Bugaboo: something that causes fear or distress out of proportion to its importance.
When it comes to running a company, when things break down executives have traditionally said “we need to improve our forecasting!” Would better forecasting accuracy be a good thing? Absolutely!
Unfortunately, most companies cannot, and will never be able to, consistently rely on highly accurate forecasts. Two months before COVID made headlines in the US, nobody was forecasting the dramatic downturn in demand. Further, there may or may not be a recession. If a recession hits, demand will decrease. When will the recession hit? How much will demand be impacted? All of these things are all but impossible to forecast with any degree of accuracy.
When companies talk about improving their forecasting, they are most often referring to demand forecasting. Demand forecasting is the process of making future estimations of how much of a given product will sell by location and time period. Organizations then convert those demand forecasts to the associated quantities of raw materials to purchase, goods to be manufactured, or finished products to ship.
Here are ten things to keep in mind in terms of the role that forecasting can have on supply chains:
While not a cure all, forecasting does matter. As demand forecasting accuracy increases, and the standard deviation associated with the forecast decreases, the need to hold “just in case” inventory also goes down. This leads to lower inventory carrying costs and thus better case flow. Further, better inventory positioning – the right inventory in the right location – improves a company’s ability to deliver what their customers want. This increases sales. In consumer goods industries, better forecasting leads to lower fines from retailers for late or incomplete deliveries. Demand forecasting should be tightly integrated to an inventory optimization application. Doing this, increases the benefits.
Companies need to understand how much it is realistically possible to improve their forecasts. It is important to benchmark forecast accuracy and similar supply chain metrics against your peers. APQC – a benchmarking, best practices, and performance improvement organization – is one place you can get these numbers. Some suppliers of demand management software can also provide excellent forecast benchmarking for selected industries.
Demand forecasts are improved with access to downstream data (point of sale, Nielsen retail data, and access to competitor promotion schedules). Other external data, like industry data or economic data, is used for other types of forecasts. Increasingly, forecasts are being improved by leveraging outside data sources rather than merely relying on a company’s internal historical shipment data.
Demand management solutions that use machine learning perform better than solutions that don’t – particularly for short term forecasts. Machine learning also makes it possible to make more granular forecasts – for example, instead of forecasting demand for the company’s products in the Eastern Region of the U.S., forecasting product sales at 10,000 stores. Machine learning is making it easier to explore whether new data sources, for example weather forecast data, lead to improved forecast accuracy.When a big event – like a pandemic or recession hits – and forecast accuracy plummets, demand management software that uses machine learning recovers more quickly. In other words, the new forecasts become accurate more quickly than traditional forecasts do.
Large consumer goods companies believe that solutions based on a graph database also have real potential. Tech giants like Google, Facebook, LinkedIn and PayPal all tap into graph databases to generate customer recommendations and ad placements based on huge volumes of customer data they have access to.
Demand models need to be continuously updated. Part of the role of a forecaster is to ask why there was an unexpected demand spike or crash. If the reason can be determined, the model needs to be updated.For example, it may be that there was a spike in restaurant chains sales at restaurants in Phoenix metro area in the second week of February. It turns out the Super Bowl happened on February 12th in Glendale, Arizona. The forecaster should not assume there will be a similar spike in sales next January in Phoenix. The Phoenix forecast for the coming year needs to be smoothed out to reflect what would occur absent that event for those restaurants.The next question the forecaster should ask is where the next Super Bowl will be played. The model can be updated to reflect a demand spike for that city during the relevant period. In short, demand models need to be living models that are continually enriched.
Demand forecasting needs to be an objective process. This sounds obvious. Sadly it is not. A CEO, for example, who has told Wall Street they will grow by 7% in the coming quarter, and then looks at a demand forecast in a business planning meeting that shows lower growth, might be tempted to pressure the demand team to up the forecast. This kind of stupidity ensures poor supply chain performance.Instead, the CEO should ask what the company could do to increase sales – cut prices, run new promotions, etc. – and examine how those actions would affect the forecast. And if no action will lead to the promised revenue bump, the executive needs to live with that. This will allow the supply chain team to use the best forecast to run their operations.
COVID taught business leaders the need to develop better agility. Fortunately, many executives now understand that achieving agility can’t be done solely by improving their forecasting. Forecasting needs to be part of a more agile integrated business planning process. A more agile process might be achieved by speeding up the forecasting process. Instead of monthly forecasts, for example, a weekly forecast can improve agility. While doing a weekly forecast for all products may be unrealistic, many organizations might find it possible to do weekly forecasts for promotional items or key products.
Agility is greatly improved through the usage of forecast scenarios. Companies need to produce a base-case scenario – a forecast for what they consider most likely to happen. But they also need a best-case scenario surrounding the sales that could happen if they catch some tail winds; and a worst-case scenario for what could happen if they face head winds. Then companies need a way to monitor those head or tail winds to know what scenario they are in. For example, a company with a monthly forecasting process could monitor their weekly orders and compare what they expected to occur with what is actually occurring.
Demand forecasts are not the only forecasts companies need if they want to improve supply chain agility. Other kinds of forecasts are also important. There need to be long-term forecasts examining what demand is expected over the coming year or years. This allows companies to determine if they will have the production and warehouse capacity to meet long-term demand.
There can be forecasts focused on whether sourcing partners will be able to deliver raw materials to the company in the period needed, or if key carriers can deliver goods to their customers by the due date promised. Demand forecast can be used as an input to forecast how many production or warehouse associates will be needed in a particular week. Many of those forecasts are not produced by demand management solutions but require other applications or tools. For some of these more novel forecasts, machine learning and Big Data can be critical. Leading companies are becoming highly innovative in producing new business forecasts to improve their operations.
In conclusion, the mindset of an organization matters. President Eisenhower, reflecting on his military career, once said “plans are worthless but planning is everything.” That kind of thinking is critical to avoiding the forecast accuracy bugaboo.