Petrobras

Petrobras, South America's largest energy company with more than 45,000 employees and 70 years of ultra-deepwater oil and gas expertise, faced a sharp internal contradiction: the company was an acknowledged leader in applying AI to exploration and production, yet its corporate administrative functions — tax, procurement, HR, and finance — remained largely manual. The tax department bore the heaviest burden. Staff were required to interpret hundreds of pages of dense, frequently updated Brazilian tax code while processing months of company financial data, with errors carrying material regulatory and financial risk. The workload was severe enough that employees routinely worked weekends throughout tax season, a practice that had persisted unbroken for 15 years.

Petrobras deployed Automation Anywhere's Automation Co-Pilot for Business Users, integrating a generative AI model directly into the tax department's filing workflow. Hundreds of pages of complex Brazilian tax code were ingested alongside months of Petrobras's own tax data, giving the model the regulatory context needed to identify discrepancies, missed credits, and filing inaccuracies. Rather than a phased multi-year rollout, the implementation was a focused departmental pilot — tax was chosen as the proving ground because it combined high rule complexity with quantifiable financial stakes. The generative AI layer automated end-to-end tax filing, replacing manual document review and cross-referencing with a system capable of applying regulatory rules at scale and surfacing actionable findings in days rather than weeks.

The tax department delivered measurable outcomes within three weeks of deployment:

• $120M in tax savings identified in the first three weeks alone
• Tax filing cycle reduced to 3 days, down from a process that previously consumed entire weekends each filing season
• First weekend-free tax season in 15 years — a direct quality-of-life improvement for the department
• $1B+ in annual savings projected as Petrobras expands the platform to HR, procurement, and finance

CIO Carlos Barreto credited the speed of results — tens of millions of dollars saved in weeks — as validation for accelerating enterprise-wide AI adoption. The savings are also earmarked in part to fund Petrobras's transition to renewable energy, including wind and solar.

• Target rule-dense, high-stakes processes first. Brazilian tax code complexity made it an ideal AI candidate — high volume of structured rules, clear right/wrong outputs, and direct financial consequences for errors.
• Feeding proprietary data alongside regulatory text is the differentiator. The model's accuracy came from combining external tax code with internal company financials, not from general-purpose knowledge alone.
• Prove ROI fast with a departmental pilot before scaling. A single-department win ($120M in three weeks) created the mandate and budget to expand across HR, procurement, and finance.
• Quantify employee burden as part of the business case. Eliminating 15 years of weekend work is a retention and morale argument that complements the financial case.