AI Validation & Clinical Safety
Independent clinical validation for health AI you can stand behind
When an AI product enters a clinical workflow, buyers ask one question: is it safe, and can you prove it? We provide independent evaluation, clinical safety assurance and the evidence pack that moves you through NHS procurement.
Who this is for
Built for teams putting AI near a clinical decision
Health-tech companies preparing for NHS adoption, medical device manufacturers needing a clinical safety case, and provider organisations deploying AI into live pathways all face the same bar: demonstrable clinical safety and credible evaluation. We give you both — independently, and in language procurement and clinical-safety officers trust.
Our work is evaluation, assurance and governance. We are deliberately independent of any single vendor or platform, so the evidence we produce carries weight with buyers.
The evaluation stack
A structured framework, not a one-off opinion
Every engagement follows a transparent, repeatable evaluation stack so findings are defensible and comparable over time.
- Intended-use & risk definitionDefine the clinical claim, target users, deployment context, data flows and what "safe" means for this product.
- Evaluation designChoose metrics, datasets and test scenarios that reflect real clinical use — not just headline accuracy on curated data.
- Performance & error analysisQuantify performance, characterise failure modes, and analyse where and for whom the model is weakest.
- Bias, safety & human factorsTest for subgroup bias, automation bias, and the human-in-the-loop conditions needed for safe use.
- Clinical safety artefactsProduce or review the DCB0129 safety case and hazard log (manufacturer) and support DCB0160 deployment safety.
- Assurance pack & readoutDeliver a procurement-ready evidence pack and a clinical readout your buyers and board can act on.
What you receive
Concrete deliverables, mapped to buyer expectations
- Intended-use and clinical risk statement
- Evaluation protocol and methodology note
- Performance, error and subgroup analysis report
- Bias and human-factors review
- DCB0129 clinical safety case & hazard log (or review)
- DCB0160 deployment safety support
- DTAC-aligned assurance summary
- Executive readout for board / procurement
What we do — and don't — claim. We provide clinical evaluation, implementation support and governance guidance. We do not provide direct patient diagnosis, triage or emergency advice, and we do not replace your regulatory or notified-body obligations. Regulatory applicability depends on intended use, deployment context and local governance.
Answers
Frequently asked questions
What does clinical validation of an AI product involve?
Clinical validation independently assesses whether a health AI tool is safe and fit for its intended clinical use. It covers intended-use definition, evaluation against representative data and workflows, performance and error analysis, bias and human-factors review, and the clinical safety artefacts (DCB0129) a buyer needs. The output is an evidence pack a procurement team can rely on.
Do you certify or regulate our AI as a medical device?
No. We provide independent clinical evaluation, safety assurance and governance guidance to support your own regulatory and procurement journey. Whether software is a regulated medical device depends on its intended purpose under MHRA rules; we help you understand applicability and prepare the evidence, but formal conformity assessment and regulatory decisions remain with you and your notified body.
What is the difference between DCB0129 and DCB0160?
DCB0129 is the clinical risk management standard for the manufacturer of a health IT system; DCB0160 applies to the healthcare organisation deploying it. Manufacturers produce a clinical safety case and hazard log under DCB0129; deploying NHS organisations manage local clinical risk under DCB0160. We support both sides.
How quickly can a validation engagement start?
A scoping call typically happens within a few business days. A focused evaluation sprint can begin within two to three weeks of agreeing scope, depending on data access and governance approvals.
Is this relevant for an ambient scribe or documentation AI?
Yes. Ambient documentation tools have specific safety, accuracy and DPIA considerations. See our dedicated Ambient Scribe Evaluation service.
Validate your AI to a standard buyers trust
Book an AI validation scoping call or request an assurance proposal.