What it actually takes: building an AI service in an NHS trust
This session sets out a case study of building an in-house AI service within an NHS Trust, from early idea through to operational delivery. It is not a transformation success story. It is a practical account of what it actually takes to turn interest in AI into a governed, functioning service.
The starting point was not a formal strategy but a practical need. Teams across the Trust were under pressure, working with fragmented data and recurring operational challenges. The early aim was to create a small, credible internal capability that could convert this interest into safe, usable solutions.
Progress depended on translating AI into terms that senior leaders and boards could engage with. The case had to be framed around risk, clinical safety, operational benefit and financial pressure rather than innovation. Gaining support required plain English communication alongside honest acknowledgement of uncertainty, trade-offs and constraints.
Board approval was only a midpoint. The harder work came afterwards: recruitment challenges, governance processes, assurance requirements, and defining an operating model that could function within day-to-day NHS realities. Delivery required aligning technical capability with clinical credibility, governance frameworks and service demand.
Building the service was not linear. It involved scepticism, organisational politics, financial scrutiny and competing priorities. Trust and credibility had to be earned continuously through engagement, transparent decision-making and a clear route from idea to deployment.
The presentation reflects on early missteps, including overemphasising AI as a capability, underestimating governance complexity, and the difficulty of recruiting the right skill mix. The broader lessons are about anchoring work in real service problems, communicating risk and value clearly, and ensuring governance and clinical engagement are in place before scaling.
AI adoption in the NHS is an iterative process shaped by real-world constraints. What determines whether it works is alignment with organisational priorities and credibility built over time.
