously changed and highlights the importance to put a framework like this into place, monitor developments, and update the projections as needed.
There are many ways to use the outputs of our framework. Here are some concrete ways we have leveraged them for our clients:
Even for those in the same industry as our example (beauty), elements of this framework will need to be customized to your specific market. For example, when applying this model to the US you may want to break down virus propagation into regions to reflect the varying severity of the COVID-19 impact geographically.
We can see below (Figure 6) how digital disruption has impacted the retail industry in APAC markets differently. Japan is part of the mature followers group, where e-commerce evolution has been modest historically. On the contrary Vietnam, Indonesia and India are lagging in terms of market maturity and digital disruption, but are expected to accelerate quickly, inspired by China the digital leader. Customizing your indicators for the “Fast Modernizers” and “Developing digitalizers” markets to include logistics and overall distribution development could be helpful in building your models. You may also want to include indicators on the growth and spending power of the upper & middle-class (discretionary spending, income gaps) when assessing the dynamics of less-mature retail markets.
For automotive players in Europe the COVID-19 crisis occurs in a challenging time, where the industry was already facing big shifts in the form of greener electric vehicles & ride-sharing ecosystems. In this example we must completely restructure the identification of relevant macro-economic categories and variables.
Among others we can think about some key categories:
In a world of unprecedented uncertainty, leveraging data & analytics can help leaders to guide their businesses through the fog. By applying a structured framework to project the recovery of your market, you can prevent the wastage of time and resources on less critical tasks and gain an edge on the competition.
—
*Ekimetrics has developed the open-source Python library pyepidemics to help people easily run their own virus propagation models based on epidemiology best practices. This initiative fuels the French government on virus containment decisions and is a valuable output of our collaboration within CoData, a global NGO aiming to promote the use of data for all areas of research.
References:
—
> Subscribe to the data science & AI newsletter!
