Value dossier automation means generating a market-specific dossier draft from a governed evidence base, with every claim traced to the source passage supporting it, for review by a market access professional. It does not mean a system that produces a submission-ready document from a data dump.
The distinction matters because the second thing is what gets demonstrated and the first thing is what gets deployed. This article covers how the working version is built, what the review step looks like, and where the effort actually goes. For the wider picture, see AI for market access and HEOR.
The pipeline
1. The governed evidence base
Everything depends on this and it is most of the work. A governed evidence base is not a document repository — it is a curated, permissioned, versioned collection in which each source is identified, its provenance recorded, and its content structured against a schema that captures indication, population, comparator, endpoint, outcome and study design. Searching a flat pile of PDFs for "progression-free survival in second-line" returns approximate matches; querying a structured base for a specific endpoint in a specific population returns the right studies.
2. The dossier structure
Each target format — a global value dossier template, an AMCP dossier, a specific HTA body's submission structure — is represented as a defined skeleton: sections, required content per section, evidence requirements, and on-label constraints for the market. This is configuration, not generation, authored once by people who know the requirements.
3. Retrieval and assembly per section
For each section, the system retrieves the relevant evidence, filtered by the market's constraints, and drafts content grounded in the retrieved passages. Each claim carries a reference to the specific passage and source version supporting it — not a bibliography at the end, a passage-level link per statement.
4. Verification before review
Automated checks run before a person sees the draft: do all citations resolve; does each cited passage actually support the claim; are figures consistent with sources; is anything off-label for this market. Errors caught here do not consume expert time.
5. Review and adaptation
The market access professional receives a draft with evidence attached. They verify claims against the displayed passages, adapt the argument for local context, make the strategic judgements, and approve. The output is a first draft that arrives with its homework shown — not a finished submission.
What it does not do
It does not construct the value argument — which comparator is defensible, what a given HTA body will accept, how to handle a weakness in the data. It does not resolve missing evidence; if the study that would support a claim does not exist, the system correctly produces nothing. And it does not remove the review step, nor fix an evidence base nobody governs — deployed over an unharmonised repository, it produces fluent, confident, untraceable drafts, the worst possible outcome because they look finished.
Where Eclypse sits: one governed evidence base generates the global dossier, regional dossiers and local value stories on demand, with provenance tracked on every claim. See the market access evidence base case study this pattern is drawn from.
The determining question is not whether dossier generation works. It is whether your evidence base is in a state where it can. See AI agent use cases in healthcare for where this fits among other working deployments.
Common questions about value dossier automation.
What is value dossier automation?
Generating a market-specific value dossier draft from a governed evidence base, with each claim traced to its source passage, for review by a market access professional.
Can AI produce a submission-ready value dossier?
No. It produces a draft grounded in approved evidence with claims traceable to sources. The professional verifies, adapts, and makes the strategic judgements.
What is a governed evidence base?
A curated, permissioned, versioned collection of evidence structured against a schema capturing indication, population, comparator, endpoint and study design, with provenance recorded per source.
What happens if the evidence to support a claim does not exist?
A correctly built system produces nothing for that claim rather than generating plausible content — the intended, safer behaviour.
Bring us your hardest submission — we'll tell you what the evidence layer would need to look like.
Book a working session
ECLYPSE
AI
Our proprietary AI orchestration platform for healthcare: one engine, a registry of reusable task modules and domain agents, and a governed knowledge base — deployed inside your walls and run by your team.