Machine Learning and Generative AI in Administrative Processes in Peru: Administrative Efficiency in the National Public Sector

Miluska Odely Rodriguez Saavedra, Juliana Mery Bautista Lopez, Wilian Quispe Nina, Antonio Víctor Morales Gonzales, Iván Cuentas Galindo, Luis Miguel Campos Ascuña, Anthony Stefano Saenz Colana, Robinson Bernardino Almanza Cabe, Paola Gabriela Lujan Tito, Sharon Veronika Liendo Teran

Public organizations in Peru have committed substantial resources to artificial intelligence over recent years, yet evidence on whether these investments produce measurable returns has remained scarce. This study evaluated the causal impact of AI adoption on administrative efficiency across 20 Peruvian national public organizations, using a quasi-experimental design combining Difference-in-Differences with Propensity Score Matching, complemented by XGBoost version 1.7.6, Random Forest, GPT-4, and SHAP explainability analysis. The sample comprised 428 civil servants across treatment and control organizations. Results showed significant efficiency gains as perceived by civil servants through validated Likert instruments: work absenteeism decreased by 9.4%, processing times by 8.7%, and administrative costs by 18.2%, all at p < 0.001 with Cohen’s d ranging from 0.55 to 0.90. The convergence between DiD and PSM estimates supports a causal reading of these effects. Four of five hypotheses were supported. AI delivered comparable efficiency gains regardless of institutional complexity, so H2 was not confirmed. Digital infrastructure significantly moderated AI effectiveness (H3: r = 0.198, p = 0.004). Higher resistance to change was significantly associated with lower efficiency outcomes (H5: r = −0.256, p < 0.001), reinforcing the role of proactive change management as a positive moderator of AI effectiveness. SHAP analysis revealed that training investment, specialized IT personnel, and resistance management together explained 51% of predictive importance, outweighing structural variables such as budget size or geographic location. These findings provide the first systematic causal evidence on AI efficiency in Peruvian public administration and offer actionable benchmarks for comparable middle-income public sectors.