Peru's electoral authority says its generative-AI assistant cut candidate-registration processing from four hours to six minutes; the system also records an auditable trail of manual changes, though the claim rests on a single institutional case with limited independent validation.

Main Finding

EleccIA, a generative-AI–based electoral judicial assistant implemented by Peru’s Jurado Nacional de Elecciones (JNE), materially speeds up electoral administrative adjudication: average automated calification time fell from 240 minutes to 6 minutes per candidate-list expediente, with a subsequent human review of ~20 minutes. The JNE reports >97% reduction in operational workload and improved traceability of manual changes, while final legal decisions remain under human responsibility.

Key Points

• Purpose and scope
  • EleccIA automates qualification of registration files for candidate lists, standardizes validation criteria, proposes draft resolutions and generates summaries (e.g., program summaries for parties); political organizations retain final editorial control.
  • Human adjudicators (Asistentes jurisdiccionales) retain final decision authority and perform review of AI outputs.
• Measured operational gains
  • Per-file automated processing time: 240 → 6 minutes.
  • Human review time per file after AI output: ≈20 minutes.
  • Institutional claim: >97% reduction in workload; capacity to process many more cases per workday.
• Ecosystem and governance elements
  • Components: citizen-centered design, legal compliance module, local secure database and cybersecurity, specialized human adjudicator, ethical conduct requirements.
  • Design guided by AI governance/ethics principles (transparency, explainability, fairness, security, privacy, accountability, robustness — e.g., ASEAN guide).
• Risks and limitations noted by the author
  • Algorithmic opacity (black-box models), possible bias amplification, digital-divide concerns, need for legal/regulatory frameworks and multidisciplinary oversight (lawyers + engineers).
• Comparative context
  • Estonia used as an illustrative example of advanced e-government and AI use in public administration (tax, elections) to highlight benefits and constraints of digitalizing electoral services.

Data & Methods

• Study type: single-case qualitative/administrative case study of EleccIA implementation at the JNE.
• Data sources used or reported in the paper:
  • Administrative process metrics supplied by JNE (processing time per expediente before and after EleccIA).
  • Institutional descriptions: JNE roles, candidate counts (10,257 candidates across 39 parties; 208 elective positions), and system architecture (legal and technical components).
  • Reference to governance frameworks and comparative literature (Estonia, ASEAN guide, academic works on AI governance).
• Methods and analysis:
  • Descriptive accounting of workflow changes, time comparisons, and organizational impacts.
  • Normative discussion of ethical and regulatory frameworks for public-sector AI adoption.
• Methodological limitations (reported or evident):
  • No randomized or controlled evaluation; evidence is observational/administrative.
  • No reported accuracy/quality metrics for AI decisions (error rates, false positives/negatives, disagreement with human adjudicators).
  • Limited information on costs (development, deployment, maintenance), datasets used to train/validate models, or external audit results.
  • Generalizability may be limited to Peru’s legal and institutional context.

Implications for AI Economics

• Productivity and cost structure
  • Large per-case time reductions imply substantial labor productivity gains and potential reductions in marginal administrative cost per expediente. Economists should monetize time savings against development and running costs to assess net welfare gains.
  • Fixed-cost nature of AI development suggests increasing returns to scale for large caseloads (e.g., thousands of candidacies).
• Labor and task reallocation
  • Routine, time-consuming validation tasks can be automated, shifting human labor toward oversight, complex adjudication, appeals, and quality control. This reallocation affects labor demand and required skill composition in electoral administration.
• Public-sector service delivery and welfare
  • Faster processing improves predictability and legal certainty, which may reduce transaction costs for parties and candidates, and lower litigation or backlog externalities—potentially increasing system efficiency and public trust.
  • However, if automation embeds systematic biases, it can distort electoral competition with economic and political distributional consequences.
• Investment, regulation, and institutional complementarities
  • Realizing benefits requires investments in secure local data infrastructure, continuous model validation, transparency/audit mechanisms, and staff retraining—these are recurring costs that affect cost–benefit calculations.
  • Regulatory frameworks (transparency, accountability, data protection) impose compliance costs but are essential to preserve legitimacy; economists should include regulatory and governance costs in impact assessments.
• Externalities and empirical research agenda
  • Important externalities to measure: effects on voter confidence, litigation rates, speed of dispute resolution, incidence of appeals, and any systematic changes in candidate eligibility outcomes.
  • Suggested empirical analyses:
    • Cost–benefit analysis: time savings × wage rates minus AI development/ops + governance costs.
    • Accuracy and fairness audit: compare AI recommendations vs. human-only decisions, track reversals on appeal, measure demographic/party-specific error patterns.
    • Causal inference: difference-in-differences or randomized audits across jurisdictions or case types to estimate causal impacts on processing time, costs, and legal outcomes.
    • Distributional impact: who gains/loses (staff, parties, small vs. large organizations), and potential political economy effects.
• Policy takeaways for economists and policymakers
  • EleccIA illustrates substantial marginal productivity potential of generative-AI tools in public administration, but rigorous impact evaluation (accuracy, bias, costs, and legitimacy) is essential before scaling.
  • Design of public-AI deployment should budget for monitoring/auditing, transparency mechanisms, human-in-the-loop safeguards, and workforce retraining to capture net social benefits while limiting risks.