Cognitive Load Optimization for Contact Center Agents Using Real-Time Monitoring and AI-Driven Workload Balancing
Citation. Arcot, S. V. (2023). Cognitive Load Optimization for Contact Center Agents Using Real-Time Monitoring and AI-Driven Workload Balancing. International Journal of Scientific Research in Computer Science, Engineering and Information Technology, 9(2), 863–879. https://doi.org/10.32628/cseit2342436
Summary. This article reports a 12-month, four-site randomized trial of an AI system that used wearable physiological signals and voice-stress features to estimate contact-center agents’ cognitive-load and stress states, then dynamically routed work. It reports gains in real-time stress classification, first-call resolution, stress reduction, turnover, and agent acceptance. The article positions continuous worker-state monitoring as both a workload-allocation mechanism and a privacy-sensitive workforce-AI design problem.
Key claims
- The article reports that multimodal models using HRV, galvanic skin response, skin temperature, movement, and voice features classified stress states with over 89% real-time accuracy, improving to 91.2% with continuous learning.
- In the reported 500-agent, 12-month randomized trial across four contact-center sites, optimized workload distribution was associated with a 15% improvement in first-call resolution, a 28% reduction in stress levels, and a 22% decrease in agent turnover.
- The proposed system uses inferred cognitive load and stress as dynamic inputs to call routing and workload balancing rather than treating task allocation as a fixed capability decision.
- The article reports 92% agent adoption, while also identifying sensitive biometric and voice data as requiring consent, data minimization, access controls, and privacy-preserving processing.
Evidence & limitations
- The article describes a randomized controlled trial with 500 full-time, English-speaking contact-center agents across financial services, healthcare support, technical assistance, and customer-service sites; it combines physiological wearables, voice analytics, operational metrics, and surveys.
- The unusually broad performance, well-being, turnover, and acceptance claims are reported by a single article. The record does not provide independently verified replication, raw data, a preregistration, or an external evaluation of the cognitive-load and stress measures.
- The source is a journal article with Crossref metadata and a DOI, but the ingested PDF has internal bibliographic inconsistencies: its first-page issue line says Volume 10, Issue 2 while running headers identify Volume 9, Issue 2. This record uses Crossref’s 2023 journal-article metadata, volume 9(2), and pages 863–879 as canonical provenance.
- Publisher landing-page access returned HTTP 403 to this environment, so public DOI/Crossref metadata verifies public provenance but an exact publisher-PDF hash comparison was unavailable.