AK
All Projects
Conversational UI

AI Voice Assistant

Designing voice interactions that feel natural, not robotic

Conversational UIAI/MLInteraction DesignMobile
🎙️

Role

UX/Interaction Designer

Duration

4 months

Platform

iOS, Android

Year

2023

Overview

A context-aware AI voice assistant app that enables hands-free productivity for on-the-go professionals. The design focused on making voice interactions feel natural through smart visual feedback, error recovery patterns, and a conversational design language that matched how people actually speak.

Problem

Most voice assistants feel like talking to a machine — rigid command structures, poor error recovery, and no visual context clues. Users reported abandoning voice features after 2–3 failed interactions. The challenge: design a voice experience that handles ambiguity gracefully and keeps users in flow.

Goal

Design a voice interface that reduces friction for common productivity tasks by 50% while maintaining a 90%+ task success rate even for non-power users — achieved through smart visual scaffolding and graceful error handling.

Design Process

1

Research

Analyzed 500+ voice interaction failure logs. Conducted "Wizard of Oz" testing to observe natural speech patterns. Identified 12 core task categories. Found users front-load intent ("remind me to...") vs. end-load ("...tomorrow at 9am").

2

Conversation Design

Mapped 80+ conversational flows covering happy paths and failure states. Wrote voice design guidelines — tone, response length, clarification patterns. Designed the "partial match" confirmation pattern to handle ambiguity without frustration.

3

Visual Design

Created a waveform animation system tied to voice confidence levels. Designed real-time transcript display with inline editing capability. Built a "context shelf" that shows what the AI currently understands, preventing confusion.

4

Testing

Ran voice usability testing in noisy environments (cafes, commute). Tested 15 participants, 8 tasks each. Achieved 91% first-attempt task success rate. Iterated on error recovery flows 4 times before reaching target metrics.

Impact & Results

+50% task completion speed vs. manual input

91% first-attempt success rate for core tasks

User retention at Day 30: 44% (industry avg: 22%)

4.8/5 rating in beta with 400 testers

Reduced "assistant abandonment" events by 67%

Key Learnings

"Visual design is even more critical in voice interfaces than in traditional UIs. The visual layer is what saves the interaction when voice fails. Spending 40% of design time on error states and recovery flows — usually underfunded — was the highest-leverage decision on this project."