How do you get demand forecasting back on track after last year derailed it? These 5 methods make it easy in Deposco.

One of the biggest hitches facing demand planners right now is, “How do I get my forecast back on track, when all I’ve ever known is forecasting on historical data… and COVID-19 blew that out of the water?” Things get especially hairy when you throw in the twist of seasonal demand planning.

If it seems like the pandemic left you nothing to work with but data outliers, fret not. 

Our new Bright Forecasting Ideas series kicks off today with 5 ways you can use Deposco’s advanced demand planning tools to easily manage data outliers and seasonality in forecasting. 

Before I dive in, let’s reminisce for a minute…

In the good ‘ole days

Before the words “Delta Variant,” “N95,” and “quarantine” invaded our everyday dialogue, demand forecasting made sense. (Generally speaking). We forecasted on logic – from a set of mathematical “if-then” scenarios, defined by prior sales history. Decisions could be backed up with data, which we then used to predict future demand. And then… COVID-19. 

“If/Then” logic suddenly became “What-if/I-Have-No-Clue” scenarios as the norm because we’d never lived anything like it. These unknowns crept into our personal lives too. And they continue to perplex us:

  • What if high school sports hadn’t been canceled last spring; would my 12th-grader be in college on a baseball scholarship right now? 
  • What if quarantine had never happened; would I have missed out on that new friend I made out walking amongst neighbors sharing my same wicked case of cabin fever?
  • What if I had purchased stock in Peloton right before COVID hit; would I have gotten rich when all the gyms shut down 2 months later? What should I invest in this year?

Point is, we have no idea. And quite possibly, we never will. With the flow of historical wisdom upended, so much data in our personal and professional lives has become an outlier. Fortunately, there are advanced demand planning tools that prove very useful in accelerating recovery for our customers. 

Advanced forecasting tools give you options

So what options do you have? Bright Forecast offers 5 advanced demand planning tools with flexible parameters that use a mix of traditional and machine learning-based forecasting algorithms, depending on what you need to do.

This flexibility helps demand planners make swift, accurate decisions in forecasting pandemic-related outliers as well as seasonal, multi-location and multi-channel demand forecasting. 

Which method to use and when 

Use these post-pandemic forecasting approaches to predict demand for any type of product—including those that are expected to recover, still recovering, will never recover, require seasonal time series, or are just a wild card.

#1:  Prophet Statistical Model with logistic growth

Use when: demand is still recovering, and you think it will eventually return to a previous level. This method models the more recent history fairly accurately for the upcoming forecast. 

Let’s say demand for an item dipped drastically at the start of the pandemic. In Bright Forecast, you would apply parameters with a specified max. The system intelligently aligns the forecast to recent history, rather than pandemic history, to return a forecast that will not exceed that max, as recovery continues.

#2:  Consider only recent history; treat it like new item

Use when: demand is not expected to recover to prior levels, but will remain closer to what it is now. In this case, treat it like a new item; consider only history from the recovery start period. 

This is a great option when the product is not expected to recover to prior levels. Simply adjust the start date for historical data to March 2020. If the item has fully recovered, Bright Forecast offers various best-fit statistical models to choose from. Just be sure to keep these two considerations in mind:

  • Many time series will trend upwards after the pandemic and will therefore project growth throughout the forecast horizon. Ask: Is an upward trend of demand present that is expected to continue? If you expect further recovery to happen, it may make more sense to use the Prophet Statistical Model with logistic growth.
  • If a time series is seasonal, you may lose visibility into the seasonality by only considering one year of history for many months. Ask: Is there important seasonal information in the older data points? 

#3:  Apply a data mask

Use when: the impacts of the pandemic are completely over and demand has recovered to normal. If you need to handle seasonal time series, you can apply/combine the data mask with a seasonal fill method (see #5 below). 

Use a data mask only when forecasting demand for items impacted by COVID. Historical data will not be relevant to the future; therefore, you’ll need to create a data mask.

In Bright Forecast, specify the date range for which the pandemic rendered the historical data invalid (in many cases, this will be March through September 2020). Then apply the data mask on the workbench. This will smooth out the historical values, showing estimates of what the values would have been if the pandemic had not happened.

#4:  Apply the Naive Statistical Model.

Use when: you are unsure about any of the above scenarios. 

This model sets the forecasts of all future values to the last observed value. It’s okay to use this model if you feel that demand has leveled off and will not continue to recover. 

#5:  Apply a seasonal data mask with Seasonal Naive Statistical Model

Use when: you’re dealing with time series that display a consistent and significant seasonality pattern, with no significant long-term trends.

Just like using a regular data mask, this method mitigates the effect of COVID on historical data. It’s different though, in that it masks values based on the year prior’s data, rather than a non-seasonal method—such as a moving average or prior-period observation. 

A regular data mask produces a more stationary adjusted historical value. Here, you can replace COVID data with historical values that contain a seasonality pattern. The system allows you to force the forecast to replicate the demand in 2019, rather than in 2020.  

Need help?

Current customers. Already a Bright Suite customer? Get personalized help with any of these methods and extend the power of Deposco.
Contact me

New to Bright Suite? Or want to learn more about Bright Forecast? Request a tailored demo today!
Get a demo

More Bright Forecasting ideas coming your way 

This year continues to be a best guess for many of our customers. Coping with the roller coaster of supply chains re-opening, then immediately threatening to shut down again, now is the best time to size up your manual forecasting tools and processes. 

  • Can they scale? 
  • Were they built for supply chain execution, or something else that requires a whole lot of heavy lifting and retrofits? 
  • Can they seamlessly plug into other areas of your business process, like the warehouse, order management and fulfillment?

Next up in this series

Stay tuned! In the coming weeks I will expand on the topic of demand forecasting, with more demand planning goodness including:

  1. Why is forecasting in Excel so hard? What’s wrong with spreadsheets, and why demand planning software now?
  2. What are the benefits of cloud-based planning? What are some examples of practical tools within these applications and how do they help companies get better with forecasting / inventory management?