Forecasting with machine learning

Customers that have historical sales or other volume data spanning multiple years can leverage the power of machine learning in LFSO to create better labor forecasts. Machine learning forecasting creates forecast models for each driver by analyzing historical data to determine the impact of seasonality and day-of-the-week trends on sales or volume performance. Compared with algorithmic forecasting methodologies, such as trend of historic averages, the machine learning model leverages more than one year of historical volume data. Incorporating additional data allows machine learning forecasting to generate more precise forecasts that give customers a more accurate picture of the number of people required to staff a given store.

In addition to sales and volume data imported from Point of Sale (POS) systems, machine learning forecasting incorporates supplemental data points provided by customers or third-party services to refine its forecasting models. Customers can input past and future sales promotions in the Promotion Calendar, which are used to predict how future promotions will affect performance. Optionally, historical weather data from a third-party weather API can be integrated into forecasting models to predict the impact of expected future weather on sales and volume results.

The machine learning model runs a training process to generate forecasting models for each store and its associated drivers. For stores with multiple drivers, training generates individual forecasting models for each driver. For example, a store could have forecasting models for predicting sales, revenue, and traffic. The training process determines the best possible prediction models for each driver based on the data available at the time of training. Training is a continuous process that is performed at regular intervals to improve prediction models as new data becomes available.