AI in Pharma Market Access: Where It Works, Where It Doesn't, and Why Brazil Is the Most Urgent Case

Brazil's pharmaceutical market is the largest and the most methodologically demanding in Latin America. CONITEC is maturing its criteria at an accelerated pace, the new real-world data infrastructure is already operational, and the region's first MEA for gene therapy has just been signed. This guide explains where AI already delivers results in market access — and why Brazil concentrates the bulk of the opportunity.

83% of life sciences professionals consider AI to be overhyped. Only 9% of biopharma leaders report having seen real return on their AI investment. And yet, in one specific pharmaceutical process — market access — results are already measurable, verifiable, and reproducible. In no other market in the region is that contradiction more relevant than in Brazil.

This article is for market access directors, HEOR managers, and medical affairs teams operating in Brazil and Latin America. It is not a trends presentation — it is an analysis of what can be done today, with what documented results, and under what methodological conditions.

1. Skepticism about AI in pharma: justified in almost every case

The widespread skepticism toward AI in the pharmaceutical industry has an empirical basis, not an emotional one. The failures are real and documented in the three areas where the most investment has been made:

• Drug discovery: decade-long timelines, near-infinite variables, binary success. AI can accelerate specific steps, but ROI does not materialize within the current investment cycle.
• Clinical trials: study design depends on regulatory criteria, heterogeneous populations, and endpoints that evolve with the scientific context. AI improves recruitment but does not transform the core process.
• Manufacturing and supply chain: the benefits are real but gradual, and they require data infrastructure that most companies have not yet built.

McKinsey puts it precisely: simply adding AI to existing processes does not generate value. The problem is not the technology — it is that AI has been applied to processes that lack the structural profile AI needs to be effective.

2. Why market access is the exception: the process profile AI needs

Pharmaceutical market access brings together the four conditions that make generative AI effective:

Pharmaceutical function	ROI timeline	Process	Suitable for generative AI?
Drug discovery	10+ years	Open-ended, infinite variables	No
Clinical trials	5–10 years	Regulated but unpredictable	No
Manufacturing / supply chain	2–5 years	Optimizable but slow	Partial
Market access / HEOR	6–18 months	Structured, evidence-based, repeatable	YES

The construction of value dossiers, cost-effectiveness models, systematic literature reviews, and budget impact analyses shares a profile that language models process better than any other type of task: synthesizing large volumes of structured information, adapting it to specific methodological criteria, and producing verifiable outputs within defined timelines.

3. Real benchmarks: documented results in market access with AI

Results are already available and come from verifiable organizations — not press releases:

Market access process	Without AI	With AI	Benchmark
Systematic review (screening)	3–6 weeks	2–4 days	~80% reduction in screening time
Cost-effectiveness model ICER (local adaptation)	4–6 weeks	3–5 days	Margin of error below 1% vs. original model
Complete Global Value Dossier	4–8 months	6–8 weeks	60% time reduction documented
Local adaptation per market	2–4 weeks/country	2–5 days/country	Scalable in parallel across multiple markets

4. The specific context of Brazil: why the window is open — and closing

In the rest of Latin America, the transformation of the HTA ecosystem is gradual. In Brazil, it is happening simultaneously on four fronts:

4.1 CONITEC is raising the methodological standard in real time

The Comissão Nacional de Incorporação de Tecnologias no SUS (CONITEC) has consolidated in recent years one of the most rigorous HTA evaluation processes in Latin America. It operates with a reference threshold of 3x GDP per capita, requires local real-world evidence for high-cost indications, and mandates budget impact analysis with Brazilian health system parameters.

The most recent milestone: in March 2025, CONITEC signed the region's first managed entry agreement (MEA) for a gene therapy — Novartis's Zolgensma. That precedent signals an unequivocal direction: greater methodological sophistication, higher demand for local evidence, and lower tolerance for dossiers built without adaptation to the Brazilian context.

4.2 The RNDS opens an unprecedented real-world data source in the region

In July 2025, Brazil launched the National Health Data Network (RNDS), a real-world data infrastructure connecting hospitalization records, pharmaceutical dispensation, outpatient procedures, and mortality data at the national scale. For market access teams, this means that for the first time it is possible to build local comparative effectiveness evidence with data from the actual Brazilian population — not extrapolations from European or North American studies.

Companies that develop AI capabilities to process and synthesize RNDS data now will be positioned to present CONITEC with evidence their competitors will not be able to replicate retroactively.

4.3 The AI Bill and the National Plan create the regulatory framework

Bill PL 2338/2023 — Brazil's Legal Framework for Artificial Intelligence — is advancing in Congress with broad support. The National AI Plan allocates approximately one-third of R$23 billion to the health sector. That budget is not just going to hospitals — it is building the data infrastructure, quality standards, and validation protocols that will define how AI is used in market access decisions in Brazil over the next five years.

4.4 The regulatory gap on AI is a window that will close

CONITEC does not yet have specific guidelines on AI-generated evidence. That does not mean it will accept anything — it means it applies its existing methodological criteria. Companies that build AI practices aligned with recognized international frameworks today will have a credibility advantage before CONITEC that will be very difficult to replicate once formal requirements exist.

