COFE + Robot App
UIUX
Product Design
AI Workflow
How do you make a robot feel less like a machine and more like your regular barista?
COFE+ explores how a coffee robot can be fast and reliable, and still feel personal. We designed a mobile ordering flow for people short on time.
Two years later I came back to it. I audited our final screens as if they were someone else’s. I checked them against WCAG 2.2 AA and redesigned the moments that broke. I worked with AI as a partner and kept every decision.
My Role
UI/UX Designer
Timeline
0 - 1: Mar 2024 - Jul 2024
Audit: Oct 2026 - Nov 2026
Industry
product
outcome
Faster Checkout Time
Higher Order Completion Rate
2024 · company test report
Pick-up code contrast
Text layers pass WCAG AA
2026 · audit and redesign
PART 01 · 2024 · Building from zero
Problem
Time-sensitive users struggled to get speed without losing a sense of being cared for.
Market RESEARCH
Every product optimizes speed. Few earn trust.
INsights
Speed builds trust. Complexity breaks it.
Speed matters more than exploration, commuters prefer predictable ordering flows over deep customization.
Decision fatigue slows down ordering, because too many options create unnecessary friction.
Automation needs emotional feedback, and robot-driven service risks feeling cold without expressive interaction cues.
Design Opportunity: A system that makes fast decisions feel effortless, but also make user feel like being taken care of while experiencing robotic efficiency.
design
How can we make ordering fast, without making it feel like just a machine?
We designed two layers that work together. A simpler flow that removes decision fatigue, and a character that gives the robot a personality.
Feature 01
The ordering flow was simplified into three stages:
Explore → Customize → Confirm.
Each step shows only the options that matter. Checkout puts review, payment and pickup in one step.
Feature 02
A fully automated flow can feel anonymous. So we designed a mascot as a team. It introduces itself and reacts at each step, adding warmth without adding steps.
Results
It worked. The company measured it.
The company ran its own tests and shared the report with us in a meeting. Ordering took 3 steps. Checkout was 25% faster. The order completion rate was 23% higher. The project was later nominated for the UX Design Award and won MUSE Gold.
Reducing options was the harder design problem than adding them.
Most commuters ordered the same drink every time. The real work was deciding what to hide by default.
Predictive ordering is still unsolved.
One user said "I just want to tap once and have my coffee ready before I even stop walking." Our flow still asked people to reselect their usual drink every time.
PART 02 · 2026 · Auditing
THE TURN
Award-winning screens can still fail at the machine.
Two years later I audited our final screens against the product's own promise. A robot should feel as easy as a barista. It broke at pickup, allergies, price, reorder and cancel.
Redesign
Same look. Fixed five broken moments
I kept our 2024 look, the mascot and the tracker. That was the strongest part of the work. Every fix below traces back to one finding.
Redesign 01
Redesign 02
Redesign 03
Redesign 04
Redesign 05
DESIGN SYSTEM
Teal is a surface, never text.
Our brand teal fails as text. So every colour became a role token with a measured contrast ratio, and type went from 11 sizes to 5.
AI Workflow
AI built and reviewed. I made the decisions.
I wrote a spec for each screen. AI built it in Figma with my design system and ran review passes. I decided what to change, then tested with people.
Where I went against the AI
Kept our 2024 look. The layout was never the problem.
Rebuilt the allergy path. I restructured the milk options myself.
Fixed what the review found. My redesign had put the danger button where the thumb confirms.
Testing
3 people. 4 tasks. All completed.
One person had never ordered coffee through an app. All 3 users ordered, set an allergy list, picked up and canceled. Each one stated the price before paying.
"It seems great to have the allergy feature."
REFLECTION
Auditing my own work was harder than auditing someone else's.
It is easy to defend a decision you remember making. Measuring before and after made my claims defensible.
AI made me faster, and more careful about what I claim.
It built what I could describe. The judgment stayed with me.
