This is not a specific client's case. It is a reconstruction, stage by stage, of where time really goes in an HTA dossier and how much each part can be compressed according to the public evidence available. The figures come from published studies, not marketing promises.

When a market access team hears "AI cuts your dossier preparation time," the sensible reaction is skepticism. Cuts what exactly? At which stage? And at what cost? This article answers those questions by building an illustrative case: a hypothetical but realistic HTA dossier, broken down into its real stages, with each savings estimate anchored in public studies. It does not describe any client's result. It describes what the available evidence allows us to claim about the potential of automation.

1. First, where time goes in an HTA dossier

An HTA dossier is not a single document. It is an evidence package that agencies require teams to assemble under strict deadlines. The core components, according to NICE, CADTH, and ICER guidance, include a systematic review of clinical literature, cost-effectiveness and budget-impact analyses, indirect comparisons where applicable, and the contextual value narrative.

Assembling this demands considerable project management effort. A typical access team brings together health economists, epidemiologists, statisticians, physicians with disease-area experience, and regulatory specialists, with experts iterating drafts and responding to the agency's points of clarification. And the deadlines are unforgiving: NICE's STA-type appraisals, for example, require submission 56 days after the invitation, within an appraisal window of around 35 weeks.

It is also worth naming the recognized bottleneck: finalizing the global economic model is usually the main obstacle to closing a Global Value Dossier. That is why many teams develop the dossier in two phases, first the clinical part and then the economic one.

With that map, we can see where automation has real room and where it does not.

2. Stage by stage: what the evidence says about savings

Systematic literature review (SLR). It is one of the most time-intensive stages and where the savings evidence is strongest. A study presented at ISPOR evaluated a synergistic approach of generative AI with human review and reported that, compared to SLRs done by humans alone, it was 20.3% faster at title and abstract screening, 61.8% faster at full-text screening, and 55.6% faster at data extraction. Another study, on an AI-assisted SLR tool, reported a cumulative 42% time savings across the end-to-end SLR process versus the traditional Excel-based approach.

The honest caveat matters: that same study reported that the auto-generated final report lacked some sections and depth, and that human feedback was valuable to align it with expectations. In other words, the savings are real, but they require human review to reach submission quality.

Economic modeling. The ISPOR Working Group on Generative AI report documents that AI can assist economic modeling (code generation, replication, sensitivity analysis), but is explicit that generative AI shows promise in automating HEOR tasks such as systematic reviews and economic modeling, while it is not yet reliable for autonomous use. Here the savings come from accelerating construction and validation, not from eliminating the economist. Given that the model is the recognized bottleneck, even a partial compression of this stage has a disproportionate effect on the overall timeline.

Generating the value narrative and the dossier. The same ISPOR report identifies dossier development and real-world evidence generation as emerging applications of generative AI. Drafting structured documents, adapting the narrative to different agency formats, and consistency across sections are tasks where assistance accelerates the work, always under expert validation.

Documented per-stage savings

The public evidence places savings in the most processing-intensive phases (SLR, model construction, drafting) in a 40% to 60% range, always with human review as the condition for reaching submission quality.

3. The illustrative case, assembled from those pieces

Imagine a dossier that, in a traditional process, takes a team several months, with the SLR and the economic model as the heaviest stages. Applying only the savings documented above to the stages where the evidence supports them:

If the SLR represents a significant portion of the timeline and automation with human review accelerates it in the 42% end-to-end range, that single stage compresses noticeably. If, in addition, the construction and validation of the economic model (the bottleneck) is partially accelerated, and the dossier drafting is assisted, the combined effect on total time is substantial.

The exact figure for total reduction depends on the specific composition of each project: how much the SLR weighs versus the model, how many comparators there are, how many agencies and markets. That is why this case is illustrative and does not promise a single percentage. What the evidence does support is the direction and the order of magnitude: documented per-stage savings between 40% and 60% in the most intensive phases, which translate into a material compression of the overall timeline when applied with judgment.

What the evidence does not support, and that is why we do not claim it, is that automation eliminates expert review or produces a submission-ready dossier without human intervention. Quite the opposite: the same studies that document the savings also document that submission quality demands the human in the process.

4. The non-negotiable condition: the human in the loop

All the evidence converges on one point. The ISPOR report is direct: generative AI should augment and not replace human expertise in the near term, and it should be adopted with strong controls. NICE, the first agency with a public position on AI in evidence, makes the principle of a competent and informed human in the process a non-negotiable requirement.

This is not an awkward limitation to disguise. It is the very source of the savings' legitimacy. An accelerated but poorly validated dossier is a risk, not an advantage: a fabricated reference or a mis-extracted data point that an agency detects costs more than the time saved. The real value is in accelerating the mechanics while preserving rigor, not in skipping rigor.

5. What to take from this case

The right question is not "does AI cut dossier time?" but "at which stages, with what evidence, and at what cost?". The honest answer:

In the processing-intensive stages (SLR, model construction, drafting), the savings are documented and substantial. In validation, methodological judgment, and evidence strategy, the human remains irreplaceable and should stay that way. The net result, well executed, is a materially shorter timeline without sacrificing defensibility before the agency.

At Quantus we build exactly on that logic: use automation to compress the stages where the evidence supports the savings, while keeping expert market access and HEOR knowledge and human validation at the center. The goal is not to produce dossiers faster for sport, but to accelerate real market access for the client's products, without putting at risk the quality the agency will examine.

References

IntuitionLabs. HTA Dossiers: A Guide to Submissions for NICE, CADTH & ICER. 2026. Available at: https://intuitionlabs.ai/articles/hta-dossier-submission-guide
Costello Medical. The Critical Role of the Global Value Dossier in EU Joint Clinical Assessment. 2025. Available at: https://www.costellomedical.com/what-we-do/value-and-access/role-of-the-global-value-dossier/
ISPOR. The Use of Large Language Models for Systematic Literature Review Automation: An Evaluation of Quality and Time Savings (ISPOR 2025, Poster Session). Available at: https://www.ispor.org/heor-resources/presentations-database/presentation-cti/ispor-2025/poster-session-2/the-use-of-large-language-models-for-systematic-literature-review-automation-an-evaluation-of-quality-and-time-savings
ISPOR. ActiveSLR: Optimizing Efficiency in Systematic Literature Reviews with Artificial Intelligence (ISPOR 2025, Poster Session). Available at: https://www.ispor.org/heor-resources/presentations-database/presentation-cti/ispor-2025/poster-session-2/activeslr-optimizing-efficiency-in-systematic-literature-reviews-with-artificial-intelligence
Fleurence RL, Wang X, Bian J, Higashi MK, Ayer T, Xu H, Dawoud D, Chhatwal J; ISPOR Working Group on Generative AI. A Taxonomy of Generative Artificial Intelligence in Health Economics and Outcomes Research: An ISPOR Working Group Report. Value in Health. 2025. Available at: https://www.sciencedirect.com/science/article/abs/pii/S1098301525023356
NICE. Use of AI in evidence generation: NICE position statement (ECD11). 2024. Available at: https://www.nice.org.uk/corporate/ecd11