Over the past few months I’ve been thinking a lot about how to prove risky assumptions, the value of building common components, and how to test, learn and deliver at pace in the NHS.
I’ve been reflecting on the successes of our work in Personalised Prevention Services (PPS) and what we should optimise for next.
(These are my own reflections, not the official PPS strategy. My views are shaped by my formative product experience in government, leading two platform components — GOV.UK Notify and GOV.UK Forms — and spending most of my days advocating for doing things well once and reusing them across government, when there’s a proven need.)
We started with a platform approach
Our vision in PPS is to make it easy for people to understand their health risks and find the right services to stay well.
When we began, we structured our work around three core parts of that journey:
- Helping people understand their risks — through health assessments and risk calculators (we completed an excellent discovery documented in our health assessments design history)
- Helping people understand what to do next — by connecting them to the most clinically and contextually relevant next step, summed up in the personalised prevention platform discovery design history
- Helping people take action and stay on track — by supporting ongoing engagement and habit formation
The early focus was on shared components — reusable building blocks that could support many prevention services. The idea was that if we built things well once, we could scale them efficiently across the system.
Why we decided to switch
In May, we shifted direction. Instead of building separate components, we focused teams on five ‘thin-slice’ end-to-end services:
- Weight management
- Lung cancer screening checks
- Talking Therapies
- AI Health Coach
- NHS Health Check online
This shift came from a simple insight: you can’t test behaviour change in a research interview.
Asking someone if they’d go out for a run after seeing a beautifully designed digital journey doesn’t tell you if they’d actually do it. Especially not on a dark November evening, when Celebrity Traitors is on. Behavioural change happens in the messy, unpredictable reality of people’s lives, not in prototypes.
The thin-slice approach allowed us to test our hypotheses in the wild, quickly, and without assuming something is a ‘capability’ that would become scalable (spot on formulated in Richard Pope’s post on capability mapping).
We’ve already seen encouraging signs. In our Talking Therapies pilot, for example, people who hadn’t previously engaged started self-referring after receiving personalised messages.
We also expected this approach to speed up delivery. Building platforms is (usually) slower than developing individual services. And smaller, self-contained teams with tighter scopes can move faster and navigate governance and funding more easily. This matters because prevention services are still largely organised around individual conditions or initiatives.
Reality check – the assurance needed for real data
What we’ve learned since moving to this thin slice use cases pilot approach is that testing in the wild is revealing and very valuable, but also slow and hard.
The thin-slice model has helped us optimise for understanding behaviour change. It’s deepened our evidence base about what helps people stick with healthy habits. And it’s aligned our work with how prevention is currently structured across the NHS and local systems, which has made progress easier to achieve.
However, pilots also bring real-world complexity, dependencies, and the kind of slow learning that comes from working with people’s actual health behaviours. The partnership, engagement and assurance steps required to get a product with real users and data, even for a simple pilot, are still very significant, and similar (or perhaps even the same?) to what’s required for a private beta. So doing true experimentation is hard.
So while this approach has given us valuable insights, it only tells part of the story.
What we lost by switching
By moving from a platform-first model to thin slices, we’ve stopped testing some of our riskiest assumptions.
We’re not currently proving that it’s possible to move from a condition-led approach to a person-led one where each individual is offered the most important next step based on both their clinical needs and personal circumstances.
For example, while a weight management service might lower someone’s cardiovascular risk, tackling anxiety through Talking Therapies first might lead to better long-term outcomes.
We’re also not actively exploring whether shared capabilities could deliver value and efficiency at scale. In particular, we may be missing opportunities to test common needs around:
- Health assessment tools — many digital self-assessments share core data sets, making reusable components feasible and valuable.
- Risk calculation tools — work led by Dan Booker-Macedo on integrating an updated CVD risk calculator (QRisk3) into GP IT systems highlights how shared tooling could support multiple health risk calculations in future.
- Connection to the next step — Ralph’s recent provocatype shows the potential of joining up prevention journeys across vaccinations, screening, behaviour change and other services. Realising that vision requires a common back-end infrastructure, so any prevention service, national or local, can plug in and be surfaced when relevant.
From experience, I know that building a true platform or shared capability doesn’t happen by accident. It needs deliberate investment in the fundamentals: self-service onboarding; effortless adoption; and funding models that incentivise growth.
Personalised Prevention is a system
For me, PPS is both a principle and a system. To deliver the vision of a truly personalised prevention experience, we need to approach it that way.
James Plunkett’s excellent blog post makes the case for several operating patterns that need to be true to realise digital-era healthcare. Using his language, delivering PPS’s vision would require platform and components thinking, as well as curation, system enablement, and seeding and spreading.
Our delivery approach in PPS should mirror the system change we’re aiming to create.
At the moment, we’ve optimised for understanding the behavioural change side of the proposition, which is valuable and difficult work. But right now we’re not testing the system-level levers that make personalised prevention truly scalable. To do both, we need to think and act as a system too.
In hindsight — and what next
If I were starting again, I might argue for a hybrid model: running thin-slice pilots to test behaviour change, while simultaneously investing in shared foundations that could later enable scale.
That’s where we’re heading next. Our Deputy Director, Emily Houghton, will lead a new Prevention in the NHS App team, exploring how to curate all aspects of prevention into a single, joined-up journey.
We’ll need to manage coordination and dependencies carefully. As soon as multiple teams are involved, challenges around flow and alignment grow — as Team Topologies highlights.
And, regardless of the structure, the biggest challenges remain the same: adoption, funding, and system change.
Closing thought
Personalised prevention is both a principle and a system. Our delivery approach should mirror that by combining real-world experimentation with the deliberate building of shared capabilities.
We’ve learned a lot from shifting our approach. The next step is to bring those lessons together to make prevention not just more personal, but more connected and scalable across the NHS.
Irina Pencheva 
Leave a comment