Forecasting

AIDRA forecasting aims to predict and anticipate demand for space in the office. To do this, AIDRA considers various historical and future data points that help define the probability of an individual going to the office, and the total number of spaces to allocate to each team on a given day.

The main driver for forecasting is attendance data. At the individual level, AIDRA analyses the pattern of attendance, weighing more heavily the attendance on recent dates. If an individual has recently started coming to the office more frequently, this is more heavily weighted, so the probability of attendance is increased, even if there is a long history of not going to the office. AIDRA supports attendance forecasting at a daily level, so each day of the week might have a different probability of attendance for an individual. Similarly, using the predicted team associated with an individual’s attendance, AIDRA assigns the most likely team for the individual depending on the day of the week.

An individual can have different likely teams on different days of the week ie, the Accounting team on Mondays, and the Finance team on Wednesdays. If an individual is forecasted to be in the office with their usual team, but then schedules in to work as part of another team, the forecast for the usual team will be 0% for that date. Each individual is assigned a permanent team, which is defined by the organisation and imported via integration. If an individual changes permanent team, their individual forecast transfers to the new permanent team.

At the team level, team attendance is the main factor affecting time series algorithms used to recommend a target for the team. The time series model combines attendance, intentions (requests for space from individuals when they schedule into the office), and headcount over time to predict the next target for a team.

Individual and team forecasting models take into consideration day of the week, trend and yearly seasonality (when enough data is available), and new predictions are generated at least daily after re-training the model with new data.

Additional data sources such as area or desk sensors can provide additional information, such as areas and zones visited by an individual. This helps to better identify the team an individual worked with, making the prediction of the team associated with an individual’s attendance more accurate.