Digital design
Mental Health Screening Platform for Adolescents
Commissioned by a school board to build a mental health screening platform from scratch — I led research, design, testing, and reporting end to end.
Scale
5,000+
students screened across schools
Efficiency
3d → 1d
screening time cut by two thirds
Usability
100%
completed independently, no help needed
Data quality
Reliable
actionable data, no redesign after launch
My role
Lead UX Researcher & Designer
Duration
4 weeks
Participants
10 adolescents, ages 12–18
Client
School board
Due to confidentiality considerations, screenshots and visual materials related to the platform design and testing process cannot be disclosed.
Design a survey adolescents could understand, complete accurately, and respond to honestly.
Wide variation in reading ability (ages 12–18)
Typing and spelling difficulties in Chinese input
Misunderstood question wording
Sensitive topics requiring careful handling
Revealing confusion through behaviour as well as self-report
Because adolescents rarely articulate confusion, direct observation was combined with self-report to capture both behaviour and experience.
Collaborative requirements gathering
Defined scope, users, and outputs with the school board before design began.
Contextual observation
Watched students use the platform in school. Real behaviour revealed what self-report missed.
Informal probing
Asked in the moment if anything was unclear — kept conversational to avoid adding pressure.
Post-session interviews
Followed up after testing to understand hesitation that observation couldn't explain.
Iterative usability testing
Self-tested before involving students, refining the design between every round.
From raw observation to patterns
After each round, notes were analysed independently — grouping themes, surfacing patterns, identifying variation across age groups. Analysis happened between every round, not just at the end.
Organised field notes by session
Structured by participant and task to spot shared moments of confusion.
Identified repeated patterns
Issues across multiple participants were prioritised for redesign.
Noted unique differences
Age-group variation shaped the decision to build two survey versions.
Fed findings into the next iteration
Each round produced a short list of changes to test next.
Four patterns emerged consistently across participants and age groups — each one directly shaped a design decision.
Language ≠ Understanding
Some phrases that appeared clear were interpreted differently by adolescents, leading to inconsistent responses.
Input Method Shapes Response Quality
Typing and spelling difficulties—especially in Chinese input—introduced friction and reduced response reliability.
Comprehension Varies Across Age Groups
Differences in cognitive and linguistic development (ages 12–18) meant that a single survey design could not work equally well for all users.
Behaviour Reveals Hidden Issues
Users rarely reported confusion directly, but hesitation, re-reading, and response patterns revealed where understanding broke down.
Every change came from observed behaviour, not assumption.
Each entry below names the behaviour that prompted the change.
Adapted for Different Age Groups
Before
Single survey for all
One version assumed similar comprehension across ages 12–18.
After
Two age-specific versions
Same constructs, adjusted wording and complexity per age group.
Simplified Language
Before
Abstract phrasing
Wording that appeared clear was interpreted inconsistently.
After
Age-appropriate wording
Rewrote unclear phrasing to match how students actually understood the questions.
Reduced Input Friction
Before
Open-text responses
Younger users struggled with typing and spelling in Chinese input.
After
Selectable options
Replaced open text with selectable answers — mouse only, no typing.
Response Format Selection
Before
Format chosen without evidence
Multiple choice, open text, and sliders considered with no basis for choosing.
After
Multiple choice adopted
Open text caused spelling friction; sliders were difficult on small screens.
Forced Completion
Before
Questions could be skipped
Incomplete responses made screening data unreliable.
After
Validated before submission
Platform blocked submission and returned students to any missed question automatically.
Reduced Bias
Risk
Patterned responses
Question order and structure considered as bias sources from the start.
Change
Structure and randomisation
Order and structure randomised where appropriate to reduce patterned responses.
What I would do differently
-
I would debrief with teachers after each round — they had a vantage point I didn't.
-
I would also document synthesis more formally, so reasoning could be shared with stakeholders throughout.