Creating a new design system to improve accessibility, efficiency and consistency

Definition Health had clear business objectives driving the need for a design system. This was both in commercial terms, such as meeting RFP requirements and NHS standards for integration, and for internal efficiency and user experience where after years of evolving the product, many inconsistencies had crept in.

These were the challenges that a design system could address:

• Existing UI not meeting NHS standards for integration
• Unnecessary barriers to inclusion
• Lack of clarity around work needed to improve accessibility
• Inefficiency in engineering stemming from accommodating different design choices
• Inability to know which patterns and components were the ‘right’ ones to use when designing changes to the UI
• No assets in Figma so everything needed to be created from scratch
• Icons from different sets, no visual consistency

With good buy-in from senior leadership, a clear commercial benefit, and a moral imperative to improve accessibility, I didn’t need to persuade anyone that a design system would be a good thing. But I did need to understand “the gap”. Where were we now and where did we want to get to?

Step 1: Project Kickoff

To align key stakeholders with the goals of the project, I used the Design System Canvas to understand and communicate scope, constraints and expectations.

Using the design system maturity scale was helpful for us to think about our starting point and track our progress so we could measure our level of success.

I also ran a Lean Coffee discussion with the engineering team to understand what design system related concerns and aspirations were on their minds.

To promote the design system initiative more widely, I presented a design systems 101 to all staff in the company so that everyone understood the importance and expected impact of this work. I wanted my colleagues to feel excited about the future of product design at Definition Health.

As the work progresed, I regularly published blog posts and updates on Slack explaining the rationale for changing the UI.

Step 2: UI Audit

Next, I completed an audit of the current LifeBox UI, capturing every distinct piece of design as screenshots that I annotated in Figma. This helped to understand deeply what the needs across the product were and how rationalising with re-usable components could help. It was also a really useful artifact for helping to estimate and scope work relating to integrating new components into the existing UI.

Step 3: Scoping and tech decisions

How should we store and manage design tokens?

We decided they were best under the control of designers so tokens were stored in a git repo and managed via Figma with the Tokens Studio Pro plugin. Tokens Studio allowed me to generate shared styles from the token set to apply across components in Figma. The deployment pipeline translates these tokens with style dictionary to turn them into usable CSS variables in the web app.

How should we name design tokens?

After doing a lot of research on this topic and studying some well established design systems, I settled on the following structure as I thought it would be able to scale well for our needs. This gave us a consistent way to approach naming tokens to speed up decision making and make naming more predictable.

system.concept.property.variant.state

e.g.

lbds.color.background.primary.default


Inspiration for this came from Nathan Curtis and his excellent blog post on naming tokens in design systems.

Should we use an off the shelf library like Google Material?

This was considered, but ultimately I felt that we would end up needing to extend it with custom components and customise it heavily to suit our needs. With this in mind, it may have ended up taking more work to maintain it than something custom built. A competitor had already built their UI with material so it would have been difficult to seem distinct from them in terms of our visual design and interaction. With the skills and backing we had in the business, going off the shelf was less advantageous for us.

‍

Step 4: Definition and delivery

I knew that establishing the foundational elements quickly would allow us to make rapid improvements and gain buy-in for future changes. Instead of a full scale UI redesign, I focused on evolving the existing styles, refreshing both the visuals and code to keep aligned with our accessibility goals. My aim was to maximise the footprint of new components in the UI to subtly evolve the interface as we worked through the backlog.

🌈 Colours

I started out by defining colours. Since we already had a well established UI, I focused on establishing semantic colours like danger, warning, info and adding in a few accents that could be used in badges and charts. I took the brand colours and used palette generation tools to make sure the new hues were harmonious as well as capable of providing good contrast to meet accessibility guidelines.

I built out the colour palette in Figma, generating tokens and shared styles to use across projects.

🖋️ Spacing and typography

Spacing and typography had to come next, as buttons depend on it for their structure and labels, giving us a huge increase in new component footprint once integrated.

Spacing

Spacing was based on a 4px baseline grid and along with basic values (4, 8, 12, 16 etc) I set up semantic configurations that would handle all 4 values in one. Thanks again Nathan Curtis for your article about applying space with intent.

Typography

The typography component was designed to be highly flexible, allowing authors to define which HTML element it renders. This separation of markup from visual presentation is particularly useful for maintaining proper heading hierarchy in the code (critical for screen reader users) while still enabling unique visual treatments when needed.

By setting up the component with variants for size, weight, color, and type (e.g., heading, body text), it became easy to use in Figma with tight control, while also providing developers with enough flexibility to avoid workarounds in edge cases.

While this approach works exceptionally well in code, it hasn’t translated as smoothly into the day-to-day realities of Figma. Over time I have encountered challenges when using variables, especially as the component becomes deeply nested in other components. Switching variants with modes and variables fails to preserve overridden text values, undermining the benefits of these advanced features. Sadly it’s a known frustration with Figma and there’s no sign that things will improve soon. In hindsight, I’d consider relying more on shared styles next time to keep things simple.

🔢 Buttons

One of the most critical elements of any web UI, buttons were sorely in need of improvement.

The primary and secondary buttons not only didn’t meet AA level contrast requirements but they were also too similar in style to visually communicate the hierarachy of actions. If all colour was taken out of the UI, they would look almost indistinguishable.

To ensure these were accessible and distinct enough from one another, I decided to make the secondary buttons an outline style. I used our new accessible colours to update the button backgrounds. Props were added for icons to be placed left or right as well as a variant for icon only.

‍

📝 Documentation

I wrote accompanying documentation in Figma to explain the available properties on each component.

I also blogged about significant updates so everyone in the company could understand why changes were being made. This kept everyone interested in the project and was a good way to demonstrate progress and celebrate milestones.

Within just a few sprints, the new design system started to deliver heaps of value for Definition Health. The UI was becoming more accessible for patients and hospital staff, developers had reliable components to work with and I had a solid UI kit built in Figma, enabling me to craft new designs at speed.

This great foundation has powered the next wave of design and dev work to keep improving accessibility in the UI and has enabled the overhaul of the most problematic parts of the interface.