Digital Pharma Lab
Contact
What AI changes in market access: monitoring, literature reviews, models, dossiers and safeguards

What AI changes in market access: monitoring, literature reviews, models, dossiers and safeguards

AI speeds up monitoring, literature reviews, modelling and the drafting of market access dossiers. What it changes, its limits and the safeguards to put in place.

November 18, 2026 · Digital Pharma Lab

Artificial intelligence mainly speeds up four market access tasks: monitoring, literature reviews, health-economic modelling and dossier drafting. It replaces neither the judgement of the teams nor the company's responsibility before the French National Authority for Health (HAS) and the Economic Committee for Health Products (CEPS). Its value depends on the safeguards in place: traced sources, human validation, protected data.

Up to date as of 18 November 2026.

Where does AI save time in market access?

Market access is a profession of documents and evidence. A team reads hundreds of publications, follows the authorities' opinions, builds models, drafts long and highly standardised dossiers, then answers the assessors' questions. A large part of this work consists of searching, sorting, summarising and rephrasing: this is precisely where language models are useful.

The gains are concentrated in four families of tasks:

  1. monitoring: opinions of the Transparency Committee and of the Economic and Public Health Evaluation Committee (CEESP), early access decisions, European reports, scientific publications;
  2. literature reviews: searching, study selection, data extraction;
  3. modelling: documentation, checking and sensitivity analysis of health-economic models;
  4. drafting: first versions of dossier sections, summaries, structured answers to the authorities' questions.

In each of these cases, AI produces a draft or a proposal. Decisions, validation and sign-off remain human.

How to use it for monitoring?

Official sources publish a great deal. The HAS puts its committees' opinions and its early access decisions online; the EU joint clinical assessment, provided for by Regulation (EU) 2021/2282, adds its own reports. An assistant can summarise each new opinion in a consistent format: indication, comparators retained, Actual Clinical Benefit (SMR) and Clinical Added Value (ASMR) ratings, points of discussion, reservations.

The value comes from consistency and traceability. A useful monitoring sheet always points to the original document, quotes the passage supporting each statement, and flags what the tool could not read. A summary without a link to its source should not circulate.

Can a literature review be entrusted to AI?

A systematic literature review remains a method, with a protocol, inclusion criteria, dual screening and controlled extraction. AI can speed up each step without changing the rules:

  • suggesting search strings, which the team reviews and adjusts;
  • pre-screening titles and abstracts, with human verification of exclusions;
  • extracting data into a grid, with sample-based or dual checking;
  • spotting inconsistencies between publications of the same trial.

The point to watch is the silent error: a missed study, a miscopied figure, an invented reference. An AI-assisted review must therefore document what the tool did and what people checked, and remain reproducible. The HAS has published, for health professionals, first keys for the use of generative AI in health (in French), which stress checking generated content.

What can it bring to health-economic models?

A health-economic model is judged on its method, its data and the clarity of its assumptions, against the 2020 HAS guide Methodological choices for economic evaluation at the HAS (in French). AI does not choose these assumptions in place of the modellers. It does, however, help to:

  • document the model, assumption by assumption, from the code or the spreadsheet;
  • review the code and spot formula or reference errors;
  • prepare and describe sensitivity and scenario analyses;
  • check consistency between the model, the technical report and the clinical dossier.

These uses address frequent CEESP reservations: poorly justified assumptions, uncertainty insufficiently explored, a model that is hard to review. We detail them in our article on the efficiency opinion.

And for drafting dossiers?

A transparency dossier, an early access application or a health-economic dossier follow precise templates (see our articles on the transparency dossier and early access). AI can produce a first version of a section from validated documents, harmonise terminology, check that each claim points to a piece of data, or prepare structured answers to the assessors' questions.

It must never invent data, a result or a reference. A simple rule helps: every figure in a dossier comes from an identified source document, and a person has checked the match. Assisted drafting remains subject to the company's usual medical, regulatory and legal review process.

What limits should be kept in mind?

Language models have known weaknesses, which matter more in market access than elsewhere, because each error can end up in a document filed with an authority.

  • They can invent a reference, a result or a quotation, with the same assurance as accurate information.
  • Their knowledge has a date: without access to up-to-date sources, they ignore a revised doctrine, a new opinion or an amended text.
  • They reproduce the biases of what they are given to read: a partial selection of articles produces a partial summary.
  • Their answers vary from one run to the next, which complicates the reproducibility of a method.

None of these limits rules out their use. They require working from documents provided to the tool, rather than from its memory, and checking what comes out.

What safeguards should be put in place?

Safeguards are what make AI usable in a profession where every statement can be challenged. The most useful ones are concrete:

  1. choose enterprise tools whose terms exclude the use of data to train models, with appropriate confidentiality commitments;
  2. never enter identifiable health data into a tool not designed for it, consistent with the General Data Protection Regulation (health data are special-category data, Article 9);
  3. require sources: each output points to the documents used, and anything unsourced is not reused;
  4. validate by a human any output intended for a dossier or an authority, with a record of that validation;
  5. document the use: tool, version, instructions, steps concerned, so that it can be explained to an assessor;
  6. train teams in the uses, limits and typical errors of language models.

An enterprise tool reduces risks; it does not exempt you from any of the other safeguards.

What does the regulatory framework say?

Regulation (EU) 2024/1689 on artificial intelligence (AI Act), amended in 2026 by the "Digital Omnibus on AI" Regulation (EU) 2026/1744, governs AI systems according to their level of risk. The uses described here, searching, summarising and drafting, do not generally fall among the high-risk systems it lists; each case should nonetheless be checked against the text. It also requires providers and deployers of AI systems (a company using an AI tool in its business is a deployer) to take measures to support the development of AI literacy among their staff, without requiring any specific level for each individual (Article 4, as replaced by Regulation 2026/1744).

The processing of personal data, and a fortiori health data, remains subject to the GDPR. Finally, responsibility for the content of a dossier filed with the HAS remains with the company, whatever tool was used to prepare it. Our AI Playbook for Pharma & MedTech details this framework, by company type and size.

Where to start?

Start with a high-volume, low-risk task, such as monitoring published opinions, before moving on to literature reviews and then drafting. Measure the time saved and the rate of corrections needed: these are two simple indicators of whether the tool helps.

To govern these uses over time, inventory your AI tools, assess their compliance posture and follow a roadmap, you can rely on AI Compass and on our AI in healthcare expertise. For the dossiers themselves, discover our market access expertise.

This article provides general information; it does not replace regulatory advice or legal advice tailored to your situation.

Sources

Sources accessed on 18 November 2026.

Our expertise on this topic

  • Market Access

    Building robust access strategies, solid health-economic evaluations, and HAS dossiers that withstand scrutiny.

  • AI in Healthcare

    Integrating artificial intelligence into healthcare organizations, from strategy to operational deployment.