Updated on June 10, 2026
Editor’s note (updated June 10, 2026): Although this article was originally written during COVID-19, the same forecasting challenges apply today whenever organizations experience abnormal demand patterns caused by tariffs, supply disruptions, promotions, weather events, or market volatility.
While COVID-19 created one of the most visible examples of abnormal demand, the same forecasting challenges appear whenever markets behave outside historical norms. Today, organizations face similar issues from tariffs, supply shortages, weather disruptions, promotions, and sudden changes in customer behavior.
Why abnormal demand hurts forecast accuracy
COVID-19 created one of the largest demand disruptions in modern supply chains, causing demand patterns to shift dramatically across industries.
Demand forecasting works best when historical patterns remain reasonably stable over time. Most statistical forecasting methods assume that historical sales contain useful signals about seasonality, growth, and customer behavior.
But during periods of abnormal demand, that assumption breaks down.
Events like COVID-19, tariff changes, supply disruptions, promotions, or sudden market shifts create demand patterns that are unlikely to repeat in the same way. When those periods remain untouched in historical data, forecasting systems can mistake temporary behavior for a lasting trend.
The result: inflated forecasts after demand spikes, suppressed forecasts after demand collapses, and lower forecast accuracy long after the disruption ends.
Examples of abnormal demand during COVID-19
The challenge wasn’t simply that demand changed — it was that demand changed in ways traditional forecasting models would interpret incorrectly.
Groceries: Demand spike → future overforecasting
The grocery industry saw a massive demand increase during the abnormal demand periods due to concerns over grocery availability. If we don’t adjust demand history during these periods, it will overestimate the demand for the same period the next year. Left untreated, these demand spikes can cause forecasting models to overestimate recovery demand far into the future.
Transportation: Demand collapse → future underforecasting
As consumers began working from home and avoiding non-essential travel, demand for transportation services dropped significantly. If we don’t adjust demand history, any AI forecast engine will underestimate demand for the same period in the next year. Left untreated, these demand drops can cause forecasting models to underestimate recovery demand well into the future.
Once abnormal periods become part of demand history, forecasting teams have two choices: manually adjust the data or use automation to estimate what normal demand would likely have been.
Three ways to adjust demand history
So how can you account for pandemic outliers to ensure better forecast accuracy in future planning periods?
1. Manual demand adjustment
This approach is the simplest. It gives you the most control but is least scalable.
Adjust historical demand item by item during abnormal demand periods. Although this method would work theoretically, it’s neither efficient nor practical. Baseline demand for each item would have to be manually “forecasted” during these periods.
2. Exclude abnormal periods
This approach is fast but risks losing useful signals: automate the historical demand adjustment process. For example, you can set up a flag to inform the AI forecasting engine to ignore sales history data during an abnormal demand period. A good forecasting engine should develop an algorithm to estimate baseline demand of an abnormal period.
3. Estimate normalized demand (recommended)
The best approach preserves seasonality and growth while removing distortion.
Ideally, you would use seasonal and yearly growth to project baseline demand. For example, if sales for March 2019 were 100 cases and year‑over‑year growth was 5 %, then a good forecast for March 2020 would be 100 × (1 + 5 %) = 105.
| Month | Actual | Adjusted |
| March 2019 | 100 | 100 |
| March 2020 (pandemic spike) | 160 | 105 |
This number would replace the actual sales history of March 2020 within the pandemic period, which will then be used to “forecast” or estimate future demand.
How New Horizon approaches abnormal demand
New Horizon has incorporated this approach in its statistical forecast engine. Using AI to estimate baseline demand during these periods, we found that our approach is both extremely efficient and accurate.
Want to see how New Horizon identifies abnormal demand and improves forecast accuracy automatically? Talk with our team to see how AI-driven forecasting handles demand disruptions without manual cleanup.