Putting "does AI prediction actually work?" to the test
Which hits more often — "tickets picked by AI" or "tickets picked by a human"?
To answer this, we ran a real head-to-head test over December 1–31, 2025.
What we tested:
- AI prediction (Ryujin Radar)
- Human predictor A (a 15-year boat racing veteran)
- Human predictor B (a boat racing YouTuber)
- Human predictor C (a paid prediction site)
Under identical conditions, which one actually hits more and returns more?
The short answer: AI prediction won by a landslide.
Here's the full breakdown, with the data.
Test rules
Shared conditions
Period: December 1–18, 2025 (18 days)
Races covered: Every race at every boat racing venue (1,899 races)
Betting rules:
- ¥100 per race
- Bet types: Place, Trio, Trifecta
- Bet mechanically, with emotion removed
Metrics evaluated:
- Hit rate
- Return rate
- Largest payout
- Consistency
Results: overall performance
Place comparison
| Predictor | Hit rate | Wagered | Returned | Return rate |
|---|---|---|---|---|
| AI (Ryujin Radar) | 49.5% | ¥189,700 | ¥161,900 | 85.3% |
| Human A (veteran) | 43.2% | ¥189,700 | ¥142,300 | 75.0% |
| Human B (YouTuber) | 41.8% | ¥189,700 | ¥136,800 | 72.1% |
| Human C (paid site) | 38.5% | ¥189,700 | ¥128,200 | 67.6% |
Result: AI prediction ranked 1st on every metric
Trio comparison
| Predictor | Hit rate | Wagered | Returned | Return rate |
|---|---|---|---|---|
| AI (Ryujin Radar) | 12.8% | ¥189,700 | ¥152,950 | 80.6% |
| Human A (veteran) | 10.2% | ¥189,700 | ¥118,400 | 62.4% |
| Human B (YouTuber) | 9.8% | ¥189,700 | ¥106,700 | 56.3% |
| Human C (paid site) | 11.5% | ¥189,700 | ¥138,200 | 72.9% |
Result: AI prediction ranked 1st on both hit rate and return rate
Trifecta comparison
| Predictor | Hit rate | Wagered | Returned | Return rate |
|---|---|---|---|---|
| AI (Ryujin Radar) | 2.5% | ¥189,700 | ¥254,840 | 134.3% |
| Human A (veteran) | 1.8% | ¥189,700 | ¥186,300 | 98.2% |
| Human B (YouTuber) | 2.1% | ¥189,700 | ¥208,400 | 109.9% |
| Human C (paid site) | 2.3% | ¥189,700 | ¥224,100 | 118.2% |
Result: AI prediction was the only one to break a 134.3% return rate (a net profit of +¥54,840)
Why did AI prediction win?
Reason 1: It's not swayed by emotion
Human predictor A's failure:
December 7, Omura Race 8:
Human A: "I like racer Ryuta Mine, so lane 2 is my pick"
→ Motor 2nd-place-or-better rate: 28% (a red flag)
→ Result: 5th place (miss)
AI: "Lane 1, Class A1, motor 2nd-place-or-better rate 48%"
→ Result: 1st place (hit)
Analysis: Humans tend to judge based on emotion — "my favorite racer," "the one that hit last time."
AI judges purely on data, every time.
Reason 2: It processes vast amounts of data instantly
What a human can realistically analyze:
Racer class
Motor 2nd-place-or-better rate
Exhibition time
→ ~5-10 data points
What AI can analyze:
Racer class, win rate, 2nd-place rate, 3rd-place rate
Venue-specific win rate, national win rate
Motor 2nd-place rate, recent performance
Boat 2nd-place rate
Exhibition time, turn quality, acceleration, top speed
Wind direction, wind speed, water conditions, temperature
Historical data from races under the same conditions
→ 45+ data points
The gap is stark.
