01 / Frame the problem
PROBLEM
People still had to find the priority and decide what to do after reading the analysis.
MY RESPONSE
I mapped the result and follow-up journey to locate gaps in information order and action priority.
CHANCE AI / CONSUMER AI
An AI answer.
A clearer next step.
Making complex AI responses easier to understand, act on, and continue.
I designed the response structure, contextual follow-up, and multimodal feedback experience. Skin Care brings these decisions together in one continuous journey.
Explore the case
MY CONTRIBUTION
From an open question
to decisions ready for delivery.
Chance is a consumer AI social product. My work connected the response, the next question, and the system states surrounding them.
02 / Structure the response
PROBLEM
Long AI output left people to sort the information themselves.
MY RESPONSE
I organized the answer around current state, priority concerns, and care steps.
03 / Align constraints
PROBLEM
More images and longer answers introduce latency, model variability, and token costs.
MY RESPONSE
I worked with product and engineering to define response scope and a ten-image input limit.
04 / Connect the journey
PROBLEM
More specific questions emerge after a person sees a result.
MY RESPONSE
I connected suggested questions, free text, and additional images to the existing analysis context.
05 / Follow through on delivery
PROBLEM
Selection, generation, and failure all need explicit feedback.
MY RESPONSE
I defined the state coverage and followed implementation and design review. Context-preserving retry remains follow-up work.
THE CHALLENGE
The analysis was ready.
The decision wasn’t.
An image analysis can describe a concern without making the next step clear. In Skin Care, the design question was how to move from reading a result to deciding what to do with it.
The team’s goal was to give people a reason to continue using Chance. I translated that goal into three connected design questions.
Continued use is a goal to validate, not a measured outcome of this case.
- 01 / UNDERSTAND
What matters here?
Bring the priority forward instead of asking people to sort through a long answer.
- 02 / CONTINUE
What can I ask next?
Keep follow-up connected to the result, with room for a person’s own question.
- 03 / STAY IN CONTROL
What is happening?
Make input limits, waiting, and failed uploads visible before people get stuck.
EXPERIENCE MODEL
Connect one analysis
to ongoing care.
PRODUCT ARCHITECTURE
Understand → decide → keep caring.
Functional value
Use uploaded visual context and skin feedback to organize visible redness, congestion, and texture concerns. The goal is to help someone identify what needs attention.
Decision value
Turn abstract analysis into a readable priority and care direction, with explanations before optional product support.
Continuing value · hypothesis
Translate one insight into a daily plan, completion records, and future adjustment. Continued care and repeat use are design goals to validate.
The intended progression is from having advice to having a plan that can be followed and recorded. It is not evidence of improved skin health or retention.
01 / ANSWER STRUCTURE
Organize the answer around
the person’s next decision.
I moved from parallel content categories to a sequence of user questions: understand the state, find the priority, then choose an action.
BEFORE / CONTENT CATEGORIES
Status · Explanation · Advice · Routine · Products
AFTER / DECISION SEQUENCE
What’s happening? → What matters? → What can I do?
THE ACTUAL BEFORE & AFTER
Change the order, not just the styling.
The earlier layout lacked a clear action, sufficient explanation, and continuing status feedback. The new sequence answers five questions: current state, explanation, priority, today’s action, and optional support.
Start with a priority
The summary brings the overall state, top focus, and primary action together.
Explain the observations
Skin Snapshot and Skin Pattern give the analysis a readable structure. Separating visible observations from AI inference remains an important content check.
Make the next action concrete
The routine turns advice into ordered steps. Its completion controls communicate an intended action, not evidence of sustained use.
Keep product support optional
Recommendations follow the explanation and routine, so shopping does not lead the experience.
A clearer answer creates a new question: how can someone go deeper without starting over?
02 / CONTEXTUAL FOLLOW-UP
The next question starts
with the answer they have.
I connected follow-up to the current analysis, with suggested questions for people who need a starting point and free input for those who already know what they want to ask.
REFERENCE STUDY
Two patterns informed the decision.
These are the references documented during the project, rather than a claim about the products’ current interfaces.
ChatGPT / One composer for a complete question
A shared entry for camera, photos, and files; attachments enter the current composer and can be previewed, removed, or supplemented before sending. Images and text belong to one question.
