This site uses different types of cookies, including analytics and functional cookies (its own and from other sites). To change your cookie settings or find out more, click here. If you continue browsing our website, you accept these cookies.
Right now, the power to solve new global challenges across industries, is at your fingertips, no matter where you're working from. Create a new topic or reply to an existing thread to share your perspective.
The COVID-19 pandemic presents unprecedented challenges for a multitude of industries who are experiencing extreme disruptions in their supply chain across the globe. As many industries depend on goods and materials being produced in China as well as other affected regions, many are scrambling to find solutions for short term shortages and massive delays, while also looking to avoid this risk in the future. Planners, buyers, and analysts are looking to speed up the delivery of high-value insights to colleagues to respond to this emergency today.
As we look to recover from this crisis, demand for particular products may increase or decrease, making the estimation of realistic final customer demand harder and more important.
Because current customer short-term and medium-term demand may be unrealistic, demand -planning teams will need to approach reshaping their demand forecasts differently than in the past. This includes reshaping demand to activate secondary suppliers, as well as reshaping demand to include realistic time to delivery.
As we transition into the “New Normal” we’ll need to find relevant data from previous historical events that provide some context to what we’re experiencing today. Consumer behavior has been irrevocably changed disrupting supply and demand chains. Identifying leading indicators, which is an economic factor that changes before our target variable, is the best solution we’ll have to forecast demand. Some examples include:
Previous recessionary types of data (Alteryx helps an analyst connect to disparate data whether it’s an old database, new information stored in the cloud, or good old fashion flat files so they can blend historical and present information.)
Previous catastrophe events like Hurricanes or Tornados and disruptions in oil prices may help if they previously affected the supply chain in a region.
Present datasets such as COVID quarantine areas, John Hopkins Data Sets, and Git Hub Data Sets for training innovative models are also helpful.
We want to create a model that best reflects the current situation at hand and contains the least amount of entropy possible. Entropy, as it relates to machine learning, is a measure of the randomness in the information being processed. The higher the entropy, the harder it is to draw any conclusions from that information.
A black swan event, which is what we’re experiencing with Covid-19, is an unpredictable event that is beyond what is normally expected of a situation and has potentially severe consequences. Black swan events are characterized by their extreme rarity and their severe impact. We can look at historical black swan events as a means of gathering insight.
From a time-series data perspective, this means slicing the available historical data so it correlates with our present experience. What were the leading indicators during a similar black swan event that can help provide useful forecasts?
We would need to look for a time when consumer behavior was like today’s environment…we haven’t really experienced anything with this ferocity in the decline of consumers so how we’ll have to adjust our model to account for that. We are truly experiencing something very new and unknown.
So, if for example, you take previous recession and black swan event data as a proxy, you’ll begin to understand what variables are leading indicators to build a forecasting model.
You will need Domaine knowledge, so be sure to include your domain experts in the process.
What do they know? What’s their intuition tell them? Then validate what they are telling you. You’ll want to use Alteryx tools from categories such as:
Let’s first start by understanding our data. What is it telling us? Let’s first figure out association, correlation, and distribution of the data set.
Now let’s apply what we learned from investigating our data to predict trends and behavior patterns.
We also need to consider if our model is time-based. Should we use a time series model to adjust for trend and seasonality over time? If yes, let’s look at an ARIMA or ETS model to help with our forecasts.
Scoring and Forecasting
Lastly, we’ll take our model and apply it to new relevant data and provide informed estimates. Ultimately, this is the most important part of the predictive lifecycle.
If you missed our webinar on this topic and want more details please go to the Session Recording: