SneakySwing:
Launching an AI Golf Coaching App from 0 to 1
SneakySwing keeps coaching going between lessons through an AI-native ecosystem. It gives students real-time feedback backed by their coach's “digital twin,” while giving coaches a scalable platform to manage students and monetize their expertise.
Download on iOS ↗My Role
Product Design
Timeline
Feb 2026 – Present
Team
TL;DR
From First Launch to First Aha Moment
I designed and shipped SneakySwing from scratch. After launch, I used data to pinpoint key drop-offs and reshaped the journey from download to the first AI report (the Aha moment), lifting activation from 26% to 43%.
Shipped
Successfully designed and launched SneakySwing to the App Store from 0 to 1.
550+
Organic downloads for the first month post-launch.
26% → 43%
Onboarding funnel improvement — from download to first AI swing report.
44%
Weekly active users sustained post-launch.
Background
Led the Product Transition from Beta to Public Launch
When I joined as the sole designer, the product had working tech but no coherent experience. With launch weeks away, I took ownership of restructuring the core flows, building design systemClick to see case study, and designing missing interfaces such as recording instructions and subscription.
The Problem
Why I Started Redesigning Onboarding
291 people downloaded in the first two weeks, but only 26% reached “AI Analysis Completed” (The Aha moment).
According to Posthog data, the major drop-off happens between sign-up completion and the first AI Analysis request. This is a critical issue to address before we aggressively go to market.
To validate our hypothesis quickly, we conducted usability tests with 20+ users and observed where they got stuck. This gave us a clear picture of the main frictions before moving into design.
Usability Test
How I Tackled the Problem
We observed 20+ users freely exploring the app without guidance, and conducted interviews to understand why they were leaving during the onboarding flow. We identified 2 major frictions:
Download
−30%
First key drop off
Value proposition wasn't clear
Users didn't understand what they were signing up for before committing.
Registered
−44%
Second key drop off
Empty home screen, no clear next step
When users land on the home screen,
they need to record or upload a swing video
to start their AI analysis.
Even with a sample video/report available,
the minimal guidance wasn't enough
for many users to take that first step.
AI Report
Design Trade-off
Personalization vs. Onboarding Friction
Golfers deeply value personalization. However, AI feedback feels generic. Should we ask profile questions (goal, skill level, handicap...) during onboarding to give AI more context?
The concern
Adding too many steps before users experience any value may risk losing them during the long onboarding process.
Decision
Moved profile setup to the report generation step. While users wait ~30 seconds for the AI report, collecting their profile questions here reduces perceived wait while giving the LLM more accurate context for a better analysis result.
Jump to see →Onboarding Redesign
From Friction to Flow
Goal: Redesign the full onboarding flow to guide users to their Aha moment and drive activation.
Communicated SneakySwing's unique positioning with the welcome screen — helping golfers improve sneakily between lessons.
26% → 43%
Onboarding completion rate
+17%
Overall funnel improvement
+10.7%
Registered → AI Report step
AI Report Pain Points
The Report Is the Aha Moment, But We Were Losing Users There
After onboarding improvements lifted activation to 43%, attention shifted to the next drop-off: the AI report itself.
Cognitive Overload
Unlike a real coach who focuses on 1–2 fixes per session, the AI report delivered every insight simultaneously, overwhelming users.
Design for one clear takeaway. Users shouldn't have to hunt for what matters.
AI Wait = Churn
Frame breakdown (15s) then motion analysis (60s+) make users churn.
Turn dead air into engagement. The goal is not just to fill time, but to reduce the felt wait, so users stay present instead of drifting away.
No Trust
Our brand was new to users with no signals about where the analysis came from. PGA coach backing existed — but wasn't surfaced.
Make the AI accountable, not a black box. Users need to know who's behind it and how it works before they believe and trust what AI says.
How much a user trusts the AI depends on how much they already know about golf.
Golfer Personas by Skill Level
When they first open the AI Report
"I don't even know if the report is right..."
- ·Unfamiliar with golf jargon
- ·Too many problems at once, no sense of priority
- ·Disengages from confusion
Needs simpler language without jargon, and a clear focus on what to fix first
"Oh wow, the report gets exactly my problem."
- ·Seeing their own flaw in the report creates an instant trust moment
- ·But unsure how to act on it or what to drill next
Our most promising segment once we pair trust with a practical, actionable next step
"Where is this data coming from?"
- ·Skeptical of AI credentials
- ·Needs individualized depth
Needs credibility signals + advanced detail
Solution Ideation
Evaluating 3 Layout Solutions
The focus was solving Cognitive Overload by redefining the report's information architecture.
I pitched 3 layout directions to the founding team, evaluating each against technical constraints, product goals, and team bandwidth, while keeping extensibility for future features in mind.
A · Tabs
B · Accordion
C · Horizontal Cards
Pros
+Balanced info load per tab
+Closest to the existing UI with lower dev cost
+Prioritizes primary content and hides secondary details
+Easy to scan and absorb without extensive scrolling
+Closest to our vision of an AI-native report
Cons
−Tab labels must be precise, or users won't find what they need
−Still feels long overall
−Content placed later gets less attention
−Higher dev cost
Horizontal Cards was the best option.
Unlike tabs or accordion, cards are inherently modular. Each card is an independent unit that can be reordered, replaced, or generated on demand.
→Aligns with our vision of a truly personalized, AI-native report where users prioritize the insights they care about, reorder cards by dragging, and let the AI generate the next insight based on what they engaged with.
Final Solution
New Features Designed to Solve Cognitive Overload, Churn, and Trust
1. Collecting Profile Questions
During report generation, a bottom sheet invites users to answer a few questions.
- →Collecting user context that makes the AI analysis more accurate and personalized
- →Gathering user and acquisition data that benefits internal teams
2. Add Credentials
During report generation, a ticker rotates trust signals, such as "Built on 1M+ analyzed swings. Validated by PGA experts."
- →Increasing trust in AI
- →Reduces felt wait time by giving users something to read
3. Help Users Reach the Last Card
The horizontal layout reduces cognitive load by showing one card at a time. To maintain completion rate, the report auto-advances every 4 seconds.
- →Higher report completion rate
- →Users retain ownership through manual carousel navigation
AI Coaching Agent Design
Defined agentic dialogue flow and input guidance for swing analysis and report clarification, distilled from coach expertise.
