About
GoodGang Labs is a start-up building an artificial-intelligence powered checkout kiosk.
The kiosk provides tailored realtime assistances ranging from customer's questions about products, stores, and payments.
Contributions
Product Design, UIUX Design, Visual Direction, Specialized UIUX
Tools
Figma, Adobe Suite, Pro-Create
Duration
6 Months, 2024—2025
Summary
As the sole product designer, I worked with GoodGang-Labs to turn a set of loosely defined ideas into a cohesive, usable product interface and experience.
In 6 months, the team and I turned a concept into a real usable product, shipped a total of 3 iterations in Korea, and secured 3 retail partners.
Increased kiosk's verbal operation comprehension from 10% to 80%, 3X transaction completion rates, and cut total checkout time by 25%.
We used Korea as a test bed not because it was convenient, but because it’s one of the most demanding places to prove a commerce experience. People are highly accustomed to fast, efficient service, they already have strong expectations for kiosks, and they won’t tolerate friction for novelty alone.
GG-Labs AI-kiosk & offline flagship store located in Seoul, Sinsa
Building an MVP
We built an MVP, observed how people used it, and iterated from real insights. The MVP served as an anchor point, revealing core pain-points that allowed for a comprehensive diagnosis.
With the ambiguous nature of our product, our team decided to aggressively build, ship and test to quickly validate feasibility of concepts. We conducted guerrilla testings with potential clients, partnerships, groups of college students, available walk in customers, and friends.
The MVP QA revealed that the kiosk had a steep learning curve, but lacked the crucial information and guidance users needed for success.
pain-point 01
When to speak wasn't clear.
The turn-taking conversation system wasn't self explanatory enough, so users spoke over the AI-agent or hesitated at wrong moments.
Only a small number of users had successful ordering experiences without staff guidance, majority of users suffered from cascading failures.
(ex: *Missed STT > Confusion > Retries > Repeat)
pain-point 02
How to operate wasn't clear.
Sessions had to be started through a touch-screen payment device on the side, and the kiosk's voice-only operation wasn't self evident.
Users always experienced initial friction trying to find out the kiosk's operation method, which resulted in initial distrust and hesitation.
pain-point 03
What to say wasn't clear.
Phrasing requirements for successful orders weren't communicated, so users spoke in fragments or varied styles.
In order for the LLM/SLM AI-agent to accurately interpret user's requests, the orders had to fulfill specific requirements such as quantity, color, name, size.
(ex: *I want to order "One"—"Silver"—"Medium Size"—"Tumbler")
pain-point 04
Mismatch in mental models and expectations.
Users Expected Efficient Transactional Interactions Rather than a Back & Forth Interaction with the AI-Agent.
Users processed dialogues before the AI-agent even finished speaking, and often felt the dialogues were too long. Users were focused on getting to the payment as soon as possible rather than conversations.
How might we
With the diagnosis, what we had to solve became clear.
"How might we make a voice-operated checkout kiosk instantly learnable, so users know when-to-speak, what-to-say, and how-to-operate the kiosk all on their own?"
I studied how people shop in-person, extracting insights to refine our AI-kiosk user-flow.
insight 01
Two types of store checkout experiences; Streamlined-Purchases vs Consultative-Purchases.
In real stores, most customers arrived with one of two intents. Some were already decided and wanted a streamlined purchase, while others were undecided and need consultation.
(ex: *convenience stores vs shoe stores)
insight 02
Helpful staffs led customers through the checkout process with clear and easy questions.
Efficient staffs asked customers specific questions rather than open-ended questions, making it easy for customers to respond.
(ex: *“What do you need?” vs “What can I help you purchase?")
I studied voice AI products and conversational interfaces from leading AI companies.
insight 01
Voice-experiences progress the user-flow using dialogues instead of screens, shifting the UI’s purpose into bringing status clarity.
Leading voice-experiences kept a minimal UI design because they don’t have “pages” to navigate, the user-flow is replaced by the conversation.
So the visuals aren’t for navigating, they’re for clarity; showing users what’s happening, what was heard, and what to do next.
insight 02
Voice-experiences optimized converting conversations into solving needs, not having a conversation.
In voice-interactions especially, users feel friction when the conversation feels like a long back-and-forth.
Users have a clear need, and want the system to understand quickly, respond briefly, and move them to completion.
1st product iteration
Based on prior insights, we shipped version-1, a single-screen interface that prioritized system and information clarity throughout the kiosk’s conversation-driven flow.
We tested this version of the product through partnerning with Mamison in the form of a pop-up retail. This was our official launch of the store & kiosk, and I observed and documented over 1700+ live sessions and collected 175+ exit surveys from users, while collecting live feedback for the next iterations.
core-interaction 01
Standby screen, indicator to show how to initiate a session.
