AI in Sanitation Infrastructure Auditing

Brazil has a historic deficit in basic sanitation: approximately 35 million people lack access to treated water, and more than 100 million live without sewage collection. To reverse this scenario, the New Sanitation Framework (Law 14.026/2020) established ambitious goals—universalizing access by 2033. To meet these targets, investments on the order of R$ 500 billion are needed in infrastructure projects, including water and sewage treatment plants (WTP and STP). In 2025, the Federal Court of Accounts (TCU) analyzed 120 sanitation construction contracts and found evidence of overpricing in 28 of them—a 23% rate that raised the alarm about the need for new control tools (TCU, operational audit report 2025).

Traditional oversight of sanitation projects has always relied on in-person inspections, analysis of budget spreadsheets, and comparison with reference tables such as SINAPI (the National System of Construction Cost and Index Research). This model, although necessary, is slow and reactive. By the time a problem is identified, the project has already advanced, the money has already been paid, and correction becomes expensive or unfeasible.

It is in this context that artificial intelligence is beginning to play a central role. Machine learning algorithms are being trained to analyze engineering designs, compare unit costs, detect schedule deviations, and even predict which projects are most likely to present irregularities. The promise is to transform auditing from reactive to predictive—and the first results are already emerging.

The challenge of overseeing sanitation projects

Basic sanitation projects are among the most fertile grounds for fraud and mismanagement in the public sector. The reasons are structural: long-term contracts, complex measurements, technical specifications that vary according to terrain, and a supply chain that makes price verification difficult.

A typical contract for building an STP involves dozens of items: excavation, foundations, concrete structures, pumping systems, pipelines, biological and chemical treatment equipment, as well as complementary works such as access roads and fencing. Each of these items has costs that vary by region, soil type, material transport distance, and seasonality. For a human auditor, verifying whether each value falls within a reasonable range is a herculean task.

In its 2025 audit, the TCU used a hybrid approach: technical teams analyzed contract samples while a computer system cross-referenced data from all 120 contracts. The system identified patterns such as unit prices above the national average, measurements exceeding what was actually executed, and time extensions without technical justification (TCU, operational audit report 2025). The result was a recommendation to suspend payments on 12 projects until the inconsistencies were clarified.

"Artificial intelligence does not replace the auditor, but empowers them to see what was previously invisible. In sanitation projects, where the amounts involved are in the billions and the social impact is immediate, this anticipatory capacity is decisive." — Excerpt from the TCU report on operational auditing of sanitation projects, 2025.

This case illustrates a paradigm shift. Instead of auditing everything with the same level of depth, the oversight body began using AI to prioritize where to concentrate efforts. Auditing becomes more efficient because the algorithm points out the highest-risk projects—and the human auditor focuses on investigating those cases in depth.

How AI analyzes engineering projects

The heart of the new approach lies in analyzing project data. A machine learning system can be trained on thousands of previous sanitation projects, including their final costs, to learn which combinations of items and prices are plausible.

The process works in stages. First, the algorithm extracts information from technical documents—budget spreadsheets, descriptive memoranda, physical-financial schedules. Next, it normalizes this data into a comparable baseline, considering factors such as geographic location and project scale. Finally, it applies statistical models to identify outliers: items whose unit cost deviates significantly from the expected distribution.

A practical example: the cost per cubic meter of reinforced concrete for treatment plant structures ranges between R$ 1,200 and R$ 1,800 in federal projects, depending on the region and structural complexity. If a contract shows a value of R$ 2,500, the system flags it automatically. The auditor then verifies whether there is a technical justification—such as the need for high-strength concrete—or whether there is evidence of overpricing.

The Office of the Comptroller General (CGU) has developed a platform that applies this logic at scale. In a test using data from 25 contracts from the Ministry of Cities, the system identified 89 items with above-expected prices, of which 72 were confirmed as overpricing after manual analysis (CGU, technical report 2025). The 81% accuracy rate demonstrates the tool's potential, although the number of contracts analyzed is still too small for generalizations.

The role of the Ministry of Cities and the CGU in the new oversight model

The Ministry of Cities, responsible for the national sanitation policy, is also adopting AI tools. In partnership with universities, the agency developed a schedule monitoring system that uses GPS data from machinery and satellite imagery to verify whether the physical progress of a project matches what is being paid.

The CGU has expanded its operations with anomaly detection algorithms for construction contracts. In 2025, the CGU reported savings of R$ 1.2 billion in sanitation contracts thanks to early identification of irregularities (CGU, annual report 2025). The amount includes cancellation of improper measurements, price renegotiation, and prevention of unjustified contract amendments.

The CGU's system cross-references data from different sources: company records, sanction history, input prices, and even meteorological information that could affect project progress. AI finds connections that would go unnoticed—such as a company that won contracts in different states with the same technical team, suggesting possible irregular intermediation.

The table above summarizes the current moment. The numbers show that AI is no longer a distant promise—it is an operational tool with measurable results.

Challenges and limitations of AI application

Despite the advances, applying AI to sanitation project auditing faces significant challenges. The first is data quality. Many older contracts are poorly digitized, with spreadsheets in incompatible formats and incomplete information. Without clean, standardized data, the algorithm loses accuracy.

The second challenge is institutional resistance. Auditors accustomed to traditional methods may see AI as a threat to their role. The TCU's experience shows that integration is more effective when the technology is presented as a support tool, not a replacement. The algorithm suggests; the auditor decides.

There is also the risk of algorithmic bias. If the model is trained on historical data that reflects regional overpricing practices, it may normalize those values as acceptable. To mitigate this risk, it is essential that models be audited periodically and that training data be reviewed by independent experts.

Conclusion: the future of predictive auditing

The application of AI to basic sanitation project auditing represents a significant evolution in how the Brazilian state controls its infrastructure investments. The initial results—with the identification of overpricing in 23% of contracts analyzed by the TCU and R$ 1.2 billion in savings by the CGU—demonstrate that the technology is already an operational reality, not merely a theoretical promise.

The path ahead involves overcoming challenges related to data quality, team training, and bias mitigation. But the direction is clear: predictive auditing, based on machine learning and large-scale data analysis, will become increasingly central to ensuring that public resources allocated to sanitation effectively reach the projects that will transform the lives of millions of Brazilians.

With sanitation universalization as a goal by 2033, AI is not just an efficiency tool—it is a strategic necessity for the country to achieve its objectives with transparency and fiscal responsibility. The coming years will be decisive in consolidating this transformation, and oversight agencies that adopt these technologies will be at the forefront of a new era of public governance.