ISG Provider Lens™ Generative AI Services - Development and Deployment Services - Large - Brazil 2025

Use cases show the evolution from experimentation to production of GenAI in Brazil

The second edition of the ISG Provider Lens® report Generative AI Services – Brazil 2025 reveals a significant transformation in the behavior of Brazilian companies regarding contracting GenAI services. After the initial experimentation phase of the previous year, characterized by the development of numerous PoCs, organizations are now demonstrating a more strategic and cautious approach, seeking solutions that can effectively bring significant productivity gains with their GenAI assistants and agents operating in production, generating measurable ROIs.

Most of the leading suppliers in this study presented success stories with tangible benefits for their customers, including gains in process automation and operational process transformation, which generated increased business and improved the level of services offered to end customers.

During the interviews and presentations for this study, cases were also reported in which PoCs did not progress to the production phase. The reasons for abandoning these projects are diverse and include:

● Inappropriate choice of technology that could be solved with systems that do not use GenAI;

● High potential for model hallucinations;

● Organizational cultural resistance;

● Low understanding of the technology by end users;

● Unpredictable costs of operating the solutions.

Project evaluation methodologies

For this reason, most suppliers have developed structured methodologies for evaluating the ROI and TCO (Total Cost of Ownership) of their projects. Although in some cases the positive return is quite clear, the prediction of GenAI operating costs remains uncertain, given that token-based billing is challenging to predict when the solution is deployed in production and used at scale.

To overcome this limitation, some leading vendors have developed more predictable billing models, such as per-interaction billing, offering greater transparency and cost control to customers.

Consolidated applications using GenAI:

Two areas that GenAI has profoundly transformed are IT itself and customer service chatbots. The first is represented by SDLC (Software Development Life Cycle) solutions and the modernization of legacy applications. The vast majority of companies have solutions for this application, promoting accelerated development and, consequently, reducing the allocation of developers to projects, allowing them to offer more competitive services to customers.

This phenomenon has two critical implications:

1. Familiarity and effectiveness: Internal use of these applications allows system integrators to become more familiar with the solution and thus develop more effective use cases for their customers.

2. Competitive advantage: As system construction requires fewer developer hours, companies can provide more competitively priced services. Thus, suppliers using SDLC will have a competitive advantage over competitors who still employ many human developer hours.

Although there is still no clear perception of the impact of using SDLC solutions on service pricing, there is an undeniable perception that system integrators need to internalize these solutions to remain competitive in the market.

The second application offered by the vast majority of suppliers, which has preconfigured solutions, is chatbots for customer service or internal use, such as HR chatbots to assist human resources departments in communicating with the organization’s employees.

These two trends were already quite strong in last year’s study and have now been consolidated with offerings from the hyperscalers themselves, such as Amazon Q Developer and Amazon Q Business. Many of these chatbots have evolved into more sophisticated versions, including features such as summaries of technical manuals for industrial facility teams.

Emerging applications of GenAI:

This year’s study highlighted the trend of combining GenAI with Data Visualization (DataViz) and ML (traditional AI), which has been dubbed GenBI. Many vendors presented solutions with names such as talk to your data or browse your data.

This type of solution uses GenAI to create structured data dashboards through natural language, with GenAI automatically generating queries on structured data in SQL or Python in the background.

Another significant field of application was production use cases for courts of law, where agents assist courts in identifying and classifying documents, orders and judgments, speeding up legal proceedings. Many of these cases utilize MultiLLM architectures to avoid single-vendor lock-in and optimize queries, leveraging the LLMs that perform best in each specific case.

GenAI is establishing itself as a powerful technology for building more user-friendly UI and also as a fundamental tool for knowledge management in organizations.

Agent orchestration and Multiagent architectures

This year’s study also confirms the trend toward using multiple GenAI agents orchestrated by architectures that utilize orchestration frameworks such as LangChain and CrewAI. Although cases of autonomous agents orchestrated in production are still relatively rare, the trend is confirmed and leading vendors are building extensive libraries of preconfigured agents to accelerate the construction of solutions for their customers.

