Automated Demand Forecasting with ML Models

Demand forecasting using machine learning (ML) models involves predicting future customer demand for products or services. If you are performing demand forecasting with ML models you are always looking to increase the accuracy of your forecast. Events have a big impact on demand and updating your models to account for events can significantly improve the accuracy of your forecasts.

Demand Forecasting Tutorial

See our tutorial on updating machine learning models with predictive event data - Automatically and dynamically update your forecast by integrating PredictHQ data directly into your demand forecasting models.

The Power of Accurate Demand Forecasting

Let’s take a look at an example. In 2019 in London when both Wimbledon and a British Summer Time festival were held we found a major impact to accommodation businesses. The two events caused a demand increase of close to 40% on the days when the events were occurring. When Dreamforce happens in San Francisco we see a 100-110% increase in demand for some hotels in the area and some coffee chains see up to 140-190%. In other industries, retailers in Iowa, for example, saw demand for beer increase by around 180% on Saturdays when multiple college sports games were taking place nearby. For parking, we’ve seen over a 60% increase in demand for London parking locations when several dozen concerts and performing arts shows were taking place within a 3-mile radius of the location.

Where customers have updated their demand forecast to account for events we’ve seen MAPE improvements in the range of 5-10% and some even greater. For example, Favor has seen a 5 - 6% improvement in MAPE.

Some customers try to source and manage event information themselves. What they commonly find is that there are many different types of events from many different sources. Sports events are different from concert events, holidays are different from severe weather events, and conferences are different from public holiday events. We provide event data from 19 different categories via our Features API and Events API. We ingest data from hundreds of providers and we provide a single product for you to integrate with to model the impact of all these different types of events.

Figuring out how to use events in your forecast can be complex. With our product and tools, we aim to reduce any R&D effort and enable you to quickly and easily identify event-based ML features for use in your models. Our Features API provides prebuilt, forecast-ready event-based ML features for your forecast. We have seen this reduce R&D and feature engineering time for months to days for our customers.

Demand forecasting covers a range of use cases and industries from dynamic pricing to inventory management. See the use cases below for more examples.

Last updated

© 2024 PredictHQ Ltd