Read the Report — State of Applied AI →
EnergyFinance

How Petrobras Uses Generative AI to Uncover $120M in Tax Savings

Petrobras applied generative AI and intelligent automation to its tax department, feeding 150 pages of Brazilian tax regulations and three months of financial data into a model built on Automation Anywhere, Amazon Bedrock, and Amazon SageMaker. In three weeks, the system identified $120 million in tax savings and filed taxes within three days—the first time in 15 years the team avoided weekend work during tax season. The company projects over $1 billion in total savings by year-end.

Impact

$120M

Tax savings identified

$1B+

Projected total savings

40%

Efficiency increase

3 days

Tax filing time

Challenge

Petrobras's tax department manually processed filings across 150 pages of complex Brazilian tax regulations, causing persistent inefficiencies—including 15 straight years of weekend filing work—and leaving significant savings opportunities unidentified within its $54 billion annual tax obligation.

Solution

Automation Anywhere, Amazon Bedrock, and Amazon SageMaker were combined to automate tax calculation, filing, and compliance: Bedrock generated reports and compliance documentation in Portuguese while SageMaker forecasted liabilities and detected anomalies in real time.

Tools & Technologies

What Leaders Say

We saved $120 million in just three weeks with Automation Anywhere's generative AI-powered solutions. This is just the beginning.

, CIO

We filed our taxes in three days. For the first time in 15 years, we didn't have to work over the weekend to file taxes.

, General Manager of Taxes & Government Participations
Get the full context.

Sign up to read complete case studies, access detailed metrics, and unlock all use cases.

Full Story

Petrobras is Brazil's state-controlled energy giant, operating for over 70 years in ultra-deepwater oil and gas exploration with more than 45,000 employees. With an annual tax obligation exceeding $54 billion, the company's tax department managed some of the most complex financial compliance requirements in South America, navigating 150 pages of intricate Brazilian tax regulations across multiple tax types and jurisdictions.

Before deploying AI, the tax team processed compliance workflows manually, parsing dense regulatory documents and historical transaction data to determine the taxable base, calculate exemptions by region and sector, and compute ICMS rates per Brazilian state. The process was slow, error-prone, and labor-intensive enough that filing taxes routinely required staff to work through weekends—a pattern that had persisted for 15 consecutive years.

Petrobras's new CIO, an active AI user, saw the tax department as an ideal pilot for intelligent automation. The team inputted three months of tax data and 150 pages of regulations into a generative AI stack anchored by Automation Anywhere, Amazon Bedrock, and Amazon SageMaker. Bedrock foundation models, fine-tuned on historical data and regulatory guidelines, generated precise tax reports, audit summaries, and compliance documentation in Portuguese. SageMaker powered predictive models that forecasted tax liabilities and detected anomalies in real time using historical financial data and external factors including oil price trends.

Three weeks after deployment, the system had uncovered $120 million in tax savings—a result that far exceeded expectations. For the first time in 15 years, the department filed taxes within three days and did so without requiring weekend work. The automation also delivered a 40% increase in overall efficiency, with Petrobras projecting more than $1 billion in total savings by year-end.

The success in the tax department validated Petrobras's broader AI and automation strategy. The company is expanding intelligent automation into HR, finance, and procurement—aiming to replicate the gains seen in tax across every corporate function. With the operations department already fully embracing automation and generative AI, Petrobras sees the tax implementation as the turning point that brought the entire enterprise into the AI era.

Similar Cases

A
Alpiq
25–30%
reduction in mean time to detect issues

Alpiq, a Swiss energy services provider with 1,200 employees, replaced five separate monitoring solutions with Datadog as its company-wide observability platform. After deploying the Datadog Mule® Integration to bring MuleSoft monitoring into Datadog, the team achieved full end-to-end visibility and reduced mean time to detect issues by 25–30%.

EnergyDDatadog
P
Pinterest
17% YoY
revenue growth

Pinterest built an AI-powered discovery engine on AWS processing 18TB daily, delivering 10 million AI recommendations per second across 10,000+ GPU instances, driving 17% revenue growth and 70% AI-driven discovery.

TechnologyAEAmazon EKSASAmazon SageMaker
A
AES
96%
audit result generation speed improvement

AES, a global energy company transitioning to renewable energy, deployed Claude on Google Cloud's Vertex AI to automate its safety audit process across thousands of assets. The multi-agent framework made audit result generation 96% faster and cut audit costs by 99% while improving accuracy by 10–20%. This transformation enabled AES to scale safety operations alongside its exponentially growing renewable energy portfolio.

EnergyGVGoogle Vertex AICClaude
O
Omnicom
90%
compute infrastructure cost reduction

Omnicom is one of the world’s largest marketing communications networks, with 75,000 employees serving over 5,000 clients across 70+ countries. The company migrated nine global data centers to AWS and built an AI-powered platform on Amazon Bedrock and Amazon SageMaker to deliver hyper-personalized campaigns at scale. The migration cut compute infrastructure costs by 90% while enabling real-time processing of 400 billion daily marketing events.

Media & EntertainmentASAmazon SageMakerABAmazon Bedrock
RS
Reno Salvage
1 afternoon
inventory system build time

Reno Salvage, an 86-year-old Nevada salvage yard, uses ChatGPT to troubleshoot equipment failures in real time, build a 1,000-item inventory system in a single afternoon, and answer complex technical customer questions on the spot — keeping operations running without costly downtime.

EnergyCChatGPT
R
Repsol
96%
harvey adoption rate

Repsol is a global multi-energy company operating in over 20 countries with a legal division of more than 200 professionals across 12 jurisdictions. After deploying Harvey across its entire legal department in 2024, Repsol achieved 96% adoption, reduced contract clause review time by more than 90%, and each lawyer now saves 4–6 hours per week—time that flows directly into strategic work and professional development.

EnergyHVHarvey VaultHHarvey
BO
Blue Origin
2,700+
ai agents deployed

Blue Origin deployed 2,700+ AI agents with 70% company-wide adoption, achieving a 90% reduction in hardware development time using Amazon Bedrock.

Aerospace & DefenseManufacturingALAWS LambdaAEAmazon EKS
A
AppFolio
80–90%
latency reduction

AppFolio is a real estate management platform serving 20,000+ customers and 8 million+ units under management. After building Realm-X Messages—an LLM-powered inbox for property managers—on Amazon Bedrock, AppFolio used Datadog LLM Observability to identify bottlenecks and cut latency by 80–90%, which drove a 300% increase in product adoption and saves property managers an average of five hours per week.

Real EstateABAmazon BedrockDLDatadog LLM Observability