Perplexity / Continue from the answer
The documented pattern moves from sources to the answer and related questions. Contextual suggestions provide a starting point, free input preserves individual intent, and each answer can lead to another question.
NEEDS A STARTING POINT
“What else can I ask?”
Relevant suggested questionsHAS A SPECIFIC QUESTION
“I want to explain my situation.”
Free text + optional images
IMAGE CONTEXT
Add what words
can’t explain.
Images and text form one follow-up. People can select, preview, remove, and supplement images before sending them with a question.
COMPLETE INPUT FLOW
Enter, select, preview, adjust.
Start from the existing result. Open the composer or image entry, take a photo or choose from the library, then select and review attachments. Remove an image or add more before sending text and images together.
- 01Keep the current analysis
- 02Add text and images
- 03Review, then send
03 / LIMITS & STATES
Make the boundary visible.
Make the state specific.
Adding more images also adds constraints. I worked through the selection limit and upload feedback so people can see what is allowed and which step needs attention.
INTERACTION DETAIL
Show the rule while selecting.
The picker displays the selected count. At ten images, further selection is unavailable.
INTERACTION DETAIL
Waiting and failure need different feedback.
Show the status on the relevant attachment. If someone sends before an upload completes, explain that they need to wait.
Recovery boundary
The supplied generation-failure UI shows a retake entry. It does not yet demonstrate a complete context-preserving retry flow. Optional notifications also depend on permission and delivery behavior that still need validation.
AI STATES & RECOVERY
Explain the interruption.
Show a way forward.
Invalid input, long processing, and generation failure are different situations. Each needs an explanation and a next step.
01 / INVALID INPUT
Translate a technical error into a useful instruction.
The earlier message exposed a system variable. The revision names the required image, explains the mismatch, and offers a retake entry.
02 / LONG PROCESSING
Waiting should not require staying.
The design offers an optional notification entry so people can choose to leave while processing continues.
03 / GENERATION FAILURE
Move from starting over to continuing.
The supplied interface currently offers “Snap Again” after failed recognition.
- Explain what happened
- Explain why
- Offer the next action
REFLECTION & VALIDATION
The screen is only one part
of the experience.
This work connected answer structure, follow-up, image input, and system feedback. Each decision needed to hold together across the journey.
MODULARITY & SYSTEM THINKING
Define shared rules before extending to other agents.
Answer structureAction planFollow-upImage inputRecovery
Shared context, input rules, and recovery paths connect these modules. Skin Care is the current application; reuse across other agents is an extension direction.
A complete journey
Input → generation → understanding → action or follow-up. Waiting and failure return paths must be considered alongside success, with model capability, tokens, latency, and cost.
Clear responsibility
AI organizes, explains, and suggests. The system reports state and limits and supports recovery. The user judges, chooses, acts, and decides whether to continue.
Delivery and evidence
The feature entered production, while evidence of long-term user and business effects remains to be collected. Delivery is the starting point for validation and iteration.
WHAT I WOULD DO EARLIER
Align image limits, waiting strategy, and recovery rules before high-fidelity design. Then observe whether people can identify the priority, explain their next step, continue with a relevant question, and return over time.
DESIGN CONTRIBUTION
One continuous AI interaction.
Response hierarchy, contextual questions, multimodal input rules, and feedback states, developed with product and engineering.
This case presents the design decisions and supplied interface evidence. It makes no measured retention or conversion claim.
NEXT TO VALIDATE
- Can someone find the main concern and explain their next step?
- Can they ask a follow-up without repeating the analysis?
- Can they distinguish waiting from failure and recover successfully?
COMPLETE DESIGN BOARDS
The full design narrative.
All ten main boards and three alternative explorations from the original Chinese portfolio. Expand a board to inspect its complete diagrams, annotations, and UI. Alternative boards remain identified as explorations.
01 / Overview and contribution
02 / Understanding and decisions
03 / Product architecture
04 / Before and after
05 / Follow-up intent and reference study
06 / Suggested questions
07 / Multimodal flow
08 / Input boundaries
09 / AI states and recovery
10 / Reflection
11 / Alternative / Reflection diagrams
12 / Alternative / Boundary detail
13 / Alternative / Recovery detail