Due to technical limitations, the order session had to be initiated through a smaller adjacent payment screen on the right side of the kiosk.
core-interaction 02
Turn-taking in action, when to wait vs when to speak your order.
A short demo of the conversation-interaction, showcasing how the kiosk switches between AI speaking and user-input
design solution 01
A single-screen layout to keep users focused.
The MVP’s separate catalog screen created an attention split, which resulted in missed speaking cues, snow-balling into cascading failures. In this iteration, I eliminated the need to scan multiple surfaces by designing a single screen layout that includes everything the user needs for ordering.
design solution 02
Turn-indicators that make "Wait" vs "Speak" clear.
I visually differentiated the user's turn and the ai-agent's turn, while also using clear text labels to remove any ambiguity.
design solution 03
A dual purpose catalog, see what you can order, and finalize what you ordered.
I integrated a two-purpose catalog so users can reference exact product names while speaking, then confidently finalize their purchase through the order summary screen
design solution 04
The conversation-driven end-to-end user flow.
Voice-UIs don't navigate through the user flow like most traditional interfaces, instead the end-to-end flow is carried by the conversation.
1st iteration's launch & user testing
Version-1 made the kiosk learnable, but testing exposed that the problem evolved into optimizing the screen's real-time UI interaction.
pain-point 01
The turn-indicator's status changes were too subtle to catch during active conversations.
State changes didn’t stand out quickly enough, so users missed their speaking window even when they understood the idea of turn-taking.
pain-point 02
The new UI elements looked too tappable, which reinforced touch-based interactions.
The catalog and turn area read as buttons, so users defaulted to tapping—despite the kiosk being voice-led.
pain-point 03
The catalog functioned as a menu, not a prompt guide, it lacked clear instructions to help users phrase their orders correctly.
Users could see product names, but the UI didn’t help them translate that information into successful order sentences.
2nd product iteration
For version-2, we refined the kiosk into a more guided ordering experience, by strengthening its visual clarity and adding guidance features to carry the user from start to finish.
In our second collaboration, we partnered with 78-Under in the form of a long-term retail to launch their next product series at our store. We had reworked a significant chunk of our product ranging from the visual interface & user-flows to major technical architectures, yielding improved usability results.
core-interaction 01
78-Under collab standby screen, session initiation indicator
Due to technical limitations, the order session had to be initiated through a smaller adjacent payment screen on the right side of the kiosk.
core-interaction 02
Addition of the on-boarding user-flow, educating users when to wait vs when to speak.
A short demo of the conversation-interaction, showcasing how the kiosk switches between AI speaking and user-input
design change 01
Polished screen hierarchy with additional guidance features.
The MVP’s separate catalog screen created an attention split, which resulted in missed speaking cues, snow-balling into cascading failures.
In this iteration, I eliminated the need to scan multiple surfaces by designing a single screen layout that includes everything the user needs for ordering.
design change 02
Turn indicators with clear color differentiation and minimal icon designs to reduce comprehension speed between "Wait" vs "Speak"
I visually differentiated the user's turn and the ai-agent's turn, while also using clear text labels to remove any ambiguity
design change 03
A multipurpose catalog, expanding into various guidance screens; On-boarding, order prompt examples, visualized order selections, and checkout.
I integrated a two-purpose catalog so users can reference exact product names while speaking, then confidently finalize their purchase through the order summary screen
design change 04
The conversation-driven end-to-end user flow; expansion to consultative responses.
Voice-UIs don't navigate through the user flow like most traditional interfaces, instead the end-to-end flow is carried by the conversation.
Handoff
Retrospect and takeaways.
take away 01
Experiencing the 0 to 1.
Lorem ipsum odor amet, consectetuer adipiscing elit. Proin tempus enim adipiscing ut netus id. Libero tortor suscipit vel ridiculus tellus molestie efficitur. Tristique sagittis sodales sollicitudin amet. Pulvinar mi sollicitudin nec sagittis. Mollis tempor quam finibus montes. Condimentum nulla pellentesque neque ornare.
take away 02
Importance of consistent high-level communication & alignment.
Lorem ipsum odor amet, consectetuer adipiscing elit. Proin tempus enim adipiscing ut netus id. Libero tortor suscipit vel ridiculus tellus molestie efficitur. Tristique sagittis sodales sollicitudin amet. Pulvinar mi sollicitudin nec sagittis. Mollis tempor quam finibus montes. Condimentum nulla pellentesque neque ornare.
This is a part of the many things I worked on, I’m happy to chat more about my process/ work over a call.
Role
Product Designer, Specialized UIUX
Year
6 months, 2024 — 2025
Work
Product Design, Strategy, Research
Next
Arctika
>