Healthdirect Symptom Checker

Making AI health triage accessible and actionable

The challenge

Healthdirect's Symptom Checker had a completion rate of just 45% — users were dropping off before getting the care guidance they needed.

The challenge was to redesign the experience around a new AI triage engine (Infermedica) while navigating Australia's strict clinical standards, legal constraints, and Healthdirect's non-diagnostic role.

Success Metrics

• Completion rate — increase from 45% baseline toward the 60–70% industry benchmark for complex digital health tools

• Usability — achieve 78%+ task success rate and resolve all severity 3–4 issues before launch

• Health literacy — users can understand medical terms and instructions without assistance

• Accessibility — meet WCAG 2.1 AA compliance

• User confidence — users feel guided and supported throughout the assessment

• Drop-off rate — no single step loses more than 20% of users

• Team efficiency — improve cross-functional delivery to reduce project delays

Accessibility

• Conducted heuristic reviews to identify usability issues before resource-intensive testing

• Tested with diverse users including people with disability and tech-wary individuals, iterating on UX and microcopy each round

• Adhered to WCAG standards for screen reader compatibility and keyboard navigation

• Applied soothing colour schemes sensitive to user emotions

• Designed AI outputs to support decision-making without feeling directive

Keeping users motivated through a long assessment

The usability testing revealed that the encouragement screen functioned as a good step in adding a human voice and propelling users to progress.

The steppers and clear instructions helped craft a supportive user experience and ease user anxiety.

Actionable care advice

• Surfaced key decision-making info — wait times, cost, and appointment requirements — to help users compare clinics

• Numbered clinics in order of recommendation

• Bolded key info for easy scanning

• Highlighted essentials on service cards — "No fee", "Bulk Billing", distance to nearest clinic

Interaction design

Below is a prototype with implementation of motion design to improve user engagement. This idea was pitched to the business to potentially hire developer who specialises in developing interactive animations.

PHASE 2

Identifying Gaps and Opportunities Across a Complex Service System

Following Phase 1 delivery, the Symptom Checker redesign became an input into a broader, time-boxed CX and service design initiative focused on understanding where gaps and opportunities existed across Healthdirect’s service ecosystem.

Rather than optimising a single product, Phase 2 stepped back to examine how people experienced digital services, GP advice, and nurse helplines together, particularly at moments of uncertainty when users were trying to understand what help was available and what to do next.

Our purposes

• Identify system-level gaps between consumer needs, service intent, and actual behaviour

• Surface future-state opportunities across services and channels

• Use research and prototyping to inform strategic decisions and funding discussions

• Explore how AI-assisted support could responsibly enhance human-led services

What I worked on

Working alongside CX leadership, I contributed by:

• Conducting consumer interviews across prioritised cohorts

• Synthesising qualitative insights to surface recurring patterns and moments of uncertainty

• Translating insights into journey maps to make system gaps visible

• Prototyping service and AI-assisted concepts to provoke discussion in user interviews

• Sharing Phase 1 findings to maintain evidence continuity

• Contributing to research presentations and cross-functional workshops

This helped teams move from incremental fixes toward a shared view of where investment could deliver the greatest value.

Designing AI for health decisions under stress

Australians turn to digital tools when anxious or uncertain — moments where AI must build trust, not add confusion.

• Digital confidence and self-care ability are unevenly distributed

• Users accept AI, but want human oversight

• More information doesn't mean more confidence under stress

How this shaped our decisions

• Design for what people can cope with, not what the system can do

• Explain why a pathway is recommended, not just what to do

• Avoid diagnostic language to prevent anxiety escalation

• Keep human support visible — self-service can feel like abandonment

• Design for the least-resourced user, not the most confident

From insight to opportunity

Consumer interviews and synthesis revealed a small number of recurring patterns.
These were translated into clear opportunity themes to guide future design and service decisions.