Challenge
NHS prevention programmes helped people get started, but struggled to sustain engagement and long-term behaviour change. Support was fragmented or generic, leaving people without the ongoing guidance to manage their health independently.
Solution
I led product design for an AI Health Coach feature within the NHS App, combining behavioural science and personalised interventions to help people build healthier habits and support lasting behaviour change.
Outcome
Delivered an AI-powered preventative healthcare experience from discovery to Alpha in under six months, translating 60+ pain points into three priorities: behaviour change, inclusion and clinical safety, with a planned reach of 39 million NHS users.
Problem
Health support was fragmented and lacked continuity.
NHS Better Health offered separate apps for different health goals, such as quitting smoking or losing weight. This meant support was split across different experiences rather than connected around people's wider health needs.
Even within individual programmes, support felt generic, making it harder to respond as people's motivation, circumstances and goals changed.
What the research revealed
People needed connected, ongoing support.
I synthesised NHS England and DHSC research, identifying 60+ pain points across four stages of the prevention journey. The findings highlighted an opportunity to bring health support together and make it more personalised, continuous and responsive to people's changing needs.
Design opportunity
Bring different areas of behaviour change into one AI Health Coach, offering connected support that adapts as people's needs evolve.
Strategic direction
I brought together behavioural, clinical and product perspectives to shape the direction of the AI coaching experience. Using the as-is journey, we mapped How Might We questions to identify opportunities and guide the product strategy.
Main workshop question…
How might we make AI coaching feel personal, useful and safe?
Three themes emerged to guide the product direction:
Behaviour change: Help people build habits that last.
Inclusion: Make support accessible to different needs and circumstances.
Clinical safety: Keep AI within clear boundaries and provide safe routes when people need more help.
Design decisions
01 | Personalisation
One coaching style wouldn’t work for everyone
Research showed people wanted different levels of warmth, direction and challenge. Working with behavioural scientists, I translated T-GROW and Motivational Interviewing into adaptable coaching styles.
Design decision
Give people control over how their coach communicates, rather than imposing one fixed personality.
Visual consideration
The experience needed to make different coaching styles understandable and easy to change without adding complexity.
02 | Conversation design
Give people room to explain what’s really happening
Predefined replies offered more control, but restricted how people could express themselves. They also made it difficult to apply behavioural psychology approaches such as the T-GROW model and Motivational Interviewing, which rely on understanding a person’s individual circumstances, motivations and readiness to change.
I moved towards free-text conversation so people could explain their circumstances in their own words, allowing the AI to respond more meaningfully and support behaviour change.
Trade-off
More natural, personalised conversations created greater clinical risk. The AI needed clear boundaries around what it could safely answer.
03 | Clinical safety
Safety should appear when people need it, not interrupt every interaction
Policy required a disclaimer checkbox whenever users mentioned symptoms, medical conditions or asked health-related questions. They had to acknowledge that the coach couldn’t provide medical advice before continuing.
User research showed this felt cold and interrupted the conversation. Users felt less supported and were less comfortable continuing.
Design decision
I replaced the checkbox with contextual warnings and escalation messages, triggered by predefined clinical risk indicators. This communicated safety boundaries without unnecessarily interrupting the conversation.
Validation
I used user research and A/B testing to compare both approaches. Users preferred contextual guidance because it felt more natural and supportive, while keeping clinical boundaries clear.
04 | Setting expectations
Make the coach’s limitations clear before the conversation begins
Testing also revealed that some users didn’t understand what the AI could and couldn’t do. I introduced a Before You Start screen to explain the coach’s purpose and capabilities upfront. This helped set expectations before users began interacting with it.
05 | Inclusive design
Inclusion was a core design consideration
I designed for WCAG AA accessibility standards, using the EHIA to identify barriers and inform inclusive design decisions across the experience.
Impact
Establishing the direction for a safer, more inclusive AI coach
39 million people: An AI-powered preventative Health Coach designed for NHS-wide scale.
Under six months: Progressed from discovery into Alpha.
60+ pain points: Synthesised into clear product priorities.
Three strategic priorities: Behaviour change, inclusion and clinical safety.
Results after launch
<6 months
Discovery to Alpha
60+
Pain points synthesised
393K
Modelled reach at 1% adoption
Colleague feedback
Feedback from colleagues I worked with on this project.
“
Members of the team regularly consult Isabella to get her steer on the UCD approach of the project and her vision for the product we are building. I have personally consulted her for direction on how to complete certain tasks and she has shown leadership skills by owning tasks and artifacts.
“
Isabella has made a significant impact on our AI Health Coach team, both in terms of results delivered and the speed with which she works. She led the development of our user pains and user needs database, analysing hundreds of pages of research and synthesising them into 60 clear pain points.