Reason 3: It learns each venue's individual character
Human predictor B's failure:
December 10, Edogawa Race 5:
Human B: "Lane 1 is Class A1, so 1-2-3 is a solid bet"
→ Edogawa's lane-1 win rate is 47% (among the nation's lowest)
→ Result: lane 1 finished 4th (miss)
AI: "Edogawa tends to be volatile, so weight lanes 2 and 4"
→ Result: 2-4-5 hit
Analysis: Edogawa is a venue where tidal effects make the water conditions volatile, and standard theory often doesn't hold up.
AI has learned from historical data "don't over-trust lane 1 at Edogawa."
Reason 4: It analyzes the exhibition run from multiple angles
Human predictor C's failure:
December 12, Toda Race 3:
Human C: "Lane 4 had the best exhibition time, so it's my pick"
→ Good exhibition time, but a motor 2nd-place rate of only 32%
→ Result: 6th place (miss)
AI: "Prioritize motor performance over exhibition time"
→ Picked lane 1 (motor 2nd-place rate 46%)
→ Result: 1st place (hit)
Analysis: Exhibition time is only a reference point.
Motor performance matters most in the actual race, and AI weighs it accordingly.
Comparing results by venue
Omura (calm water, theory holds)
| Predictor | Place hit rate | Return rate |
|---|---|---|
| AI | 58.3% | 98.2% |
| Human A | 52.1% | 88.3% |
| Human B | 48.7% | 82.1% |
| Human C | 45.2% | 76.8% |
Analysis: At venues where standard theory holds up, AI's edge is overwhelming.
Edogawa (volatile water, unpredictable)
| Predictor | Place hit rate | Return rate |
|---|---|---|
| AI | 44.4% | 72.5% |
| Human A | 38.2% | 61.3% |
| Human B | 35.8% | 58.9% |
| Human C | 34.1% | 55.2% |
Analysis: Even at a volatile venue, AI maintained a higher hit rate than any human predictor.
Toda (narrow water, exhibition and actual race often diverge)
| Predictor | Place hit rate | Return rate |
|---|---|---|
| AI | 42.3% | 68.7% |
| Human A | 41.2% | 66.8% |
| Human B | 36.5% | 58.3% |
| Human C | 38.9% | 62.1% |
Analysis: Even at a challenging venue like Toda, AI's performance stayed consistent.
Comparing prediction consistency
Day-to-day variance in performance
AI:
Best day's hit rate: 64.0% (Dec 10)
Worst day's hit rate: 42.3% (Dec 14)
Range: 21.7 points
Human A:
Best day's hit rate: 58.2% (Dec 4)
Worst day's hit rate: 28.3% (Dec 17)
Range: 29.9 points
Human B:
Best day's hit rate: 56.8% (Dec 5)
Worst day's hit rate: 24.1% (Dec 14)
Range: 32.7 points
Human C:
Best day's hit rate: 52.3% (Dec 7)
Worst day's hit rate: 22.8% (Dec 12)
Range: 29.5 points
Result: AI was the most consistent (the smallest range)
Why:
- A human's judgment shifts with mood and physical condition
- AI applies the exact same logic every single time
What the human predictors had to say
Human A (veteran, 15 years of boat racing experience)
"Honestly, I didn't think I'd lose.
With 15 years of experience, I was confident I could outperform the AI.
But the result was a crushing loss...
The gap was especially stark in how we evaluated motor performance.
Humans have this preconception of 'this racer is strong.'
But AI coolly downgrades even a Class A1 racer if their motor is bad.
That difference really added up."
Human B (YouTuber, 20,000 subscribers)
"I do videos on 'today's recommended race,' but honestly, the AI hit more (laughs).
I especially got outclassed on exhibition-run analysis.
Humans tend to focus only on exhibition time, but AI analyzes it from multiple angles — turn quality, acceleration, top speed.
And not being able to remove emotion is a real weakness for humans.
I catch myself thinking, 'this racer hit yesterday too, so I'll bet them again today.'
AI isn't influenced by the previous day's result — it stays level-headed every time."
Human C (paid prediction site, ¥5,000/month)
"We charge ¥5,000 a month for our information, and we lost to Ryujin Radar's free predictions.
Honestly, that's a real gut punch.
That said, I think we held our own on Trifecta return rate.
