Petrobras identifies $120M in tax savings and cuts filing to 3 days

Petrobras's tax department manually processed filings against 150 pages of complex Brazilian tax regulations across multiple tax types and jurisdictions, within a substantial annual tax obligation. The process was slow and error-prone—requiring staff to work weekends for 15 consecutive years—and left significant savings opportunities unidentified.

Petrobras deployed a generative AI and intelligent automation stack combining Automation Anywhere, Amazon Bedrock, and Amazon SageMaker. The team fed three months of financial/tax data and 150 pages of regulations into the system: Bedrock generated tax reports, audit summaries, and compliance documentation in Portuguese, while SageMaker forecasted liabilities and detected anomalies in real time.

Within three weeks, the system identified $120 million in tax savings. For the first time in 15 years, the department filed taxes within three days without weekend work, and overall efficiency increased 40%. Petrobras projects substantial total savings by year-end and is expanding the approach to HR, finance, and procurement.

• Tax compliance with dense, multi-jurisdiction regulations is a high-ROI pilot for generative AI plus RPA when historical filings and regulatory text can be loaded into the model.
• Pairing foundation models (report/document generation) with predictive ML (liability forecasting and anomaly detection) produced both savings discovery and faster filing cycles.
• A visible early win in tax helped validate enterprise-wide automation and generative AI adoption beyond the pilot function.