# A Survey of Passive Sensing in the Workplace

**Authors:** Subigya Nepal, Gonzalo J. Martinez, Shayan Mirjafari, Koustuv Saha, Vedant Das Swain, Xuhai Xu, Pino G. Audia, Munmun De Choudhury, Anind K. Dey, Aaron Striegel, Andrew T. Campbell  
**arXiv:** [2201.03074](https://arxiv.org/abs/2201.03074)  
**Date:** 2022-01-09  
**Summary page:** https://www.summarizepaper.com/en/paper/2201.03074v1/

## Key points

- Workplace expected to undergo significant changes with advancing technology
- Passive sensing technology can assess human behavior and improve cognitive and physical capabilities at work
- Research focuses on using passive sensing to evaluate employee wellbeing and productivity
- Das Swain et al. study used Bluetooth beacons to analyze desk and away from desk sessions, time spent at work and home, measured organization fit
- Mirjafari et al. introduced organizational personas by clustering personality facets with activity facets, found activity facets statistically significant in individual task performance
- Mirjafari et al. classified individuals into higher or lower performers using fitness trackers, phone agents, and beacons; achieved high discrimination ability with combined features from both sources
- Mirjafari et al. used passive sensing data to predict day-to-day job performance dynamics with a 75% F1 score; employed Gradient Analysis for actionable interpretations of results

## Summary

As new technology continues to advance, the workplace is expected to undergo significant changes in the future. One area that shows promise for improvement is the use of passive sensing technology to assess human behavior and promote better cognitive and physical capabilities at work. Recent research has focused on utilizing passive sensing in the workplace to evaluate the wellbeing and productivity of employees. One notable study conducted by Das Swain et al. involved 249 individuals over a period of 62 days. Bluetooth beacons were used to analyze desk and away from desk sessions per hour, as well as time spent at work and home. The researchers found that by analyzing individual routines and comparing them to their organization's normative behavior, they could measure organization fit. This measure, along with Big Five Personality measures, predicted up to 28% of the variance in Individual Routine Behavior (IRB), with routine fit showing a significant positive association with IRB. Another study by Mirjafari et al. examined 603 information workers over 110 days. They introduced the concept of organizational personas, which involved clustering personality facets constructed from Big Five Personality data with activity facets constructed from multi-modal data collected through fitness trackers, phone agents, and Bluetooth beacons. While activity facets accounted for a small portion of the variance in Individual Task Performance (ITP), they were statistically significant. Mirjafari et al. also conducted a similar study involving 554 information workers using a phone agent, beacons, and a Garmin fitness tracker. By applying K-Means clustering on participant averages of various performance scores, they were able to classify individuals into higher or lower performers within their organizations. Results showed that features from fitness trackers alone could discriminate performers with an AUROC (Area Under Receiver Operating Characteristic) score of .72; features from phone agents achieved an AUROC score of .65; and combining features from both sources increased discrimination ability to an AUROC score of .83. In another study by Mirjafari et al., 298 information workers were assessed using passive sensing data from mobile phones and a Garmin fitness tracker. The researchers used auto-encoder generated features to predict day-to-day job performance dynamics, achieving an F1 score of 75% with the final model. They also employed Gradient Analysis, a technique commonly used in computer vision, to provide actionable interpretations of the model's results. Nepal et al.