OCTO: ML-Powered Fraud and Cost Anomaly Detection

The goal of the project was to move TEMPO's fraud and cost control from manual, reactive reviews to a predictive, data-driven model. OCTO was designed to use Machine Learning to detect anomalies, cost distortions, and potential fraud across assistance service requests, and to direct analysts to the cases that matter most.

The project aimed to reduce financial losses, cut manual review effort, speed up investigations, and give the company the ability to scale its operations without losing control over costs and margins.

2026

OCTO delivered measurable financial and operational impact:

• Approximately BRL 3 million in annual cost avoidance, thanks to earlier and more accurate detection of anomalies, cost distortions, and potential fraud.
• Around 25% reduction in fraud-analysis headcount, as ML-based prioritization replaced broad manual reviews.
• Faster, more focused investigations: analysts now work on a prioritized list of suspicious cases instead of reviewing requests at random.
• Stronger cost control and scalability: the company can grow its volume of service requests while keeping controls consistent.

Overall, TEMPO shifted to a proactive approach to fraud prevention, with better decisions, protected margins, and a solid foundation for further use of AI in operations.

OCTO was developed using a combination of Machine Learning models, cloud infrastructure, data processing pipelines, APIs, and analytics tools integrated with TEMPO’s operational ecosystem.

The solution processes historical and transactional data from assistance systems and is fully integrated with TEMPO's existing platforms, with security and traceability built in.

The project does not use solutions from the Global CIO catalog

Fraud and cost anomalies have no single pattern, so the team had to consolidate historical and operational data, select relevant variables, and train and validate models on real cases. A key challenge was balancing sensitivity with usability: catching meaningful anomalies without overwhelming analysts with false positives. The solution also had to be integrated into existing systems and investigation processes, which required close collaboration between Technology, Data, Operations, and Fraud teams to turn model outputs into actionable work.

Before the project, fraud and cost control at TEMPO relied mainly on manual, reactive analysis. Broad reviews of service requests were time-consuming, required a large team, and often detected problems too late.

To address this, TEMPO built OCTO, a Machine Learning solution that analyzes large volumes of operational and financial data, compares it against historical patterns, and flags cases that deviate from expected behavior. These cases are prioritized and sent to specialized teams for investigation.

The implementation required data consolidation, model training and validation, and careful tuning to keep false positives low. The solution was built on a secure, scalable cloud architecture with full traceability and integration into TEMPO's existing operational platforms. Close cooperation between technology and business teams was essential to make the model's output part of daily work.

As a result, analysts now focus on the most relevant cases, investigations are faster and more consistent, and the company avoids around BRL 3 million in costs annually while reducing fraud-analysis headcount by about 25%.

OCTO is deployed across TEMPO’s operations in Brazil, supporting the analysis of assistance service requests from different regions of the country. Because the solution is digital and cloud-based, its Machine Learning models can analyze operational and financial patterns at scale, regardless of where the service was delivered.

This nationwide scope is especially relevant for TEMPO’s business, as fraud patterns, service costs, provider behavior, and operational anomalies may vary significantly by region. OCTO enables a centralized intelligence layer while preserving visibility into regional differences.