- Automatic Forecast predictions
- The Forecast algorithms
- Where will the prediction be made visible and how?
- How do I activate this?
- How do I configure the algorithm?
- What other configuration is needed?
- How do I initiate it manually?
- How do I follow up on the historical runs and logs?
Automatic Forecast predictions
As of Neo version 0082, we have made it possible to have your Forecast predictions in the future to be done automatically based on your historical data.
The Forecast algorithms
There are currently three different methods you can use to generate Forecasts, two are performed by Quinyx, and the third is an option to connect a third party prediction service to Quinyx, allowing the forecast to be generated elsewhere then presented in Quinyx.
When adding a new Forecast algorithm (see How do I configure the algorithm below), you are presented with three choices:
Last Year Same Day
This algorithm automatically copies the actual data for your forecast driver from the previous year as a forecast for the current year-to-date ahead. You have an option to add a modifier as an increase or decrease in percentage on the previous years data.
Quinyx Forecasting algorithm
This is a machine learning algorithm that intelligently and automatically predicts the forecast for your selected driver for the next 90 days from when it is run. You can also configure this method to re-run automatically on your preferred day. This method looks at all available actual data for the selected forecast driver in order to determine the suggested forecast.
You must have at least 60 days worth of historical data for your forecast driver in order to use Quinyx Forecasting Algorithm.
When forecasting a driver (e.g. sales or footfall) for a specific unit, the current algorithms use only data pertaining to that unit. Algorithms try to find regular patterns (seasonalities) and the overall trend. Seasonalities can occur on different time scales. For example, you can have a yearly seasonality (e.g. sales are higher around Christmas, lower during summer), weekly seasonality (e.g. sales are higher on weekends) or daily seasonality (e.g. sales are higher at lunchtime). Multiple seasonalities can be present for the same time. Many units experience significant holiday effects and the algorithms try to estimate that by using data from the same holiday of previous year(s). In general, the algorithms will make more accurate forecasts when they have more historical data. Quinyx recommends 24-36 months of data. However, with 60 days of data, a forecast will be produced. Currently, the algorithm only make 90 days forecast prediction into the “future” from the last data uploaded.
This method allows a third party solution to push forecasted data into Quinyx through the API using POST /predicted-data:
With this operation, you can upload predictions from external tools used for predicting sales, transactions, etc., that optimal staffing rules will use to define staffing needs:
- externalForecastVariableId - ID for a variable as set up for units in Account settings→Forecast→Variables→External ID.
- externalForecastConfigurationId -
- externalUnitId - External ID for a unit as set up for units in Classic → Settings → Tables → integration keys
- externalSectionId (optional) - External ID for a section as set up for units in Classic → Settings → Tables → integration keys.
- runIdentifier - A value to be set if external system sends multiple predictions for the same date range.
- runTimestamp - A value to be set to define when the predicted data was generated. In the future, Quinyx will use this value to compare different predictions if they exist.
- Data - the amount of the external prediction.
- Timestamp - the time of where the amount should be placed in UTC format 2020-04-09T20:45:00+01:00.
Where will the prediction be made visible and how?
The outcome of the algorithms prediction will be used in statistics variable as the prediction for that specific variable. If you would have a driver that is sales, when you have uploaded 24-36 months of historical data so that the algorithm has many seasonalities to work with it will produce this as a row in the table view in statistics or a line in the graph view.
Read more about forecast in statistics here.
How do I activate this?
If you already have Neo Forecast up and running today please contact support and they will help you with the behind the scenes activation of the algorithms. After that is done you will only have to follow the instructions that follow.
If you do not have Neo Forecast today but is eager to know how using automated predictions together with optimal staffing rules to help your managers make even better schedules for your employees, then contact email@example.com
How do I configure the algorithm?
The algorithm is configured by creating a new configuration. This is done from Account settings -> Forecast configuration -> Add new forecast configuration
Forecast formula now offers the choice "Quinyx forecasting algorithm", when selecting that formula you are presented with additional choices. This to be able to schedule when the algorithm runs
What other configuration is needed?
In order for the algorithm to be able to take public and bank holidays into account the country needs to be correctly configured on the unit(s). Read more about how to configure this here
How do I initiate it manually?
In some cases you might not want to wait a full week before the algorithm to run. This could be because of testing during an implementation phase or that lots of new input data has been added historically that may affect the future predictions. This is available from Account settings -> Forecast configuration and by interacting with the more details icon.
In this detail panel you will be able to see the statuses of the previous runs and can initiate a new run
How do I follow up on the historical runs and logs?
You will be able to follow up the log from Account settings -> Forecast configuration and by interacting with the more details icon as described above.