5. Market access functions with the greatest AI impact in the Brazilian context

Systematic reviews for local clinical evidence

CONITEC requires local effectiveness evidence for high-cost medications. Systematic reviews that include studies with Brazilian populations require searches in local databases (Lilacs, SciELO, BVS) in addition to international ones. AI can manage that volume of cross-database search with greater speed and consistency than the manual process.

Economic modeling with SUS parameters

Adapting a cost-effectiveness model to the Brazilian context requires calibrating specific parameters: SIGTAP unit costs (Sistema de Gerenciamento da Tabela de Procedimentos do SUS), local disease epidemiology, comparators available in the national therapeutic formulary, and the reference threshold of 3x GDP per capita. AI can perform that adaptation in two to five days with less than 1% error relative to the base model.

Value narrative calibrated for CONITEC vs. private health plans

Brazil's health system is dual: the public SUS and a robust private sector with more than 47 million beneficiaries. CONITEC's decision criteria — budget efficiency, SUS impact, comparators from the national formulary — are radically different from those of a medical director at a private health plan operator. AI enables generating differentiated versions of the same value argument for each audience without multiplying working time.

6. Methodological rigor: what CONITEC expects and how AI meets it

CONITEC does not have specific guidelines on AI, but its methodological quality criteria are explicit and rigorous. The relevant international frameworks in 2026 set the standard:

• ELEVATE-GenAI (ISPOR Working Group, 2025): standardized reporting framework for the use of LLMs in HEOR. Establishes transparency, traceability, and validation criteria that CONITEC can verify.
• CHEERS-AI (ISPOR, 2025): extension of the CHEERS checklist for economic evaluations incorporating AI. Specifies how to report AI use in cost-effectiveness models.
• NICE precedent (August 2025): NICE launched HTA Innovation Laboratory projects exploring AI in economic modeling and HTA automation. Its position is clear: AI use must be declared, transparent, and reproducible.

Frequently asked questions

Why does market access have a different profile from drug discovery or clinical trials for AI application?

Because market access has exactly the process profile that generative AI needs to create value: structured, based on verifiable evidence, repeatable with known variations, and with measurable short-term results. Drug discovery and clinical trials are open-ended processes with decade-long timelines and uncontrollable variables — the opposite of AI's natural territory.

How does the CONITEC submission process differ from other LATAM HTA agencies?

CONITEC is the most formalized and rigorous process in the region. Unlike IETS in Colombia or CENETEC in Mexico, CONITEC has an explicit reference threshold (3x GDP per capita), requires budget impact analysis with SUS parameters, and publishes regularly updated methodological guidelines. This means local dossier adaptation is not optional — it is determinative for approval.

What changed with the launch of the RNDS for market access teams in Brazil?

The National Health Data Network launched in July 2025 creates for the first time the possibility of generating comparative effectiveness evidence with real data from the Brazilian population — hospitalization, dispensation, procedures, mortality. This changes the nature of the argument before CONITEC: from extrapolating evidence from international studies to demonstrating effectiveness in the population the SUS actually serves.

What did the Zolgensma MEA with CONITEC in March 2025 mean?

The first managed entry agreement for gene therapy in Brazil establishes that CONITEC is willing to approve very high-cost technologies under conditions of uncertainty — provided there is a mechanism for outcome monitoring and evidence-based price adjustment. For companies with high-cost therapies in the pipeline, this precedent means that access to the Brazilian market no longer depends exclusively on clinical trial evidence — it depends on the ability to generate and monitor local evidence on an ongoing basis.

Are there specific CONITEC or ANVISA guidelines on AI-generated evidence?

Not yet. Neither CONITEC nor ANVISA has published specific guidelines on the use of AI in health technology assessment processes. What does exist is the international precedent from NICE (UK), and ISPOR's methodological frameworks (ELEVATE-GenAI, CHEERS-AI) that establish how to document and validate AI use in a way agencies can evaluate. Aligning with those standards today is the recommended practice before formal requirements exist.

Conclusion: Brazil is not the easiest LATAM market — it is the most important

Brazil's pharmaceutical market is the most demanding in Latin America in terms of HTA methodological rigor. CONITEC has clear criteria, high standards, and evaluation cycles that do not wait. That is not an obstacle — it is an advantage for those who arrive well prepared.

Generative AI does not lower the standard of what CONITEC requires. What it does is reduce the time a team needs to meet it — from four to eight months to six to eight weeks. In the context of CONITEC's evaluation cycles, that difference is not operational — it is strategic.

The new RNDS infrastructure and the Zolgensma MEA precedent signal that the next 24 months will redefine the rules of pharmaceutical market access in Brazil. Companies that arrive with solid local evidence, economic models calibrated to the SUS, and AI-trained teams will build an advantage their competitors will not be able to replicate retroactively.

The window is open. In Brazil, more than in any other market in the region, it will not be open indefinitely.