AI Demand Forecasting
The goal of the AI Demand Forecasting module is to create forecasts on a 15, 30, or 60-minute or daily level based on historical data. These forecasts can be used to create a headcount as a basis for scheduling. Headcount can be calculated either through Quinyx Labour Standards or Static/Dynamic Rule solution or through customer-specific calculations outside of Quinyx.
Quinyx Demand Forecasting solution incorporates the concept of hyperlocal forecasting. Hyperlocal forecasting assumes that, although different units forecast the same data variable, e.g., sales, each individual location has a unique data pattern and customer/user behavior. Therefore, a single forecasting method is not necessarily the best forecasting method for all data variables in all locations.
Quinyx Demand Forecasting solution includes multiple forecasting methods, where each method analyses historical data in a different way in order to predict future data values. Therefore some specific methods are more suitable in certain scenarios than others. The forecasting methods range from, for example, Machine Learning Models to Time Series Models. The method that provides the most accurate forecasts for a specific variable in a specific unit, based on the historical data, is selected. Each location and variable combination is therefore considered individually, and the best method is selected for each of them.
The Demand Forecasts are retrained on a regular basis, depending on at what interval your actual data is updated. Each time the forecast models are retrained, the most recent actual/historical data is incorporated, all forecasting methods are run through the method selection process, and once again, the method providing the most accurate forecasts is selected. That way, new data patterns can be captured while the forecasts remain the most accurate compared to actual data.
- There need to be input data variables and forecast configuration variables representing the variables to be forecasted configured in Quinyx
- There needs to be sufficient historical data of good quality. Sufficient historical data can mean multiple years if you have yearly or seasonal trends in your data. More historical data is required if the COVID-19 pandemic had a large impact on your data levels. Good quality data means data without too much missing data or unrealistic data.
- The Demand Forecasting module needs to be enabled and configured by Quinyx.
Configuring Demand Forecast
Aside from the prerequisites above, no specific configuration is required in order to be able to create Demand Forecasts. However, you can make sure that the forecasts consider your opening hours and any events and/or public holidays.
If you have configured opening hours, the Demand Forecasting solution can ensure that no forecasts are created outside your opening hours. This is important in cases where opening hours regularly change and where the new opening hours are shorter than the previous opening hours. Opening hours can be configured either in Quinyx or in Pythia (AI Optimization).
If you have configured events in Quinyx in order to increase or decrease the expected variable values, the effect of the event is automatically applied to the forecast variables in Quinyx. However, if you want to ensure that the optimal headcount is updated according to your events, you need to either manually run the Demand Forecasting algorithm to incorporate the event in the optimal headcount calculation, or you need to wait until the Demand Forecast is automatically run to create new forecasts.
If you have configured events in Pythia (AI Optimization) in order to increase or decrease the expected variable values, you need to run the Demand Forecasting algorithm to ensure that the effects of those events are applied. You then need to either manually run the Demand Forecasting algorithm to incorporate the event in the optimal headcount calculation, or you need to wait until the Demand Forecast is automatically run to create new forecasts.
Creating a Demand Forecast
Forecasts can be created for a single or multiple data variables either automatically or manually. Generally, the automatic frequency for the creation of the forecasts is configured by Quinyx at the initial configuration of your Demand Forecasting module. When configuring the automatic creation details, you can indicate how often the algorithm should be run, how far in advance the algorithm should create forecasts for (e.g. one month in advance), and for which period length the forecast should be created (e.g. one week at a time).
Whenever there are any changes to the algorithm input considerations, such as opening hours or events, and you would like to ensure that they are considered, you can also run the Demand Forecast manually from Quinyx or Pythia (AI Optimization) at any point in time.
Analyzing the Results
The resulting forecasts can be viewed either in Quinyx within the schedule statistics or forecast graphs and tables for the specific forecast configuration variables or in Pythia (AI Optimization).
Compare the created forecast with your expectations for that variable at each point in time. If the forecast values are higher or lower than expected, then check if there has been enough historical data to forecast based on or whether there should be an event considered which has not been added.