AI Golf Coach — Case Study

From raw data
to one clear
action

A deterministic golf coaching system that turns performance metrics into one focused next step — instead of overwhelming players with numbers and competing advice.

Role
Product Design
UX & Design Flow
Scope
Coach, Analysis,
Results, Session Goal
Outcome
Stable Engine v1
Baseline
Coach
Analysis
M
Ready ↗ Updated 2 min ago
Improve focus metrics to change your results
⚡ Current Focus
Guiding
Improve Smash Factor
Limited evidence — cause unclear
Currently at 1.25 · Strike consistency
Hit 5 shots focusing on a centre strike
+9% potential
improvement potential
Cue · Centre strike
Focus Metrics · Tap to switch
Smash Factor
1.25
Outside optimal range
Current focus
Attack Angle
−1.0°
Too shallow
Dispersion
12 m
Outside optimal range
Ready to train
Smash Factor
+9% potential
Start coaching →
Live MVP
See the coaching flow working.
Experience the live MVP and see how the system turns raw data into one clear next action.
Try the live MVP →
System Design

One metric.
One action. One session.

Instead of a dashboard of competing advice, the system selects one focus metric and translates it into one immediate action. Each metric has its own threshold logic and confidence states.

Smash Factor
Strike efficiency.
Ball Speed ÷ Club Speed
Supporting data
Ball Speed Club Speed
Precise split: contact vs timing
Start No BS/CS and no typed SF
→
Guiding Ratio 1.20–1.35, no precise pattern
→
Precise: Contact BS gap > 10% vs expected
→
Precise: Timing BS + CS present, below club threshold
→
No issue SF > 1.35 (global)
Coach
Analysis
Ready
↗ Updated 2 min ago
Smash Factor
Precise
Strike consistency
Poor contact
BS 110 · CS 100
Strike the ball first, then the turf
Don't worry about distance yet
Begin 5 shot focus →
Show explanation ∨
Attack Angle
Steep or shallow delivery through impact
Supporting data
Launch Angle Spin Rate
Start AA missing, or outside optimal with no LA or SR
→
Guiding Outside optimal (club-aware), no precise pattern
→
Precise: Shallow AA above club shallow threshold, LA > 20, SR ≤ 6500
→
Precise: Steep AA below club steep threshold, LA < 16, SR > 6500
→
No issue AA inside optimal range
Coach
Analysis
Ready
↗ Updated 2 min ago
Attack Angle
Precise
Too shallow attack
Feel your hands ahead of the ball at impact
Consistent ground contact is the foundation
Begin 5 shot focus →
Show explanation ∨
Dispersion
Shot spread. How tight is the landing zone?
Supporting data
Face to Path Club Path
Start D > 10 or no dispersion data
→
Guiding 5 < D ≤ 10, or no dominant pattern
→
Precise: Face |FtP| ≥ |CP| + 1
→
Precise: Path |CP| ≥ |FtP| + 1
→
No issue Dispersion ≤ 5
Coach
Analysis
Ready
↗ Updated 2 min ago
Dispersion
Precise
Face issue
Clubface issue
Feel the face closing earlier through impact
Tighter spread means closer to target
Begin 5 shot focus →
Show explanation ∨
Confidence Model

Precision requires
evidence.

The system scales coaching tone across four states. It does not claim to be precise when supporting data is missing. Confidence is earned by the data, not assumed.

01
Start
Low confidence. Supporting data may be missing, or the issue may be too severe to classify precisely yet. Use a simple foundational drill.
Example action
"Hit 5 shots focusing on clean contact"
02
Guiding
Moderate confidence. Issue present but no precise pattern yet. Observational guidance. Avoids overclaiming.
Example action
"Notice where the club hits the ground"
03
Precise
High confidence. Supporting metrics justify a specific recommendation. Cause identified.
Example action
"Feel your hands ahead of the ball at impact"
04
No issue
Player is already performing well. Reinforce what works. Do not invent a problem to solve.
Example action
"Keep the same strike"
Reinforcement
When performance is good,
reinforce it
Smash Factor above the global 1.35 threshold. The system confirms what's working instead of finding a new issue to fix.
Ready
↗ Updated 2 min ago
Smash Factor
No issue
Strike consistency
Keep the same strike
Don't worry about distance yet
Continue training →
Move to next focus (Attack Angle)
Evidence-based precision
Precise only when the data
supports it
Attack Angle with Launch Angle and Spin Rate present. Shallow diagnosis confirmed. Specific instruction given.
Ready
↗ Updated 2 min ago
Attack Angle
Precise
Too shallow attack
Feel your hands ahead of the ball at impact
Consistent ground contact is the foundation
Begin 5 shot focus →
Honest under uncertainty
No direction data,
no directional claim
Dispersion needs work but Face to Path and Club Path are missing. The system stays observational and avoids guessing.
Ready
↗ Updated 2 min ago
Dispersion
Guiding
Limited evidence
Hit 5 shots and notice if the ball starts left or right
Dispersion controls your scoring zone
Begin 5 shot focus →
Outcome

A stable baseline that proved the concept.

Engine v1 established a coherent coaching system. One focus per session. Confidence scaled by evidence. Consistent status language across Coach, Results, and Analysis. No fake certainty.

✓
One focus metric per session
The system never overwhelms. One metric, one action, one session.
⚡
Confidence scaled by evidence
Four confidence states. Precision only when supporting data justifies it.
◎
Consistent across all screens
Coach, Results, and Analysis share one source of truth for metric status and confidence states.
✗
No state carryover bugs
Every session evaluates from current data only. No false confidence from previous sessions.
Coach
Coach
Analysis
M
Ready
↗ Updated 2 min ago
Session complete · 7 iron
Results
Edit feedback
Smash Factor — This Session
1.10
→
1.25
↗ Improved this session
What changed
Smash Factor moved closer to optimal. Stay consistent — results will follow.
What to keep
Current approach is working. Stay with it.
Effort vs Outcome
Good execution. Feel matches the data — keep it up.
Continue with
Attack Angle
Continue →
Reflection & Next
Engine v1 — Baseline locked

What this
project taught.

Designing a deterministic coaching system forced a fundamentally different design question: not "how do I display this data" but "how do I help the player decide what to do with it". That reframing shaped every product decision.

Decision-first design is harder than data-first design
Building a system that outputs one action requires more product clarity than building a dashboard. Every threshold, every state, every coaching copy choice is a product decision.
Trust is a design material
The confidence model is not a technical feature — it is a trust mechanism. Scaling feedback based on evidence quality is what makes the coaching feel credible rather than generic.
Logic and UI must be one system
State carryover bugs proved that when logic and UI drift apart, the experience breaks. Coach, Results, and Analysis had to share one source of truth.
Possible next
01
Engine v2 — Session history
Track metric trends across sessions. Let the system learn which focus areas improve fastest for each player.
02
Expanded metric set
Add Club Path, Dynamic Loft, and Spin Axis as focus metrics. Each would require the same confidence model treatment.
03
Coach tone personalisation
Let players choose between technical and feel-based coaching language while keeping the same underlying logic.
04
Multi-club session logic
Allow a session to span more than one club and carry context between them without leaking metric values.
Live MVP
See the system in action.
Experience the working MVP and see how the coaching flow turns raw performance data into one clear next action.
Open the live MVP  →