£900,000 AI Award Targets the Evidence Gap Before Health Trials Begin

Vamstar, a London life-sciences AI company, says it has secured £900,000 through the Sovereign AI programme to develop a platform intended to bring market-access questions into clinical trial planning.

The company’s central proposition is straightforward: developers should understand the evidence, health-economic and pricing questions a payer may ask before a trial protocol is locked. In its announcement, Vamstar says the funding will take an earlier proof of concept to a working demonstrator over the next 12 months.

The award matters less as another AI funding headline than as a test of whether commercial evidence can be made useful early enough to influence development decisions. In pharmaceuticals, biotech and medtech, the cost of discovering a material evidence gap late in the process can be far higher than the cost of asking a better question at the design stage.

The commercial decision comes earlier than the submission

Clinical development and market access are often treated as adjacent workstreams. One establishes whether a technology performs as intended; the other later makes the case that it represents good value for a health system.

In reality, the two are tightly connected. Choice of comparator, patient population, outcome measure, follow-up period and treatment pathway can all shape whether later evidence is relevant to a reimbursement or adoption decision. Once recruitment begins, changing those choices can be difficult, costly or impossible.

That does not mean a payer’s eventual decision can be predicted with certainty. Health technology assessment (HTA) involves judgement as well as analysis, while standards and local context vary between markets. But earlier testing of plausible evidence requirements could help teams identify where their planned trial has weak coverage of the questions that will affect adoption.

This fits with the discipline set out in NICE’s health technology evaluation manual. NICE expects evidence to be relevant to the decision problem, transparently reported, systematically assembled and methodologically sound. Its methods also recognise that assessments may draw on clinical studies, real-world evidence, registries and economic modelling, depending on the question at hand.

Vamstar’s claim is about connecting fragmented evidence

According to Vamstar, its proposed cloud platform will combine clinical-trial evidence, health-economic modelling and market and pricing information. It plans to extend the demonstrator across six therapeutic areas and make it available to pharmaceutical, biotech, medtech and NHS organisations.

The intended commercial use is not simply faster document production. It is decision support: helping teams compare design options against likely evidence expectations and the price or value proposition those options could support.

That is a more useful framing for AI in this setting. The hard problem in HTA is rarely a shortage of material alone. It is the need to define the right decision question, establish what data is fit for purpose, make assumptions visible and understand uncertainty. A system that merely retrieves information at speed does not resolve those tasks.

Vamstar says its earlier Innovate UK-funded proof of concept automated HTA evidence retrieval and health-economic modelling, with results consistent with published NICE outcomes at more than 90% reproducibility for ICER and QALY measures. The company says that work was presented at an ABPI conference in 2026. That performance claim should be treated as Vamstar’s own until independently published validation provides the methodology, comparator set and error analysis needed to assess it.

Traceability is the practical standard

For regulated health decisions, a persuasive answer is not enough. Users need to see the underlying source, understand how it was selected and challenge the assumptions that connect evidence to a recommendation.

Vamstar says each output will link back to source material and that the platform will follow NICE methods and governance requirements. Those are sensible design aims, particularly where AI is used to assemble or interpret evidence across clinical and commercial sources. But traceability should be viewed as a minimum operating condition, rather than evidence that a conclusion is correct.

The relevant test for prospective users will be practical. Can a market-access lead identify the source supporting an input? Can a health economist alter an assumption, inspect the effect on the model and explain the change? Can a clinical team see when an apparently attractive design choice creates uncertainty elsewhere? And can the organisation document the human judgement behind the final decision?

The wider Sovereign AI initiative is designed to support UK-based AI companies as they grow and scale from Britain. Vamstar’s announcement places that policy ambition in a demanding health-technology setting, where the value of a platform will depend on its reliability within real development and assessment workflows rather than on the sophistication of its model architecture alone.

The demonstrator has a clear burden of proof

Vamstar is right to focus on the period before a protocol is fixed. That is where evidence planning can still alter a commercial outcome, and where clinical, regulatory and market-access teams have most to gain from working from the same view of the decision.

The next stage is therefore not simply technical deployment. The demonstrator will need to show that it produces auditable, clinically credible and commercially useful insights without obscuring uncertainty or encouraging teams to mistake a modelled forecast for a payer decision.

If it can do that, the platform could make evidence strategy more connected to product development. If it cannot, it risks becoming another layer between teams and the careful judgement that HTA requires.

This article is based on information distributed through Pressat. It has been edited by Expert View Media for clarity, context and length.