AI-Assisted Screening of Citizen Complaints and Requests

The Problem

The Ombudsperson's Office is responsible for analysing thousands of citizen submissions each year, many of which contain complex narratives, allegations of irregularities, requests for public services, or sensitive personal information. The initial screening process requires analysts to evaluate multiple factors simultaneously, including the type of submission, institutional jurisdiction, legal requirements, privacy considerations, and the appropriate course of action.

This work is essential but highly repetitive and resource-intensive. The growing volume of submissions increased the risk of inconsistencies in analysis, longer processing times, and the allocation of highly qualified staff to routine tasks that could potentially be supported by technology.

The challenge was therefore twofold: improving the efficiency and consistency of the screening process while preserving the quality, accountability, and human judgment required for public-sector decision-making. We also sought to understand how Generative AI could be applied responsibly in a sensitive public-service environment where transparency, fairness, and human oversight are critical.

The Solution

To address these challenges, we developed a Generative AI agent to support the initial technical screening of citizen submissions received by the Ombudsperson's Office. Rather than replacing human judgment, the solution was designed as a decision-support tool to assist analysts in performing repetitive and time-consuming assessment tasks more consistently and efficiently.

The AI agent was trained to analyse citizen submissions and provide structured support in key aspects of the screening process, including identifying the type of submission, assessing institutional jurisdiction, detecting sensitive information, highlighting potential indicators of irregularities, suggesting appropriate handling procedures, and generating standardised technical summaries.

We chose this approach because the screening process involves the analysis of large volumes of unstructured text and requires the application of multiple legal, procedural, and operational criteria. Generative AI offered an opportunity to increase consistency, reduce the time spent on routine activities, and allow public servants to focus on tasks requiring specialised expertise, critical thinking, and human judgment.

The project was implemented as a pilot initiative with human oversight at every stage. All recommendations and outputs generated by the AI remain subject to review, validation, and final decision-making by public servants. This approach enabled us to explore the benefits of Artificial Intelligence while maintaining accountability, transparency, and compliance with public-sector governance requirements. This project costs approximately $30 a month.

The Impact

1. The ability to analyse complex submissions and generate structured outputs

Although the project is still in its pilot phase, it has already demonstrated significant potential to improve the efficiency, consistency, and quality of the Ombudsperson's Office screening process. The AI agent has shown the ability to analyse complex citizen submissions and generate structured outputs that support analysts in identifying submission types, assessing jurisdiction, detecting sensitive information, and preparing technical summaries. This has reduced the effort required for repetitive assessment tasks and promoted greater consistency in applying screening criteria.

2. Valuable organisational learning on responsible use of Generative AI

Beyond operational gains, the pilot has generated valuable organisational learning regarding the governance and responsible use of Generative AI in public administration. It has helped the organisation better understand the opportunities, limitations, and risks of applying AI to sensitive public service processes, while reinforcing the importance of human oversight and validation.

3. Contributing to a broader cultural shift toward data-driven approaches

The project has also contributed to a broader cultural shift within the organisation, encouraging the exploration of data-driven and AI-assisted approaches to improve public service delivery. As the initiative progresses, the team expects to measure its impact more systematically across areas such as processing time, analysis consistency, and resource allocation.

Lessons Learned

• The AI agent added the most value in structuring unstructured input. One of the most successful aspects of the project was the AI agent's ability to process lengthy, unstructured citizen submissions and transform them into structured analyses that support Ombudsperson analysts' work. The tool demonstrated particular value in identifying key information, suggesting classification criteria, and generating standardised summaries, thereby increasing consistency across the screening process.
• Testing Generative AI in a real public-sector environment provided practical governance experience. Another positive outcome was the opportunity to test Generative AI in a real public-sector environment and gain practical experience with governance, risk management, and human oversight. The pilot reinforced the importance of using AI as a support tool rather than as a replacement for human judgment.
• Complex and ambiguous cases exposed the limits of current AI models. One of the main challenges was dealing with highly complex or ambiguous cases that require contextual interpretation, legal reasoning, or nuanced assessment. These situations highlighted the limitations of current AI models and confirmed the need for human review and validation of all outputs. The team also learned that the quality of the results depends heavily on prompt design, testing, and continuous refinement.
• Earlier investment in validated test cases and user involvement would strengthen the approach. If starting the project today, the team would invest earlier in building a larger repository of validated cases for testing and evaluation, involve end users more extensively throughout the design process, and develop more robust mechanisms for continuous performance monitoring. Greater emphasis would also be placed from the outset on measuring outcomes such as time savings, improvements in consistency, and user adoption to better quantify the project's impact.

The project was developed within the regulatory framework of Brazil's Federal Executive Branch Public Ombudsman System, which is coordinated by the Office of the Federal Ombudsman-General, part of the Office of the Comptroller General of the Union (CGU). The Federal Ombudsman-General is responsible for establishing the guidelines, standards, and procedures that govern public ombudsman activities across the federal government. The initiative operates within the legal framework established by Brazil's Access to Information Law (Law No. 12,527/2011), the Code for the Protection and Defence of Public Service Users (Law No. 13,460/2017), and other regulations governing transparency, citizen participation, public accountability, and the handling of citizen feedback. The use of Artificial Intelligence was designed to support, rather than replace, human decision-making, ensuring compliance with principles of accountability, transparency, and responsible public-sector AI adoption.