Data privacy in AI
Updated
by
Leigh Hutchens
Our algorithms prioritize data privacy, storing only learned patterns and strictly limiting access to customer-specific information. Explore more about each algorithm, as detailed in our tips. Keep reading to find out more!
Demand Forecasting
The demand forecasting algorithm learns patterns in historical data and uses those learned patterns to make forecasts about how those time series are going to develop in the future. To maximize the accuracy of the algorithm, a separate model is trained on the data belonging to an individual variable. That model will only be used to make forecasts for the variable that it has been trained on. The demand forecasting algorithm does not process any user data (either when training or making forecasts) but is applied on variables such as the number of items sold, revenues, etc. The trained model itself only stores the patterns that have been learned from the time series data and does not store the data itself. The forecasts made by the algorithm are used as input for the optimal headcount calculations and thus contribute to highly accurate future headcount requirements.
The demand forecasting algorithm contains 38 different methods that are trained on the time series data, after which the most performant method is selected and used to make forecasts. The used methods contain classical time series forecasting methods, such as ARIMA, (trend) exponential smoothing, and moving average methods, but also machine learning methods, such as random forests, gradient boosting, and extreme gradient boosting.
Auto Schedule
The auto schedule algorithm is designed to take headcount requirements as input and return a set of shifts that optimally cover the requirements. The algorithm tries to find the shifts and breaks that follow a set of rules (shift lengths, break rules, etc) and objectives (service level, costs, employee happiness). The rules and objectives are set by our customers and are an input to the algorithm that is only used for that specific customer. No data is shared between customers and old schedules are not used to learn from. The algorithm is fully deterministic: provided the same input, the algorithm returns the same output.
Auto Assign
The purpose of the auto assign algorithm is to assign a set of anonymous (unassigned) shifts to a set of employees. The assignment is done based on a set of rules that can be configured. The set of rules usually consists of labor laws, as well as customer-specific scheduling rules. The goal of the algorithm is to do the assignment in the most optimal way. Depending on the configuration, the algorithm can optimize for costs, employee happiness, proficiency, or a mix of those.
The rules and objectives are set by our customers and are an input to the February 23, 2022 algorithm. The shift and employee data is also input to the algorithm and does not contain any sensitive information, such as names, phone numbers, etc. This data (including the rules and objectives) is strictly limited to the customer to whom it belongs. After solving the optimization problem, the algorithm returns the set of assigned shifts, shifts that are not assigned to an employee, and potential violations in any of the defined rules
Additional information
You can read even more:
