Generative AI in Finance: Real Use Cases in Brazilian Banks and Fintechs

The Brazilian financial market is undergoing a silent but profound transformation. While most executives are still debating whether or not to adopt generative AI, some banks and fintechs are already reaping concrete results — reduced operational costs, improved customer experience and more precise credit decisions. I have worked directly with institutions such as BTG, B3, Safra, XP and Inter, and what I see in the field is very different from what you read in press releases. In this article, I will present real use cases of generative AI in the financial market, covering what works, what is still a promise and where the greatest opportunities lie for those making decisions today.

Why the Financial Sector Is the Ideal Environment for LLMs

Before diving into use cases, it is important to understand why the financial sector is structurally favorable for LLMs (Large Language Models). Banks and fintechs deal with three assets that feed AI models: massive volumes of structured data, rich transactional history and continuous demand for natural language processing — whether in contracts, regulatory documents, customer service or risk analysis.

Brazil has an even more favorable context. With over 150 million digital accounts opened following PIX and Open Finance, the volume of data available for training and inference has grown exponentially. At the same time, regulatory pressure from the Central Bank — with initiatives such as Drex and Open Finance — is forcing institutions to modernize their data architectures, creating a more solid technical foundation for AI implementations.

It is no coincidence that the financial sector is, globally, one of the largest investors in generative AI. According to McKinsey, AI use cases in the banking sector have the potential to generate between US$ 200 billion and US$ 340 billion in annual value. In Brazil, this figure is proportionally significant given the size and sophistication of our financial system.

Customer Service: Beyond the Basic Chatbot

The first and most visible use case of artificial intelligence in Brazilian banks is customer service. But there is a fundamental difference between a rules-based chatbot that responds "I didn't understand your question" and an LLM-based assistant that actually solves problems.

Itaú Unibanco, for example, evolved its virtual assistant to incorporate language models capable of interpreting ambiguous intentions, cross-referencing information from the customer's history and resolving complex requests without escalating to a human agent. The result reported by the institution was a reduction of approximately 30% in the volume of calls transferred to human agents in certain customer journeys.

In the fintech segment, Nubank has been using generative AI to personalize communication at scale — not only in reactive customer service, but in generating contextualized explanations for charges, credit limits and invoice variations. The difference is subtle but powerful: instead of a generic message, the customer receives an explanation built around their specific behavior.

The critical point here is not technology — it is data architecture. Institutions that still operate with information silos between CRM, core banking systems and digital channels will struggle to deliver a coherent AI experience. The quality of the LLM's output is directly proportional to the quality and integration of the data that feeds it.

Credit Analysis and Risk Management with Generative AI

This is, in my assessment, the use case with the greatest measurable financial return in the short term. Banking automation with AI applied to credit analysis is not new — predictive models have been used for decades. What changes with generative AI is the ability to process unstructured data within the risk equation.

Invoice documents, financial statements in PDF format, supply contracts, free-text registration assessments — all of this was excluded from traditional models because it could not be easily structured. With LLMs, it is possible to extract risk signals from these documents and incorporate them into the credit score in an automated way.

BTG Pactual has been exploring this frontier especially in the corporate credit segment, where the analysis of complex documents is intensive in both time and human expertise. Accelerating the analysis cycle has a direct impact on commercial competitiveness — institutions that respond faster capture more business.

In the SME credit market, fintechs such as Creditas and Jungle Finance have been using generative AI to interpret bank statements and Open Finance data in a more sophisticated way, identifying cash flow patterns that traditional models failed to capture. This makes it possible to approve credit for companies that would be rejected by conventional methods — while keeping default rates under control.

Generative AI does not replace the senior credit analyst. It transforms a junior analyst into someone with the analytical capacity of a senior analyst, multiplying the team's productivity without compromising the quality of the decision.

Compliance, Regulatory Affairs and Anti-Money Laundering

If there is one area where LLMs in the financial sector solve a genuinely costly and complex problem, it is compliance. Brazilian banks spend billions of reais annually to meet the requirements of the Central Bank, the CVM, SUSEP and COAF. A large part of that cost is repetitive human work: reading regulatory documents, adapting processes, monitoring transactions and producing regulatory reports.

Intelligent automation with generative AI is tackling this pain point on multiple fronts:

• Suspicious transaction monitoring: LLMs can analyze transaction patterns and generate explanatory narratives about why a given operation may be suspicious, accelerating the work of AML (Anti-Money Laundering) teams and reducing false positives.
• Reading and interpreting regulatory documents: A Central Bank circular published today can be summarized, interpreted and have its impact mapped onto internal processes within hours, not weeks.
• Generating regulatory reports: Reports for COAF, the Central Bank and other regulators can have drafts generated automatically, with the analyst focusing on review rather than document production.
• KYC (Know Your Customer): The analysis of onboarding documents, including the verification of inconsistencies in statements and documents, can be automated with significantly greater accuracy than manual processes.

