Navigating Algorithmic Fallibility: Evaluating Implementation and Guidance Strategies in AI-Augmented Service Frontlines

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Despite substantial investments in collaborative human–artificial intelligence (C-AI), implementation outcomes remain mixed due to people- and leadership-based hurdles. Thematic qualitative analyses of frontline employees and their managers reveal two implementation strategies related to frontline employee empowerment and AI use guidance. Drawing on agency and cognitive evaluation theories, we conceptualize C-AI implementation as a governance design challenge. We examine how empowerment and AI use guidance shape frontline AI compliance decision-making, especially when AI fallibility manifests through incorrect recommendations. Across a large-scale experimental chat simulation involving 730 frontline employees and 6,500+ decisions in AI-assisted customer service and using hierarchical mixed-effects models, we demonstrate that outcome-control-oriented empowerment increases customer-centric decisions and reduces compliance with erroneous AI recommendations. Prescriptive guidance improves baseline decision quality by clarifying evaluative criteria but can attenuate empowerment’s protective effect by reinforcing procedural conformity. Importantly, exposure to incorrect AI recommendations significantly reduces subsequent customer-centric decisions, suggesting that algorithmic fallibility destabilizes trust and disrupts frontline employee decision calibration. Empowerment partially buffers this post-error decline, whereas prescriptive guidance exacerbates it. Collectively, the findings highlight that effective C-AI implementation requires balancing algorithmic monitoring with calibrated frontline employee discretion to sustain customer-centric outcomes.