Chan

About

Work

Product Discovery ‧ AI Strategy ‧ Health Tech

Exploring AI as a solution for long-term care communication

Improving conversion through continuous redesign and iteration

ROLE

Product Designer in a team of 7 people

SCOPE

Strategy, Research, UX/UI

PLATFORM

Mobile app

TIMELINE

4-day workshop, Mar-Apr 2026

What the project expected to explore.

In collaboration with the Head of the Digital Long-Term Care Department at Mackay Junior College, this project looked into the opportunities to improve communication between home care providers and family caregivers.

We also adopted AI-assisted workflow to speed up design process and explore the pros and cons of AI application on the customer-facing health-tech product.

Various demends boosted in Taiwan’s long-term care ecosystem

Taiwan's government projects that the number of people requiring long-term care will increase from 922,948 in 2026 to 1,295,969 by 2035, representing a nearly 40% increase over the next 10 years.

However, the long-term care workforce is already struggling to keep pace with this growing demand. Care workers are increasingly responsible for multiple cases, while case managers face mounting workloads and burnout. As a result, there is an urgent need for digital tools that can improve communication and care coordination.

Defining project scope and limitations

A start from selecting the most feasible MVP focus

The long-term care ecosystem spans healthcare providers, residential care facilities, community care services, home care services and family caregivers. Among these sectors, home care services and family caregivers were identified as the most suitable context for a MVP due to their user needs, business potential, adoption feasibility and opportunities for rapid validation.

Examine the communication flows in the ecosystem

After gaining an understanding of Taiwan's long-term care policies and exploring potential product opportunities, we conducted in-depth interviews with nine family caregivers across eight caregiving families.

These families represented twelve care recipients with a wide range of care needs, including mild physical disabilities, dementia, Parkinson's disease, mental illness, cancer, stroke, and chronic conditions. Two of the participants were sisters from the same family, allowing us to compare how different family members experienced and managed information needs within the same caregiving context.

Interviewing family caregivers

Established design systems, usability principles and examples from comparable products served as external reference points. Material Design was not treated as a framework to follow uncritically. Instead, its recognised patterns helped explain why certain interactions had become familiar to users and how similar problems were being addressed elsewhere.

Combined with competitor examples, available behavioural data and concrete prototypes, these references gave stakeholders a clearer basis for evaluating proposals and enabled engineers to work from more clearly defined designs before implementation.

Based on the interviews, four recurring challenges emerged:

Multi-stakeholder communication (mentioned by nearly all participants)
Information fragmentation (6 of 9 participants)
Medication and condition tracking (4 of 9 participants)
Navigating long-term care services (4 of 9 participants)

All these challenges created information anxiety and increased the decision-making burden for family caregivers, regardless of the care setting or medical condition.

Defining Product Strategy

After uncovering key insights from family caregivers and care professionals, we began defining the target users for the care communication platform.

The focus of target users is on family caregivers supporting people with mild to moderate care needs under Taiwan's long-term care system. Compared with caregivers of highly dependent patients, this group represents a larger user base, is generally more receptive to digital tools and makes it easier to validate key product hypotheses through an MVP.

The product direction thus centred on one core goal:

Reduce caregivers' information anxiety without introducing additional cognitive load or operational risks.

From the main goal, 4 design principles were developed to guide feature ideation:

Enable seamless care handovers so every caregiver stays informed
Organise care information into clear, prioritised and trustworthy records
Help caregivers focus on what matters most to support better care decisions
Reduce unnecessary cognitive load throughout the caregiving journey

These principles also became the foundation of how I collaborated with AI throughout the project. Rather than generating ideas from scratch, they provided the primary context for prompting AI, ensuring that exploration remained aligned with user needs and product goals.

Ideating and prototyping with AI

During the ideation phase, the proposed features were grouped into four functional areas, ensuring each design decision aligned with the project's core objectives.

01/ Information Input

Support both structured medical instructions and everyday caregiving activities.
To reduce documentation effort, caregivers could quickly capture care records through voice input without relying on extensive typing.

