The challenge
NHS Wales, a 300-strong technical organisation responsible for digital innovation and a strategic portfolio of national products underpinning healthcare delivery, was facing the challenges common to institutions of this type: legacy infrastructure requiring modernisation, fragmented data systems making evidence-based decisions difficult, and technology procurement and delivery patterns that weren't generating value at the scale of investment.
The mandate was transformation, not incremental improvement. That meant making hard decisions about which investments to protect, which systems to replace, which vendor relationships to restructure, and how to build a technology organisation capable of sustaining change beyond the initial programme.
The scale
Operating at national scale means the tolerance for error in technology architecture decisions is close to zero. When infrastructure underpins services used by millions of people, delivery risk has real consequences, not just commercial ones. This creates a different kind of engineering discipline: one that values predictability, auditability, and resilience as primary constraints, not afterthoughts.
It also means working within governance structures that most technology leaders, and most consultants, have never encountered. Procurement rules, regulatory obligations, clinical safety considerations, data governance requirements that extend across dozens of integrated systems and thousands of endpoints. Building something that works in this environment requires a different kind of architecture thinking.
What was delivered
The technology transformation programme delivered £28M+ in annual savings through a combination of:
- Infrastructure rationalisation: consolidating a fragmented estate onto modern, maintainable platforms
- Vendor contract restructuring: renegotiating and, where appropriate, exiting commercial arrangements that weren't delivering value
- Process automation: identifying high-volume administrative workflows amenable to automation and building systems to support them
- Data architecture modernisation: addressing the fragmentation that was preventing analytical capability and evidence-based management
Beyond the financial outcome, the programme established a technology capability (people, processes, and infrastructure) capable of sustaining delivery without dependency on external consultancy.
Alongside the transformation programme, I served as Technical Lead for the Welsh Government Commission for AI in Health & Social Care, creating the national AI adoption strategy for healthcare in Wales.
What this teaches about AI implementation
The patterns that cause public sector technology projects to underdeliver are the same patterns that cause enterprise AI projects to underdeliver:
Data fragmentation is the real problem. Every major transformation programme I've led has uncovered the same root cause: data that cannot be reliably retrieved, combined, or analysed. Before any AI or automation project, the data architecture question has to be answered honestly.
Stakeholder trust is built incrementally. In a public sector context, you cannot mandate adoption. You have to demonstrate value early, to a small group, and let that evidence do the work. The same principle applies to any organisation rolling out AI to a sceptical team.
Technology governance isn't bureaucracy. The governance structures that slow down public sector delivery serve a purpose, and when AI systems are making consequential decisions, governance matters commercially as well as ethically. The EU AI Act compliance questions that startups are now facing are a version of this challenge that public sector organisations have navigated for decades.
Procurement is a strategic capability. The difference between a vendor contract that creates value and one that creates lock-in is usually made in the first 90 days of a relationship. Getting this right, especially for AI infrastructure where the landscape is moving fast, is one of the most impactful decisions a CTO can influence.
Why this matters beyond public sector
Most organisations I work with aren't operating at this scale, but the principles are the same. Architecture decisions that don't account for data governance, vendor lock-in, and regulatory exposure create problems at any scale. They surface during due diligence, during audits, and during the conversations that determine whether an AI investment was worth it.
The engineering discipline that comes from environments like this one, where the cost of getting it wrong is very high, is what I bring to every engagement.
Related: How I approach technology transformation · AI Transformation Sprint