As these libraries become available to most vendors, they differentiate themselves through their specialized frameworks for correcting biases, eliminating hallucinations and avoiding inappropriate content.

Paradigm shift in productivity

The use of autonomous agents has proven to be a business lever different from what was initially expected. Many companies believed that they would see productivity gains by directly replacing human labor with digital agents. Success stories have shown that productivity gains come primarily from human workers empowered by AI, rather than from their complete replacement.

Reliable and ethical AI as a strategic priority

The issue of reliable and ethical use of AI has been treated as a priority by suppliers. This concern is not limited to compliance with growing industry regulation (Brazil is in the process of approving an AI regulatory framework), but reflects a strategic understanding that reliability is critical to largescale business adoption.

The solutions developed include:

● Automatic bias detection;

● Explainability with tools such as SHAP and LIME;

● Implementation of human-in-the-loop for critical decisions;

● Complete auditability of decision-making processes.

In this context, the use of Graph retrievalaugmented generations (RAGs) has emerged, that is, the use of graphs or knowledge graphs to organize and hierarchize information in database vectorization processes or model learning with customer data. Partnerships with Neo4j, one of the leading graph solutions on the market, have been cited more frequently by suppliers.

Set of partnerships with new logos

Regarding partnerships, most suppliers in the Large Providers segment maintain partnerships with the three hyperscalers (AWS, Microsoft and Google) and have added NVIDIA to their portfolio, which provides mature development tools and superior performance in model configuration.

Midsize Providers generally do not have the resources to maintain such a broad spectrum of partnerships and therefore establish partnerships with one or two hyperscalers. However, this does not limit their ability to deliver effective solutions, as most models and platforms have APIs and can work with workloads on any hyperscaler, especially with the development of protocols such as Anthropic’s Model Context Protocol (MCP).

These protocols are open standards that standardize how AI applications connect to external data sources, essentially functioning as a communication protocol that allows LLMs such as Claude to access contextual information in a secure and structured manner.

The focus of midmarket vendor partnerships is much more important from a commercial standpoint than a technical one, as hyperscalers provide qualified commercial leads for these vendors.

Prospects for Small Language Models

The use of SLMs (Small Language Models) in the architecture of GenAI solutions is still limited. Most of the problems addressed by vendors are solved using LLMs, which are still more cost-effective than the SLMs available on the market.

The launch of IBM’s Granite may change this scenario, but the availability of pretrained models is still limited and the costs of training SLMs themselves are high for most use cases. The most notable cases include highly specific chatbots, such as the one developed by the University of São Paulo, AWS and Hospital das Clínicas de São Paulo (HCFMUSP), based on GenAI to optimize the screening of healthrelated legal proceedings. The complexity of the subject matter justified the choice of a specialized SLM.

Impact on data governance and traditional AI

Many vendors reported that the interest generated by GenAI ultimately drove ML and traditional AI projects, in which fundamental problems, such as data governance, were addressed during the design of GenAI applications.

Customers became more aware of the need for rigorous governance processes and began to explore the use of statistical models with structured data to solve their business problems. The initial interest in GenAI ultimately led to a greater potential for organizations to become data-driven and adopt traditional data and analytics practices.

In this new phase of the Brazilian GenAI market, suppliers are assuming an even more strategic advisory role to responsibly and effectively explore the potential of the technology. Suppliers need to prepare to overcome challenges related to solution reliability, cost unpredictability and cultural resistance among customers.

The transition from the experimental phase to production implementation marks a moment of market maturity, where delivering real and measurable value becomes the primary competitive differentiator. Companies that manage to balance technological innovation with robust governance and sustainable business models will be better positioned to lead this new phase of the GenAI revolution in Brazil.

Access to the full report requires a subscription to ISG Research. Please contact us for subscription inquiries.