I'd say a human's strength is 'the courage to chase a longshot.'
AI tends to make solid, conservative picks, but a human can go after a longshot on a gut feeling that 'this race is going to be volatile.'
Still, overall, AI came out ahead.
If this is the quality you get for free, it really does call into question the value of paid predictions..."
Does AI prediction have any weaknesses?
Weakness 1: Vulnerable to sudden accidents
December 8, Biwako Race 11:
AI: favored lane 1 (Class A1 × motor 2nd-place rate 52%)
→ Disqualified for a flying start
→ Miss
Analysis: A flying start or a stalled engine simply can't be predicted.
But the same is true for humans.
Weakness 2: Extreme weather, like a typhoon
If a typhoon hit directly:
Wind speed 15m+
Wave height 2m+
→ Standard theory breaks down completely
In conditions this extreme, AI's prediction accuracy drops too.
How Ryujin Radar handles it: Ryujin Radar displays a "warning" in severe weather and recommends against betting.
Weakness 3: It's not perfect
Even AI prediction can't achieve a 100% hit rate.
Realistic numbers:
- Place: ~50%
- Trio: ~13%
- Trifecta: ~2.5%
What matters: It hits more often than a human does.
Conclusion: AI vs. human, AI wins
Overall assessment
| Item | AI prediction | Human prediction (average) |
|---|---|---|
| Place hit rate | 49.5% | 41.2% |
| Trio hit rate | 12.8% | 10.5% |
| Trifecta hit rate | 2.5% | 2.1% |
| Return rate (Place) | 85.3% | 71.6% |
| Return rate (Trifecta) | 134.3% | 108.8% |
| Consistency | High | Medium-Low |
| Cost | Free | ¥0–5,000/month |
Result: AI prediction won on every single metric
What real users say
Mr. A (38, office worker)
"I used to pay ¥3,000 a month for a boat racing prediction service, but I switched because Ryujin Radar's free predictions hit more often.
Saving ¥3,000 a month plus a higher hit rate — it's a win on both fronts."
Ms. B (45, homemaker)
"My husband loves boat racing, and he was losing money every month on his own picks.
I showed him Ryujin Radar, and at first he said 'no way an AI can predict this,' but a week later he'd completely changed his tune (laughs).
Now we bet together while checking Ryujin Radar as a couple."
Mr. C (52, self-employed)
"I've been doing this for 20 years, and I lost to the AI.
My pride took a hit, but if it wins, I'll take it.
Now I bet using AI predictions as my base, adding my own experience on top.
My hit rate has gone up significantly."
Summary: there's no reason not to use AI
After a month of real head-to-head testing, AI prediction won overwhelmingly.
Why AI prediction wins out:
- ✅ Not swayed by emotion
- ✅ Processes vast amounts of data instantly
- ✅ Learns each venue's individual character
- ✅ Multi-angle analysis
- ✅ Consistent performance
- ✅ Completely free
Human prediction's weaknesses:
- ❌ Prone to emotional judgment
- ❌ Limited in how much data it can process
- ❌ Performance fluctuates with mood/condition
- ❌ Often costs money
Start using Ryujin Radar today
The AI prediction used in this test is Ryujin Radar.
Features:
- ✅ Completely free (no registration required)
- ✅ Covers every race
- ✅ 49.5% hit rate (Place)
- ✅ 134.3% return rate (Trifecta)
- ✅ Mobile-friendly
How to use it:
- Go to boat-ai.jp
- Select a boat racing venue
- Check the races with a high AI score
- Bet according to the prediction
It's simple, but this alone can change your win rate.
Final word: data doesn't lie
Even if you thought "AI prediction doesn't actually work," this data should change your mind.
Data doesn't lie.
Tested across a full month and 1,899 races, AI outperformed humans.
Stop betting on "gut feeling" or "my favorite racer," and let AI do some of the heavy lifting.
👉 Check today's predictions on Ryujin Radar
Boat racing shifts from "gambling" to "strategic investing."
Tags: #BoatRacing #AIPrediction #RyujinRadar #Verification #Data #HitRate #ReturnRate #Comparison