AI Audio Editing Mobile App
Key Insights — From navigating AI products to completing a task
Gaudio Studio's web experience was built around separate AI products. For the iOS and Android launch, I used behavioral data to rethink that structure, designed the core mobile experience, and led implementation through release.
The Problem
Lower Mobile Engagement
Product-First Navigation
Competing Information & Actions
What I did
Behavior-Driven IA
Mobile Interaction Design
Cross-Platform Delivery
Business Outcomes
iOS & Android Launch
Released the task-first experience on both platforms.
18.4% vs. 11.8%
Users completing 3+ tasks: app vs. mobile web.
17% → 27%
Mobile share of overall usage.
Background
A web-first AI audio product moving to mobile
Gaudio Studio helps users separate vocals, extract instruments, and clean up audio with AI. After two years as a web-first beta, the team was preparing the paid product and its first native mobile apps.
An earlier mobile concept had stalled when its owner left. I took over the project and revisited the experience before bringing it to launch.
Gaudio Studio's web-first experience before the mobile launch.
The Problem
Mobile users were already there, but doing less
Mobile accounted for 17% of usage, yet users completed fewer tasks and converted to paid less often than desktop users.
Mobile
Desktop
Tasks per user
1.74
2.63
Paid conversion
1.4%
4.9%
The demand for mobile access was already visible. I wanted to understand what was preventing users from doing more once they arrived.
Behavior Analysis
Two patterns behind the mobile experience
I analyzed eight months of product behavior to examine how users selected tasks, what files they uploaded, and where the existing interface might be creating friction.
Finding 01. Product categories did not match user intent
Stem Splitter and Vocal Remover had separate entry points to serve web acquisition, but many users appeared to want the same result from both.
27.5% of Stem Splitter requests selected the same output offered by Vocal Remover.
Among users who tried the same file in both products, 76% retried within 30 minutes, with a median interval of 6 minutes.
Support inquiries continued to show confusion between the two products.
The pattern was similar across desktop and mobile. This pointed to a problem with the product structure itself, not simply the smaller screen.
Finding 02. Mobile users brought different files and needs
Mobile users were much more likely to upload videos and less likely to use lossless audio.
Upload Type
Mobile
Desktop
Video
29%
9.7%
Lossless WAV
3.5%
17.9%
Yet guidance for high-quality audio files occupied permanent screen space, alongside credits and multiple actions competing for attention.
The data did not mean that audio quality no longer mattered. It meant that the guidance needed to appear in the context where it was useful.
Design Goal
Make the task the starting point
The redesign focused on two changes: removing unnecessary product decisions and making the information on each screen relevant to the task at hand.
INFORMATION ARCHITECTURE
From choosing a product to choosing an output
Separate product pages made sense on the web, where users often arrived through search. But there was no reason to preserve those distinctions in the app's task flow.
I removed the initial product selection and organized the interface around what users wanted to extract.
Choosing vocals and accompaniment led to the Vocal Remover workflow. Choosing individual instruments led to Stem Splitter. The underlying AI products remained distinct, but users no longer needed to understand that distinction before beginning.
Before and after · Product-based navigation replaced by output selection.
Interaction Design
Show what matters, when it matters
With the task flow consolidated, I revisited the information competing for space around the primary action. Three decisions shaped the mobile interface.
01. See all options without scrolling
Replaced full-width instrument lists with a two-column grid of output choices. Users could compare the available options in one view and select one with a tap.
02. Keep credits clear and actions focused
Reduced three credit values to the cost of the task and the remaining balance. Insufficient credits became visible through a color change, while a single Submit button replaced competing actions.
03. Move guidance into context
Moved high-quality audio guidance out of the main screen and into a bottom sheet shown when users selected Vocal HQ, where source quality mattered most.
Three interaction changes · Output selection, credit visibility, and contextual guidance.
Design System
Reusable patterns for different AI tasks
The mobile interface needed to support different AI models without introducing a new interaction for each one.
I built the core task experience around larger touch targets, compact information groups, and explicit selection and processing states.
Foundations
Typography, spacing, and component states were adapted for smaller screens rather than directly copied from the web.
Selection Chips
A reusable chip-and-grid pattern let users select vocals, instruments, or multiple stems within the same interaction structure.
The number of outputs could change by model without requiring users to learn a different selection method.
Cross-Platform Delivery
One experience across two native platforms
Designing the core task was only part of getting the app into users' hands. Mobile introduced platform-specific requirements for media access, background processing, payments, and measurement.
01. Handling mobile media and processing
Because video represented 29% of mobile uploads, I treated Camera Roll as a primary source alongside Files and URL.
Permission requests appeared only when needed, with clear recovery paths for denied access, unsupported formats, and failed uploads.
Audio processing could take several minutes. Users only needed to keep the app open during upload; once processing moved to the server, they could leave and return to check the status in Songs.
02. Pricing and measurement across platforms
The web and apps used the same time-based credit unit, but mobile offered smaller purchase bundles to suit shorter, more frequent tasks.
I defined the mobile purchase requirements and worked with finance to reconcile Apple's and Google's pricing and settlement differences with the existing system.
I also aligned task and conversion tracking across web, iOS, and Android, including users moving between platforms. This made it possible to compare behavior without treating the same person as separate users.
Platform requirements · Media access, processing states, purchases, and analytics.
Final Experience
A shorter path from file to result
The finished app brought file selection, output choices, credit information, and submission into one task-oriented flow.
Instead of navigating between AI products, users could begin with the result they wanted and follow the same interaction pattern across different audio tasks.
Final iOS and Android experience.
Outcome
More repeat use, but no meaningful lift in conversion
After launch, app users showed stronger repeat behavior than mobile web users, and mobile accounted for a larger share of overall usage.
18.4% completed 3+ tasks
Compared with 11.8% on mobile web and 16.6% on desktop.17% → 27% mobile share
Growth in mobile's share of overall usage.No meaningful improvement in paid conversion
The usage gains did not translate into a comparable change in payment behavior.
The results were encouraging for the task-first experience, though they did not establish that the redesign alone caused the difference in repeat use.
Reflection
A simpler experience was not enough to change payment behavior
The launch showed that users could engage more deeply with the product without becoming more likely to pay.
I had addressed friction within the task flow, but the underlying credit model remained largely unchanged. Improving conversion would require a separate investigation into pricing, purchase timing, and how users understood the value of each task.
What I'd explore next
The same product-selection confusion existed on the web. I would test whether the task-first structure could simplify that experience too, rather than keeping different ways of starting the same task across platforms.
Thanks for coming by.
© 2026 Hailey Moon