Demand Planning Breaks Down During Recessions
In 2010, I talked to a supply chain executive from a consumer electronics accessories company who mentioned the terrible return on investment he had gotten from implementing a demand planning solution. I was surprised. Most of the time I hear about pay back periods of less than two years from implementing supply chain solutions. He pointed out that the implementation had occurred during the great recession that occurred during 2007-2009. Demand planning breaks down during recessions. Based on recent news, this topic is worth thinking about.
The last nine recessions have been preceded by a yield rate inversion where the yield on 10-year Treasury bonds dipped below the 2-year Treasury. The stock market tumbled 800 points, or almost 3.1%, to 25,579 on August 14th. This was due in large part to recession fears sparked by the inverted yield curve. The yield curve inverts roughly 14 months before a recession.
One of the worst things a company can do when the world is tipping into a recession is to react too slowly. Demand planners are, not surprisingly, much better at forecasting in a steady state environment when things are not changing too much. As Dr. Madhav Durbha, Group Vice President of Industry Strategy at LLamasoft points out, planners are typically working off a two- or three-year history of shipments and orders. They assume the future will look similar to the past. LLamasoft is a leading provider of supply chain planning solutions, including intermediate to long range demand sensing and modeling. “These inside out approaches leave plenty of blind spots and companies are caught flatfooted when things turn rapidly,” Dr. Durbha explains.
“The path of least resistance is to assume tomorrow will be like yesterday, but with a slight upward trend. I witnessed this during my work with several customers during the great recession a decade ago. It is not that demand planners did not have the intuition that a recession was coming. Smarter demand planners saw it coming. But they were too afraid to speak up. They didn’t have facts to back them up.”
Robert Byrne, the Vice President of Supply Chain Solutions at E2open, pointed out that the demand models work off data. No demand model he has seen has included bond yield curves. For machine learning to work well, it needs to be a big data application. In addition to doing forecasting based on historical sales, consumer goods companies leverage other data sets such as their retail customer’s point of sale, recent shipments of products from their warehouses to their stores, the retailer’s orders, syndicated data, and store inventory. Many of these data sets are accessed daily, or even several times a day, so the dynamic nature of demand is captured to a much higher degree than traditional forecasting techniques.
Machine learning is used in the E2open demand management solution. The engine is making many forecasts simultaneously in different planning horizons. So, there can be a forecast for demand for liquid detergent in a 100-ounce container to the Walmart store in Tuscaloosa tomorrow, one week out, and one month out. There can also be a forecast for how much of that detergent will be needed at the distribution center in Cullman, Alabama tomorrow, one week, and one month from now. Other forecasts are being done for other big retail customers and channels.
For the forecast of what is needed in the Tuscaloosa store tomorrow, it may be that the data source with the best predictive power is point of sale. For a forecast of how much of that detergent will be needed in the Cullman warehouse in a month, a different algorithm might be used that weights the data sets such that the statistical sales forecast is most important, recent shipments the second most important; customer orders third, and so forth. The data sets that have predictive power may change over time based on external factors as well as the lifecycle of a product.
A recession is the kind of event that is very likely to change the data sets that have the greatest predictive power. The more at bats a machine learning application has, the better it gets. That means when a big event like a recession occurs, there is apt to be a lag before the demand engine can recalibrate and begin to approach former levels of accuracy. It can take demand models a while to react to the changing environment.
Nevertheless, companies that use downstream data, actual consumption data at the point of purchase, will react more quickly, and carry less inventory, than companies using order and sales history. Even so, there will be a lag before the engine catches up to the changing circumstances. There will be less lag, however, for companies that forecast more often. Companies that create weekly, or even daily forecasts, will carry less inventory than those relying on monthly forecasts. This assumes, however, that other parts of the organization are able to make use of that forecast. A manufacturer with a monthly production schedule that is locked down, is not helped by weekly forecasting.
Downstream data is important. But other data sources matter too. Planners need to increasingly rely on external data/factors – “an outside in approach” - to augment demand planning according to Dr. Durbha of LLamasoft. He pointed out that there are many sources of data with predictive value that are rarely used. In fact, the reason I reached out to LLamasoft for this article was because their Demand Guru product and its’ access to Big Data. Demand Guru has access to over 550,000 data sets across all countries. Many of these data sets are based on industry and macroeconomic data.
Surrounding a recession, macro indicators such as housing starts, GDP, employment levels, customer confidence, etc. should be tested. Data sets that could have predictive value can be explored by looking at what happened to sales at the company during past recessions. Then machine learning can be applied to the data, and external causal factors with predictive value can be identified and prioritized from an impact perspective. At that point, a demand planner is armed with the ammunition the need to argue for a more conservative demand plan.
Where LLamasoft has gotten good traction with their demand modeling solution is for intermediate to longer term forecasting. Much forecasting is focused on what will happen in the coming weeks or month. The use of industry and macroeconomic data seems particularly well suited to forecasts that encompass a period stretching from a few months to strategic plans going out years.
What do you do around forecasts that occur in tactical and strategic time frames? Dr Durbha pointed out that figuring out how to respond to different scenarios is key. If GDP goes down 1%, but our sales have not been much affected, how do we react? If they go down 2%, what do we do. “Based on what I’m sensing,” Dr Durbha said, “how far can I postpone a decision? For example, if a leading indicator points to recession, can I hold off on a CAPEX decision?” One example of deferring a capital expense (CAPEX), might be to decide not to spend money to increase a plant’s capacity. “Such decisions can be data driven and fact based.”
In conclusion, there is no magic bullet for forecasting when a recession will hit and how hard it will impact a company’s sales. But demand management solutions that use machine learning and can access macroeconomic and industry data are apt to perform better.