Uncovering the training paradox: nonlinear and context-dependent effects of training on frontline performance in a public service organization

Purpose Training is widely regarded as a key driver of organizational performance; however, its performance consequences may be nonlinear and context dependent in frontline public service settings. Drawing on paradox theory, human capital theory and public management scholarship on operational capacity, this study examines whether training can both support capability development and coincide with short-term performance strain within frontline operational teams. Design/methodology/approach Using a cross-sectional dataset from 1,144 postal units within a national public service organization, this study employs gradient-boosted machine learning models, supplementary theory-driven robustness checks and SHAP (SHapley Additive exPlanations) to examine predictive nonlinear associations between training intensity and team performance, as well as the contextual conditions shaping these patterns. Because the data are cross-sectional and observational, all estimates are framed as predictive, non-causal evidence. Findings Results indicate a nonlinear, inverted U-shaped association between training intensity and team performance. Moderate levels of training are associated with higher KPI completion, whereas very high levels of training are associated with lower KPI completion, consistent with workload displacement and time-resource disruption. The pattern is strongly conditional on context: training is more favorable in units with sufficient staffing capacity and higher employee education levels, while it becomes weaker or unfavorable in understaffed or lower-capability units. Managerial experience further strengthens the conversion of employee human capital into performance. Research limitations/implications Predictive results indicate a nonlinear, inverted-U-shaped association between training intensity and team performance. Moderate levels of training are associated with higher KPI completion, whereas very high levels of training are associated with lower KPI completion, consistent with workload displacement and time-resource disruption. The pattern is strongly conditional on context: training is more favorable in units with sufficient staffing capacity and becomes weaker or unfavorable in understaffed units, while employee education profiles and managerial experience further condition predicted performance. Practical implications The findings highlight the importance of aligning training intensity with staffing capacity, employee capabilities and managerial coordination. Organizations should avoid overly concentrated training schedules, assess readiness before deployment, and treat temporary KPI declines as signals that may require workload adjustment rather than as definitive evidence that training has failed. Originality/value This study provides large-scale public-sector evidence on nonlinear training-performance dynamics and demonstrates how explainable machine learning can be used cautiously and transparently to enrich theory development in human capital, public management and operations research.