Name in original language
Vinculação de Manifestações de Ouvidoria a Serviços Públicos com Uso de Inteligência Artificial
Initiative overview
Fala.BR is the single-entry platform for ombudsman submissions to Brazil’s Federal Executive Branch. All federal institutions operate their ombudsman units through this centralised system, which has received more than five million submissions in the past five years. Despite this scale, fewer than 3% of submissions are currently linked to a specific public service. Without this connection, citizen reports cannot be systematically used to assess service performance or identify concrete opportunities for improvement from the user’s perspective. Manual classification at this scale is not feasible.
The initiative seeks to automatically identify which public service each submission refers to and tag it in a structured way. By linking citizen messages to specific services, the solution enables targeted analysis of recurring problems, unmet expectations and service delivery gaps. This supports evidence-based improvements and reduces the distance between service users and service managers.
As the central body of the Federal Executive Branch Ombudsman System (SisOuv), the Office of the Comptroller General (CGU) has system-wide visibility over submissions and is uniquely positioned to implement this solution. The AI model uses both official service descriptions from the Gov.BR portal and the text written by citizens to establish the most relevant match. The initiative covers both historical data and new submissions.
It is being integrated into the Fala.BR platform so that, at the time of registration, citizens receive suggestions of the public service most closely related to their message, which they can confirm or adjust. This structured linkage will enable all federal institutions to access service-level analytics, strengthening monitoring, service evaluation and continuous improvement. The initiative complies with Brazil’s General Data Protection Law and existing ombudsman regulatory frameworks.
Results, outcomes and impacts
The AI model was validated on more than 61,000 ombudsman submissions across 227 public services, achieving 81% top-1 and 89% top-2 accuracy in service matching. For high-volume services (over 1,000 submissions), accuracy reaches 95.9%. Results are already used internally by CGU to support oversight and service monitoring. Once the five-million historical backlog is classified and real-time tagging is implemented, federal institutions will gain structured visibility into how citizen submissions relate to their services, strengthening evidence-based service improvement.
