Demand forecasting has derailed: Here’s how to get back on track

demand forecasting has derailed heres how to get back on track img

Imagine that the traditional way of forecasting customer demand is like a train running down a track. For years, it had a smooth uninterrupted journey and then—boom—the train suddenly collided with COVID and derailed. Demand patterns dramatically changed, as did the ability to accurately predict future sales.

For some, the collision caused the train to leave the tracks. We all witnessed companies scramble to keep pace with shifting supply chains and consumer behavior. Many became overstocked with inventory after demand had waned, leading to steep discounts to move stagnant products. As an analytics leader who works with consumer companies, I also saw the opposite, with examples of staggering dips in on-shelf availability and missed sales exceeding 10% of overall revenue.

Getting blindsided by variables such as a global pandemic or economic recession is largely out of our control, but human error or negligence can also cause a train to derail. To avoid this and to get (and keep) accurate demand forecasting back on track, companies need to completely rethink the status quo.


Historically, demand forecasting involved looking at and extrapolating past sales data. This approach has utility, but the COVID collision suddenly changed what consumers purchased—as well as when and how—making past sales data far less reliable than ever. A CRI report found that “69% of retailers and 66% of consumer products companies had difficulties in demand planning due to lack of accurate and up-to-date information on fluctuating customer demand.”

Additionally, demand forecasting involves a great deal of human hypotheses. Demand planning and marketing teams have highly specialized skills, but they have natural human biases. Human-centered forecasting is constrained by the limits of human knowledge. Subject matter expertise is crucial, but it only accounts for a few of the drivers that impact demand.

Purely human-based demand forecasting is also subject to the tendency of people to try to game the system. The tactic of overinflating in order to stay ahead of demand may seem like a good idea, but can backfire. An example cited in Harvard Business Review is of an automotive manufacturer that routinely inflated its orders to a particular supplier by 10% to 15%. In response, its supplier decided to fill only about 90% of the orders.

So, we must look to data. Reliance on historical sales data and basic time series methods to forecast demand fails spectacularly, however, when demand patterns fluctuate significantly. This is also true for product categories that are erratic or intermittent, or when entirely new products are introduced that have no sales history. So, how can organizations adjust and get back on track?


Examining data from a variety of sources (both historic and real time) and understanding how those different pieces fit together provides a more well-rounded view of what organizations might expect in the future. Organizations need to use and corroborate data across multiple channels (online, offline, mobile, curbside, etc.) to understand true demand by category.

External data, earned data, and paid data sources are all crucial. Companies can use internal data to carefully select sales history data that accounts for pre-pandemic, pandemic, and “new normal” behaviors. Online shipping data can provide better insights for demand patterns.

Along with sales data, considering the impact of marketing and promotions can provide deep insights, particularly if that data is broken down into above-the-line, below-the-line, and digital campaign information. Social media, SEO, and search trends may also be correlated to consumer behavior and provide valuable information in demand forecasting.

With all these different sources of data, it’s possible (and increasingly helpful) to employ machine learning technology to better predict future demands. Predictive models allow for real-time adjustments, so that a flawed demand forecast can be corrected with minimal financial impact. According to an analysis by McKinsey: “Applying AI-driven forecasting to supply chain management, for example, can reduce errors by between 20 and 50 percent—and translate into a reduction in lost sales and product unavailability of up to 65 percent.”


Consumer behavior and preferences continue to change rapidly, and macroeconomic conditions remain uncertain. This makes demand forecasting challenging whether you are a legacy big-box retailer or a digitally native D2C brand. This year’s holiday season will serve as a testing ground for many companies as they adjust from last year’s missteps, but the challenges around demand forecasting apply well past any single shopping season.

By leveraging all available data to make real-time decisions, organizations can prevent future derailments. This will require more advanced use of data analytics and the adoption of more flexible demand models to account for outliers and unexpected disruptions in demand patterns. It will also require more creative logistical strategies such as converting underutilized store space into e-commerce fulfillment centers.

One thing remains certain in our uncertain world: Companies must move past relying on past sales data and have a connected view of all their available data to increase the accuracy of their sales predictions and minimize the negative impact of unexpected surprises down the track.

Source: Fast Company