Case study
National Public Health Publications Service
A service designed around the people who need public-health guidance, not around the organisational chart that produced it.
Evidence status: confirmedDisclosure: Anonymised
- Role
- Programme Delivery Lead & Service Designer
- Timeframe
- Multi-year programme
- Sector
- National public health
- Capabilities
- Service design · Programme delivery · Team leadership · Cloud architecture · Semantic search
- Evidence
- confirmed
- Disclosure
- anonymised
System view
Architecture and relationships
Diagram not yet published
A publishable system diagram is not yet available. The case study describes the architecture and organisational relationships without exposing client detail.
Decision records
The choices that shaped the work
Decision 01
Make the service user-need led, not content-first
- Alternatives
- Digitise the existing publication estate; organise the service around directorate structures; or start from the tasks users are trying to complete and design the content and service around them.
- Chosen
- Start from user needs and the tasks people bring to the service, then shape content, search and support around those tasks.
- Why
- A content-first migration would have preserved the organisation's internal logic and carried its support burden into a new interface. The measurable goal was fewer people needing to ask where something was.
What would change my mind
Evidence that the dominant user need was a narrow, known publication lookup rather than task-based discovery, or a statutory constraint that made a user-need-led structure impossible to maintain.
Context
A national public-health publications service is not a library in the ordinary sense. It is the place where guidance, evidence and operational information meet the people who have to act on them. If the service is hard to navigate, the cost does not stay inside the website. It becomes support requests, duplicated advice and slower decisions.
The programme needed a service that could carry a large and varied publication estate without making every user learn the organisation’s internal structure first.
Challenge
The existing estate had grown around publishing teams and directorates. Content was findable if you already knew what you were looking for. The service also carried a steady load of inbound publication and support requests, many of which were really questions about where something lived or which version was current.
The programme had to move from Alpha to Beta under the Government Digital Service assessment model. That meant proving user need, service design, accessibility, security and operational readiness, not just shipping a new front end. The delivery team was cross-functional and large: roughly 10–15+ people across architecture, engineering, content, design, delivery and support.
The ambiguity was commercial as well as technical. The service had to earn continued investment while the team was still learning what users needed.
My role
I was Programme Delivery Lead and Service Designer. I owned overall delivery, the financials and line management of the 10–15+ cross-functional team. I unblocked architecture, DevOps, frontend, backend and AI engineering, and I expanded the account as the work moved from a project into a broader service relationship.
That combination mattered. Service design without delivery authority produces a beautiful deck. Delivery without service design produces a faster version of the wrong thing. I held both.
How it fits together
The service was built with Django, Wagtail, React and TypeScript. The platform ran on AWS: ECS Fargate for application workloads, Aurora PostgreSQL for structured data, S3 for assets, CloudFront for delivery, API Gateway and Lambda for service endpoints, EventBridge and SQS for asynchronous work, and CloudWatch for operational visibility. Microsoft Entra External ID and MSAL handled identity where the service needed it.
Semantic search and personalisation were part of the service direction: helping people find the right publication without needing to know the publishing team’s vocabulary. The content model and the support model had to change together. A better search box over an unchanged content estate would have moved the problem, not solved it.
The trade-offs
The decision record accompanying this case study is make the service user-need led, not content-first. It set the shape of the content model, the search experience and the support route.
The alternative was a migration. It would have been easier to plan and easier to sell internally: move the existing pages, keep the existing taxonomy, add search. It would also have preserved the reasons people were asking for help. We chose the harder path because the outcome we wanted was not a new interface. It was fewer people needing to ask.
Trust and failure
Public-health content carries a high cost of being wrong. Accessibility, version control, publication governance and secure access were part of the service, not a compliance layer added at the end. The GDS assessments forced us to show evidence for the experience, not just describe it.
The most useful failure signal was support demand. A service that looks calm while the support queue grows is not working. We treated the request volume as a design metric, which made it harder to celebrate a launch that had simply moved the work elsewhere.
Delivery
Delivery ran from Alpha into Beta with a cross-functional team and a staged assessment route. The Alpha and Beta GDS assessments passed first attempt. That did not mean the work was easy. It meant the team had assembled the evidence and made the decisions before the assessment, rather than treating the assessment as the design process.
I ran the delivery rhythm, owned the financials and line-managed the team. I also worked directly with the engineering leads when architecture, DevOps or frontend decisions were blocking progress. The account expanded as the service proved it could carry more than the original brief.
Evidence
- Moved from Alpha to Beta.
- GDS Alpha and Beta assessments passed first attempt.
- Inbound publication and support requests reduced from around 50 per week to fewer than 5.
- £2.2M project value.
- £900K follow-on work.
- £3.1M account expansion across the UK Health portfolio.
The commercial figures are included because they are part of the delivery story, not because they are the point of it. The point is that the service changed the operating load.
Reflection
The strongest part of the work was refusing to treat the existing content estate as the requirement. The weakest part was how long it took to make the support model part of the design.
What did not work
A content-first migration would have been faster to start and slower to finish. We also learned that semantic search is not a substitute for a coherent content model. If the underlying material is organised around internal ownership, better retrieval can make the confusion more efficient.
I would bring the support team into the design earlier and treat their request log as primary research. I would also define the service’s evidence baseline before Alpha, so the improvement in support demand could be measured against a stable starting point rather than reconstructed later.
Related proof
- National surveillance and surge readiness: technical leadership in the same public-health portfolio.
- Responsible AI technical assurance: assurance work with a global consumer-goods organisation.
Evidence plate
Evidence visual not yet published
Evidence in this case study is drawn from approved programme records. Figures are attributed to their source and are not recombined into new claims.