Review of 85 studies: HR algorithms tied to efficiency, and to lower gig-worker well-being

The picture split by setting. Recruitment and selection showed the strongest link (0.35), followed by retention management (0.31) and performance management (0.18). In platform work, algorithmic management scored 0.42 on efficiency but -0.25 on employee well-being and job satisfaction.

The results vary widely between studies (I² = 74%), so the average does not apply evenly everywhere. The authors say AI also carries risks of systemic discrimination, privacy invasion and procedural injustice, and they propose a Human-in-the-Loop model for HR decisions.

The authors followed PRISMA guidelines and used the Hunter and Schmidt psychometric meta-analysis method, which corrects for measurement and sampling error. They checked publication bias with a funnel plot and Rosenthal's fail-safe N, and rated primary-study quality with an adapted Newcastle-Ottawa Scale.

Studies had to report quantitative data on decision quality, efficiency or employee attitudes, with enough statistics to compute effects. Qualitative-only studies fed the systematic review but not the meta-analysis. The subgroup analyses split studies by HR function and by AI maturity, from descriptive to prescriptive and autonomous systems.

The strongest efficiency results came from standardised, high-volume tasks such as candidate screening and scheduling, while performance management, which relies on proxies like keystrokes and active time, did worst. The authors link this to metric fixation, where employees game the proxy, and to worker resistance to being monitored algorithmically.

In the discussion, the authors warn that learning from biased historical hiring data can replicate past inequalities. They also say organisations adopting platform-style AI may inherit the precarity of gig work. Their practice recommendations include model documentation, data minimisation, analytics training for HR staff and tailored labour standards for platform work.