What is the Digital Blueprint for NHS Wales?

A free, public systems-dynamics analysis of Digital Health and Care Wales (DHCW), the national body that builds and runs the digital systems behind NHS Wales. It diagnoses why the organisation keeps failing to deliver, and sets out six sequenced interventions, a target architecture and operating model, and a costed 36-month reform plan. It is published at bluenhs.org under CC BY 4.0, with two working papers archived on Zenodo (DOIs 10.5281/zenodo.21444618 and 10.5281/zenodo.21445084).

Who wrote it, and what is it based on?

Rafal Bergman, the former Chief Product and Technology Officer at DHCW. It is built only from the public record: Welsh Government publications, Audit Wales reports, Senedd records and the organisation's own published board papers, 89 numbered primary sources in all. No confidential material is used, and every claim can be traced to a source anyone can read.

What is systems dynamics, and why use it here?

A discipline founded by Jay Forrester at MIT and made practical by Donella Meadows. Instead of asking what broke, it asks what structure keeps producing the breakage: what accumulates or drains over time, which feedback loops drive it, and which recurring traps those loops form. It suits problems that survive every conventional fix, because it explains why the fixes were absorbed and shows where an intervention would actually change behaviour.

Is this the same as the Blueprint from a Knowledge Audit?

No. The Blueprint delivered by a Knowledge Audit is a private, ranked plan for where AI will pay first in one business. The Digital Blueprint for NHS Wales is a public analysis of a national health IT body. They share a habit of mind, diagnose before you build, but they are different pieces of work.

Does this method apply to a company?

Yes. The traps are not specific to the public sector. AI pilots that never reach production because a vendor substitutes for a missing capability, delivery standards that erode quarter by quarter, and AI adoption measured by activity instead of outcomes are the same structures at business scale. The diagnostic phase of a CTO Zen engagement uses the same approach, sized to the organisation.

Digital Blueprint for NHS Wales. A systems diagnosis of national health IT

The question

Digital Health and Care Wales (DHCW) is the national body that builds and runs the digital systems behind NHS Wales. Over five years it received around £600M of public funding and grew its headcount by 87%. In April 2026 Welsh Government escalated it to Level 4, Targeted Intervention. Every obvious fix had been tried along the way: more people, new vendors, new strategies. None changed the outcome.

So the useful question was not "what broke?". It was the one I ask in any organisation where repeated fixes fail: what structure keeps producing the failure?

I knew the organisation from the inside, having served as its Chief Product and Technology Officer. After leaving, I set out to answer that question in public. The analysis is built only from the public record, with no confidential material. Every claim traces to a source anyone can read.

The method

Systems dynamics, the discipline Jay Forrester founded at MIT and Donella Meadows made practical, replaces "what broke?" with "what pattern keeps producing breakage?". I applied it in four steps.

Stocks. What accumulates or drains over time. Not only money and headcount, but trust, delivery capability and institutional knowledge. I mapped 18 of them. The visible ones looked healthy. The ones that decide whether anything ships were depleting.
Feedback loops. The circular chains of cause and effect that drain those stocks. I documented 11 reinforcing loops in two clusters: five that produce delivery failure, and six that protect that failure from the mechanisms meant to correct it.
Traps. When loops interact they form recognisable archetypes, the system traps Meadows described. Seven were active at the same time.
Leverage points. Meadows ranked the places to intervene in a system, from shallow (budgets, headcount, targets) to deep (information flows, rules, goals).

That last step is where most reform goes wrong. Parameter-level fixes sit at the shallow end. The organisation absorbs them and carries on behaving exactly as before, which is why adding hundreds of staff changed nothing. The analysis looks for the points deep enough that behaviour has to change, and sequences them so that one reform is not quietly undone by the loop next to it.

The deliverable

The Digital Blueprint for NHS Wales is free to read and reuse under CC BY 4.0. It is structured like an engagement deliverable, not an essay:

A diagnosis. The 18 stocks, 11 loops and seven traps, each tied to numbered evidence.
Six sequenced interventions over 36 months, ordered by dependency. One is aimed at the funder rather than the delivery body, because reforming either one alone reproduces the problem.
A target architecture and operating model. A federated, standards-led design on the pattern Denmark and Estonia already run: a small national standards body, clinical delivery owned by the health boards, and open interoperability between them. Every component is in operational use somewhere in northern Europe today.
A costed plan. Reform costed at £5–15M one-off, against an estimated £100–150M a year in direct waste under the status quo, with a transition plan, a 36-month timeline and a monitoring framework that tells premature abandonment apart from genuine failure.
An evidence base. 89 numbered primary sources: Welsh Government publications, Audit Wales reports, Senedd records and the organisation's own board papers.
Two working papers archived on Zenodo, each stating its method, limitations and falsifiable predictions: the diagnosis (doi:10.5281/zenodo.21444618) and the design (doi:10.5281/zenodo.21445084).

What it shows

You do not need to run a national health service for any of this to matter. The same structures turn up in every technology and AI transformation I have worked on.

Adding headcount does not fix a delivery system. Hiring is a flow. Delivery capability is a stock, and it is drained by whatever loop is already running. Leave the loop in place and new people learn the old behaviour within a year. Companies do the same when they hire an AI team before fixing how decisions get made.

Governance before tooling. The deepest interventions were about information: what gets measured, what gets published, and whether bad news can travel upwards. A new platform does not change any of that. If an organisation cannot surface problems honestly, better tools only help it produce the same results faster.

Measure the stock, not the flow. Hiring rate, spend and activity are easy to report and easy to mistake for progress. Capability, trust and knowledge are what you are actually trying to build.

Respect the delays. The right intervention often looks like a failure for the first two quarters. Agree up front what success looks like at 12 and 36 months, so you neither abandon the right change too early nor keep funding the wrong one for years. It is the same reason I agree kill criteria before any AI deployment.

Separate standards from delivery. A thin centre that sets standards, with teams that own delivery, beats a central body that tries to build everything. That holds for an enterprise AI platform team as much as for a national health system.

How this applies to your organisation

The same method sits behind the diagnostic phase of every engagement, sized to the business: find the loop producing the problem, name the trap, then find the point deep enough to change behaviour. I set out how in How I work.

The full analysis is at bluenhs.org. Start with the method or the seven traps. And if your organisation keeps applying fixes that do not stick, let's talk.