Output Arcade
Output Arcade 2
Fixing content discovery that was quietly killing trial-to-paid conversion for 300,000 monthly producers.
63% → 18% search abandonment · 32% increase in paid subscriptions
Arcade launched strong. But paid subscriptions started tapering.
That pointed to a product problem.
Average time from first login to first meaningful export was 2.2 weeks. The trial window was one week. Producers were churning before they ever had a reason to stay.
For every 23 samples a producer opened, they saved one. Users were exploring constantly, but almost nothing was landing.
On new kits not featured on the homepage.
Nearly two-thirds of all searches ended without engaging a single result — a direct signal that results weren't trustworthy.
Average time to reach a first meaningful action. For a 1-week trial, that's too slow to convert.
We believe that by improving our content discovery, users will reach their first song export within the trial window.
How might we help users quickly discover and explore new music samples that aligns with their taste, especially as we release new lines each week?
Starting With Search
Three research signals pointed at the same root cause, but we didn't chase all three at once. We started with search (63% abandonment) because it was the most contained: it wouldn't touch the existing browse experience, and if we were wrong, it'd be cheap to find out.
Our hypothesis was simple: results were scattered across two bars, one for packs, one for tracks. Consolidate them, and producers find things faster.
We had a UX engineer wire a mid-fi version to real backend data so the test would hold up, and had our researcher run sessions against it.
The metric didn't move. Producers got results. They just weren't the good results.
What We Found
Results caused by bad data
The tags had been added by hand over years, with no shared standard. “Percussion” and “percussive” were separate tags with separate results, so a search returned a slice of the library.
What Producers Told Us
How samples were searched
Duplicates were half of the problem. Our UX researcher and I interviewed eight paying subscribers, chosen from our customer experience surveys across a range of satisfaction.
The tags were descriptors such as “tense” and “organic.” Producers started a search from something concrete: an instrument, a genre, a mood, a key, or a BPM. BPM and key already appeared on every result, and producers could not filter by them.
3 Findings & What Each One Changed
Get Users Closer To Music
35% of producers could not explain the difference between Lines and Kits, and they had to pick a category before seeing any content. Navigation was rebuilt around intent: Play, Browse, and Your Stuff.
Search By Category
Duplicate tags split the results. The library was re-tagged into five categories: instrument, genre, mood, key, and BPM. Each one became a filter.
Nothing To Match On
Descriptor tags gave the feed no reliable signal, so recency was the only ranking. The same five fields now match content to each producer.
Navigation
Closer To The Music
Play, Browse, and Your Stuff
Three clear sections replacing a sidebar-driven structure, organized around what someone's doing, not around Output's internal content types. Simple enough to internalize on first use.
Removed Sidebar
Eliminated the persistent sidebar that required producers to orient before they could create. This also cut discovery-related support tickets significantly.
Unified the Kit Player and Browser
Previously two separate views — merged into one continuous experience. Less context switching between exploring and playing.
Search
Organizing Chaos
Global Search, Available Everywhere
We introduced a unified global search that works across the entire library — from kits and lines to individual samples. Users can now search from any screen, including within their personal libraries and Output's full catalog.
Rebuilding The Taxonomy
The five categories came from how producers start a search. The content team re-tagged the library against them, and each one became a filter.
Results That Reduce Steps
In interviews, producers previewed a sample, dragged it into Play, returned to search, and repeated until the keys were full. Play and search were separate screens, so every sample meant leaving search.
We brought the keyboard into the results.
Homefeed
Tailored Content Powered By Data
Unified View
Merged Feed and Lines into a single Browse view, eliminating an unnecessary context-switch.
Multiple Entry Points
Some producers open Arcade already knowing what they want — an instrument, a mood, a BPM range for the track they're on. Others open it to be shown something. We built for both: instrument, genre, mood, and BPM entry points at the top for the first group, trending and editorial carousels below for the second.
Hybrid Feed
Algorithmic recommendations layered with editorial curation. Fully algorithmic risked irrelevant content. Fully editorial wouldn't scale. The hybrid gave us relevance at scale with a quality check on top.
Weekly Kit Mix
The old "newest kits" row had one variable driving it: recency. We went from four visible tiles to five, and from recency to engagement — saves, plays, what similar producers responded to. Refreshing weekly instead of chasing each day's release gives a full week of content a real window to be seen, instead of getting buried by whatever ships next.
Impact
Results that validated the diagnosis
Producers were finding music samples much quicker and tailored to them.
Search shifted from a last resort to a primary entry point — decreasing the numbers to only 18% abandonment.
Users finding value faster, creating new songs quicker. Still outside of the trial window but a huge improvement.
Users were finding the magic moment, which gave them clear value of the product.
Prototyping
The Figma prototype served two purposes: user testing, and engineering reference. The interactions were complex enough that written specs were losing detail in translation. A live prototype gave engineers a ground truth to reference during build and QA.
Components
Design system & Cross-platform
Arcade runs on web and as a plugin inside DAWs. UI constraints differ across surfaces — the mental model has to stay consistent. I built the design system in Figma with named components tied to variables, keeping taxonomy and discovery language consistent across both. Developers had a single source of truth.
Reflections
Surface problems often have infrastructure roots
The UI improvements were downstream of fixing the metadata. Doing UI work first would have been polished and ineffective.
Design has to sell infrastructure investment
Getting the content team to re-tag the library required reframing it as product infrastructure, not cleanup. Designers are often the advocates for invisible work that makes everything else possible.
A prototype is a communication tool, not just a test artifact
Building a Figma prototype as an engineering reference eliminated more implementation confusion than any amount of written documentation could have.
Research & Project Goals
Based on our user research methods of conducting 1:1 interviews with producers of our existing product and studio sit-ins. We discovered 3 distincted opportunities to streamline the creative process of music making and increase discoverability.