Redesigned Interactive Mode to make it easier to discover and sustain AI conversations,
increasing entry by 5× and average conversation time by 5.5×.
MY ROLE
Product Design
Interaction Design
Design System
UX Research
Prototyping
TIMELINE
Apr 2024 - Jul 2024
TEAM
2 Product Designers (Me)
1 Product Owner
2 Product Managers
6 ML Engineers
8 iOS/Android/Web Engineers
TOOLS
Figma
After Effects
Photoshop
Match Group, the company behind Tinder, Hinge, and Match, explored an AI dating experience where users build ongoing relationships with AI companions. I joined as a product designer in the early, pre-launch stage, designing Interactive Mode, where users go on role-play dates with their AI instead of just chatting.
Through user testing and behavioral data from 1,000+ users, we found two challenges: users couldn't find the mode during a call, and those who entered didn't know what to say next. I redesigned the experience to make AI dates easier to start and easier to keep going, driving a 5× increase in Interactive Mode entry and a 5.5× increase in average conversation time.
Match Group is the #1 global online dating company behind brands like Tinder and Hinge. Its portfolio serves different relationship needs and audiences, with a shared mission to help people build meaningful connections through technology.
Tinder lost 100,000 subscribers after the pandemic, and 79% of daters said they often feel burnt out by the effort of finding a meaningful match. Meanwhile, AI character chat grew 250% in four years, and many users were already treating AI characters as romantic partners. 67% said they used these services to create their ideal partner. These relationships lived only in text. We created Mel, an AI video dating service that brings users' ideal partners to life through realistic video calls.
To validate the potential of a subscription model, we tested whether users could sustain 10-minute AI video calls. The average call lasted only 5 minutes, as many didn't know what to talk about. But among users who talked for over 30 minutes, 90.1% used role-play and imagined dating scenarios.
Rather than making the AI more realistic, we decided to support what users already wanted. We designed Interactive Mode, where users go on role-play dates with their AI, like "You're a rock star and I'm your fan."
Our team’s vision was a natural, seamless conversation, so we initially used dialogue-only entry into Interactive Mode rather than UI guidance.
However, in user testing, most users didn't recognize the AI's suggestion as a way in. Those who did enter often paused, unsure what to say next.
As Interactive Mode was both the service’s core feature and the foundation of its business model, our primary goal was to help at least half of call users enter the mode. Our secondary goal was to extend conversation time through continued interaction.
Working with PM, design, and ML/engineering, we aligned user needs with product goals and technical constraints to define three design directions.
Product need
Surface Interactive Mode as soon as the call begins.
UX consideration
Keep entry natural instead of requiring exact commands.
Technical constraint
Avoid interaction patterns that require heavy implementation.
→ Persistent entry + contextual prompts
Product need
Give users reasons to continue after entering.
UX consideration
Help users decide what scenario to start and what to say next.
Technical capability
AI could generate scenarios and context-aware dialogue suggestions.
→ Scenario options + suggested replies
Market signal
AI chat users frustrated when conversation context was lost during longer interactions.
Technical capability
MEL could retain previous conversation context through long-term memory.
UX opportunity
Make that continuity visible so users could easily return to previous scenarios.
→ Interactive Mode history
Providing two entry points, a main CTA and “Request a Line”, will increase entry into Interactive Mode
Providing scenario options and dialogue suggestions will help users continue conversations more easily, while conversation history will make it easier to return
To balance seamlessness with discoverability, I explored three entry button designs and chose a curiosity-driven CTA.
Hidden behind an extra step
Users had to tap an unfamiliar icon
before discovering the two modes.
Too disruptive
The prominent prompt made the feature clear,
but competed with the conversation.
Curiosity-driven CTA
An animated, branded CTA invited exploration without interrupting the conversation.
Onboarding addresses the unfamiliar pattern, the persistent button fixes low discoverability, and the contextual nudge makes the transition clear.
Describing a desired scenario through conversation was unfamiliar UX to users, making it difficult to get started. We introduced two ways to start and supported both high-intent users and users who needed inspiration.
Even after choosing a scenario, users could pause while deciding what to say next. We added context-aware suggested replies to give users an easy starting point and keep conversation flowing.
After users left Interactive Mode, there was no clear way to return to a previous scenario. We introduced My Conversation, which automatically saves past sessions with the latest image and an AI-generated summary, making it easy for users to recognize and revisit their previous experiences.
Increase in entry rate
5x
13.38%→68.15%
The share of users who tried Interactive Mode immediately after entering chat rose significantly. Users also reported feeling closer to their AI partner when their shared history was remembered.
Conversation time longer
5.5x
5.4 min→29.5 min
Scenario options and suggested replies reduced pauses and helped users keep conversations moving—extending the average chat duration to nearly 30 minutes.
We initially assumed users wanted AI conversations to feel like real video calls, so we designed a minimal interface around that experience. However, usage remained below expectations, while behavioral data showed that users with longer conversations were more drawn to imaginative scenarios beyond everyday reality.
This taught us that in an emerging market without established mental models, design decisions must be guided by observed behavior rather than assumptions. Although time constraints prevented formal A/B testing, future iterations should validate user responses systematically to inform more impactful UX/UI decisions.