Bradesco and Safra, with whom I had close contact on modernization projects, face the challenge of maintaining large teams dedicated to compliance. Generative AI does not eliminate these teams, but it radically redefines what each professional does — shifting operational work toward higher-value analytical work.

Financial Content Generation and Portfolio Personalization

XP Investimentos and BTG Pactual are benchmarks for how the Brazilian financial sector uses generative AI to democratize access to sophisticated financial analysis. The traditional challenge was straightforward: qualified analysts are scarce and expensive. With LLMs, it is possible to scale the production of analytical content without linearly scaling human costs.

Some concrete cases I observe in the market:

• Asset analysis reports: Automated generation of first drafts of analyses on equities, real estate investment funds and fixed income, with the human analyst adding qualitative judgment and validating conclusions.
• Fund manager letters: Investment funds using AI to structure periodic communications with shareholders, personalized according to each investor's profile and portfolio.
• Personalized recommendations: Platforms that use LLMs to explain, in accessible language, why a particular product makes sense for a specific investor's profile — going beyond basic regulatory suitability.
• Smart market alerts: Contextualized notifications that explain the impact of a macroeconomic event on the client's specific portfolio, rather than just delivering the generic news.

Livelo and MaxMilhas, with whom I have also worked, apply similar logic in the context of loyalty programs and travel — using generative AI to personalize offers and communications at scale. The lesson is transferable: wherever there is a large customer base with distinct profiles and a need for personalized communication, LLMs deliver fast ROI.

The Real Challenges Nobody Mentions in Press Releases

It would be dishonest of me to present only the successes. In the implementations I have accompanied and led, the obstacles are real and need to be on the radar of anyone evaluating this path.

The first challenge is data quality. There is no effective generative AI built on bad data. Banks with decades of operation frequently have data fragmented across legacy systems, with issues of inconsistency, duplication and lack of context. Before any LLM project, an honest exercise in data diagnosis and quality assurance is necessary — and that work tends to reveal problems that IT departments already knew existed, but never had sufficient urgency to address.

The second challenge is governance and explainability. The Central Bank has been signaling growing attention to the use of AI in decisions that affect consumers — especially in credit and insurance. Decisions generated by LLMs need to be auditable and explainable, which requires more carefully designed architectures than simply connecting an LLM API to the core banking system.

The third challenge is organizational culture. Across more than 300 transformation projects I have led throughout my career, I can state with conviction: technology is rarely the limiting factor. What stalls AI implementations in banks is resistance from departments that perceive automation as a threat, a lack of clarity around project ownership and the absence of executive leadership willing to sponsor changes that go beyond the pilot stage.

The greatest risk is not adopting generative AI too quickly. It is running endless pilots, without scale, without impact, and concluding that "the technology didn't work" when the problem was never the technology.

How Decision-Makers Should Think About Generative AI Today

If you are a CEO, CTO or founder of a financial institution, the question is no longer "should I invest in generative AI?" The question is "where and how should I invest to achieve real returns over the next 12 to 18 months?"

My practical recommendation, based on what works in the field:

• Start with internal use cases. Assistants for compliance, credit and analysis teams carry lower regulatory risk and deliver measurable productivity gains quickly. The organizational learning that comes from these projects is invaluable for customer-facing initiatives.
• Invest in data architecture before LLMs. If your data is not integrated, accessible and of guaranteed minimum quality, any AI project will underutilize the technology's potential. Data is the infrastructure of AI.
• Define success metrics before you start. Reduction in analysis time, decrease in escalation rates in customer service, increase in credit approval rates with controlled default — choose metrics that the business cares about and measure from day one.
• Build or hire internal AI capability. Relying exclusively on vendors for strategic AI projects is a risk. The institutions that advance the most are those that combine external partners with internal teams who deeply understand the business.

The Brazilian financial market has the opportunity to be among the most advanced in the world in adopting generative AI — not out of trend-following, but because the context is favorable: a robust data foundation, intense competitive pressure between traditional banks and fintechs, and a regulatory environment that, while rigorous, is not hostile to innovation when navigated well.

The window of competitive differentiation that exists today is real, but it is not permanent. Institutions that structure their AI strategy well over the next 18 months will create advantages that are difficult to replicate. Those that wait for the technology to "mature further" may find that maturity has arrived — but for their competitors.

If you want to understand where your institution stands on this journey and which next steps will be most profitable, get in touch. This is exactly the kind of strategic conversation I have with CEOs, CTOs and founders in the financial sector — no product pitch, no hidden agenda, focused entirely on what genuinely drives results for your business.