02/ Information Output

Leverage AI to organise, categorise and summarise care records, enabling caregivers to quickly understand a patient's condition.
Key physiological indicators required by doctors and therapists were also consolidated into dedicated health summaries for easier review.
In urgent situations, the system provides contextual guidance to support immediate decision-making.

03/ Care Recipient Management

To minimise cognitive load and reduce operational errors, the interface presents detailed information for only one care recipient at a time outside the home screen.
Cross-patient information is limited to high-level summaries and alerts on the dashboard.
All record creation and editing clearly indicates the active care recipient.

04/ Information Accessibility

Considering that many caregivers are middle-aged or older adults, the interface features larger touch targets and adjustable text sizes to improve readability and ease of interaction.

View prototype →

Expert Review & Validation

The first usability test focused on AI-generated daily summaries and alerts as the core product experience. Two professional home care workers in their 50s participated in a one-hour usability test and interview, leading to 4 key findings:

Voice input significantly reduced documentation effort, receiving a rating of 10/10.
Categorised records did reduce the burden to dogest information.
Trust in AI-generated summaries remained limited. Participants viewed them as reference information rather than a reliable source, with an average rating of 5.5/10.
In emergency situations, such as falls or loss of consciousness, caregivers would immediately contact emergency medical services instead of relying on the app.

Redefining AI's Role through Product Decisions

The findings led to another HMW:

How might we define AI's role in a way that builds user trust while maintaining the credibility of AI-generated summaries?

Ensuring trustworthy information input

After voice transcription, caregivers review and confirm both the recorded content and its categorisation before submission, ensuring the accuracy of the original care record. The original audio and transcripted record stay along with input records as references.

Redefining AI's responsibilities

AI only processes user-confirmed information and automatically generates daily summaries after each new record is added.

To maintain a clear product boundary, AI is positioned as a documentation and communication assistant rather than a clinical decision-making tool. It does not provide recommendations or make judgments in situations requiring medical expertise, such as emergencies.

Improving Information Visibility

Physiological data is visualised to help caregivers understand long-term trends at a glance. Despite the fact that these records were originally intended for healthcare professionals, presenting them visually also makes meaningful changes easier for family caregivers to understand.

Improving transparency

To help users better understand AI's capabilities and limitations, a dedicated page explaining the role of AI was added to the home screen.

Second Validation

The second round of usability testing involved 6 family caregivers. Compared with the first prototype, the refined version received higher ratings across all evaluation criteria.

End-to-end experience from voice recording to AI summaries and data visualisation (10/10)
Willingness to recommend and willingness to pay (8/10)

The findings suggest that the redesigned experience effectively reduced information anxiety while improving confidence in daily care communication.

Instead of ending with a finished product, this project concluded with a validated product direction, a refined understanding of AI's role in healthcare and a stronger product foundation for future long-term care innovation.

What I learnt from the project

AI's role evolves with product understanding.

This project introduced me to the concept of AI governance, which starts with product decisions. Before deciding how AI should behave, I first had to define where AI should intervene, where it shouldn't and what role it should play within the product. Without careful judgment and a clear definition of AI's role, AI can easily create uncertainty, reduce trust or even mislead users in healthcare scenarios.

AI accelerates execution. Designers define direction.

In the process, AI worked as a collaborative design partner from organising research and prototyping to synthesising validation findings. While AI significantly improved execution speed, the most valuable design decisions still relied on human judgment: framing the problem, balancing stakeholder needs, making product trade-offs, and defining the product vision. The product thinking still determined product strategy and what was worth building in the end.

AI should not be treated as a default solution but as a product hypothesis to be validated.

This project changed the way I evaluate AI in product design. Rather than asking "How can we use AI?", I now start with "Does AI create meaningful value in this context?" Only after understanding user needs, business goals, costs and potential trade-offs can AI become a product decision. AI should be introduced because it solves a meaningful problem, not simply because it is the latest technology.