Surface · 01
Companion
A glanceable status card answers ‘am I okay right now?’ with colour, plain-English verdict, and one suggested action. The chart is one tap below.
Case Study · 2024iOS · Health14 weeks
GlucoSense: turning 1,440 glucose readings a day into one calm answer.
A personal CGM companion for iPhone, redesigned from a numbers-only data app into a decision-ready coach people actually open before breakfast.
+38%
D7 retention
4.8★
App Store
−52s
Time to insight
01Project Overview
A two-quarter redesign sprint for a category-defining CGM app.
GlucoSense is a free iPhone companion for continuous glucose monitor (CGM) wearers. The team had a beloved data engine and a loyal niche audience, but first-time users were churning at 62% in week one.
Our brief: rebuild the first-run experience and core dashboard so a non-clinical user could feel oriented in under sixty seconds.
My Role
Lead Product Designer
End-to-end: research, IA, interaction design, prototyping, hand-off.
Duration
14 weeks
Mar – Jun 2024
Platform
iOS 17+
SwiftUI · HealthKit · Dexcom & Libre
Team & Collaborators
1 Designer, 2 iOS Eng, 1 Clinical Advisor, 1 PM/Founder
Methods Used
Diary StudiesInterviewsCo-designUsabilityAnalytics
Tools
Figma · FigJam · Maze · Lookback · Mixpanel · Notion
CASE_STUDY_REF: GS-2024-V3
02Problem Statement
A data firehose, asked to do the work of a coach.
CGMs stream a glucose reading every minute — 1,440 data points a day. GlucoSense v1 surfaced all of them in dense line charts borrowed from hospital dashboards. Power users loved it. Everyone else asked the same three questions in support tickets: ‘Is this number good? What did I do wrong? What do I do next?’
“I open it, I see a squiggle, I close it. I don't know what I'm supposed to feel.”
01
Numeracy gap
68% of new users could not name their target range without checking settings.
02
Anxiety spikes
Spike alerts arrived without context, training users to silence notifications by day three.
03
No next step
The dashboard answered ‘what’ but never ‘so what’ — users left without a decision.
03Research & Discovery
Twelve people, two weeks, one recurring sentence.
A mixed-methods discovery: a 14-day diary study with 12 CGM wearers (Type 1, Type 2, metabolic-curious), 8 follow-up depth interviews, a teardown of 6 competitors, and a quant pass on 90 days of in-product Mixpanel data.
Primary Persona
Maria, 47 — “The curious newcomer”
Diagnosed pre-diabetic six months ago. Wears her first CGM. Wants to understand her body without becoming her own endocrinologist.
Goal
Stay in range
Frustration
Medical jargon
Trigger
Morning coffee
Win
A calm green day
- 01
The morning glance
81% of sessions started within 4 minutes of waking. Users wanted one number, one verdict, one suggestion — not a chart.
- 02
Meals as the unit of meaning
People didn't think in glucose, they thought in meals. Charts only made sense when overlaid with food.
- 03
Range > number
‘Time in range’ outperformed any single reading as a confidence signal — yet it was buried three taps deep.
- 04
Alerts as punishment
Every interviewee had muted spike alerts. They read them as judgment, not guidance.
Competitive landscape
| Product | Strength | Weakness | Opening |
|---|---|---|---|
| Levels | Editorial polish, scoring system | $199/mo, food-logging heavy | Sensorless onboarding |
| Nutrisense | Human coach in-app | Coach gating, slow loops | Self-serve coaching |
| Veri | Clean Apple-like UI | Number-first, no narrative | Story over score |
| Dexcom (default) | Clinical accuracy | Built for clinicians, not patients | Translate the data |
04Design Process
Sketch fast, test sooner, kill darlings cheaply.
We worked in two-week loops. Each loop opened with a divergent FigJam session and closed with five unmoderated Maze tests. We shipped four prototypes, killed two outright, and merged the survivors into v2.
- 01
Reframe the question
Replaced ‘show all data’ with ‘answer one question per session.’ Built a decision tree of the five questions users actually asked.
- 02
Verdict-first dashboard
Prototyped a single Status card — colour, verdict, suggestion — backed by a collapsible chart. ‘You're steady’ beat ‘In range’ by 31%.
- 03
Meals before metrics
Moved meal logging from a buried tab to a primary tile. Pre-filled with Apple Health and recent meals to keep entry under 6 seconds.
- 04
Compassionate alerts
Rewrote every system notification with our clinical advisor. Removed the word ‘spike.’ Added a ‘What might have caused this?’ trail.
The hardest call
Cutting the historical chart from the home screen.
Power users revolted in beta. We compromised: kept the verdict-first card by default, added a one-tap ‘Show the data’ that remembered the preference per user. Power-user NPS recovered within two weeks.
What I'd do differently
Recruit Type 1 users earlier.
Our first round skewed metabolic-curious. A late-stage T1 panel surfaced edge cases (post-bolus drops) that forced a v2.1 patch. Should have stratified the panel from day one.
05The Solution
One screen. One verdict. One next step.
Surface · 02
Meals
Photo, voice, or recent-meal entry in under six seconds. Each meal is auto-correlated with the glucose response — cause and effect, not just numbers.
Surface · 03
Insights
A weekly digest written like a friend: ‘Tuesdays are your steadiest days. Late dinners after 9pm cost you 14% of your range.’
Designed for clarity.
Cause and effect, not just numbers.
Design System
Teal
Mint
Coral
Cream
- · Fraunces display + Inter UI
- · 4pt spatial grid, 14pt min body
- · Haptic taxonomy: soft / firm / celebratory
- · Dark mode shipped at launch
Prototype
Final hi-fi Figma prototype with 38 connected frames, used for the last two rounds of moderated testing and engineering hand-off.
06Impact & Results
Measured 90 days after launch.
+38%
D7 retention
vs. v1 baseline
4.8★
App Store rating
104 reviews, US
−52s
Time to first insight
from cold open
2.4×
Meal logs per user
weekly average
“Finally a glucose app that talks to me like a human. The morning card is the only health thing I open every day.”
“I stopped being scared of the alerts. I actually read them now.”
Lessons learned
- 01
A verdict is a feature.
The single highest-leverage decision was replacing a number with a sentence. It changed every downstream interaction.
- 02
Onboarding is a research instrument.
The first-run flow became our highest-signal place to test new metaphors — cheaper than usability sessions, faster than interviews.
- 03
Clinical voice ≠ clinical content.
Patients want a human tone with clinical rigor underneath. Our advisor reviewed copy weekly to keep both intact.
Next up
My Sequent Please — Cardiology Knowledge