Video summary

I Tried E-Sports Betting with AI

Main summary

Key takeaways

Finance

Premise & Objective

  • Goal: Test whether an AI-driven sports betting “trading bot” can produce positive long-term returns.
  • Context: The creator notes that only about 3% of people are profitable long-term in sports gambling.

Niche Selection (Targeting a Data-Based Edge)

The bot restricts betting to esports / video games only, focusing on:

  • League of Legends (LoL)
  • Counter-Strike (CS)
  • Dota

Core Framework & Rules (Step-by-Step)

1) Universe Constraint

  • Bet only on esports with abundant daily data.

2) Bankroll Management

  • Use specific units/measurements/confidence scores to avoid early “blow ups.”
  • Starting bankroll: $1,000, fully controlled by the bot.

3) Model & Build Approach

  • Use Claude AI (Anthropic) to design:
    • an esports edge model
    • a full data pipeline
  • Include tooling such as:
    • EV (expected value) calculator
    • bet log
    • market analysis + live signals

4) Data Sources Mentioned

  • bo3.gg
  • HLTV map stats
  • OpenDota
  • Stratz
  • PandaScore (used as a web app feeding the live terminal/pipeline)

5) Bet Selection Logic Improvements

Key additions and adjustments during the challenge:

  • Add factors like:
    • “rustiness” (time since last match)
    • head-to-head matchups
    • maps
  • Identify and address an early problem:
    • The bot over-favored underdogs around 30–40% win likelihood (too close to coin-flip).
  • Underdog guard:
    • Do not take bets below 45% implied/model win probability.
  • High-conviction rule:
    • Mark bets “high conviction” when:
      • open market odds imply ~70%, and
      • the model predicts ≥70%
  • Backtesting & filtering:
    • Revisit earlier matchups
    • Blacklist some leagues
    • Adjust game-volume thresholds
  • Micro tier (late-stage risk control):
    • Use a small % of bankroll when confidence is below 70% but above 60%
  • Increase play opportunities:
    • Initially trade games with >10k volume
    • Later allow more plays per day / more games

Key Performance Numbers & Timeline (Day-by-Day)

Start / Early Loss

  • Day 1: Bot goes 0 and 2, down ~$70
    • Bankroll: $1,000 → ~ $820
  • Next stretch: Goes 0 and 4, balance reported as:
    • $763.44
    • Down $200+ on the day-one timeframe

Partial Recovery

  • Day 3 (LoL rebound):
    • Goes 3 and 1 on League of Legends
    • Profit: ~$15
    • Bankroll noted as ~down $225 vs earlier ~$250

Mid-Challenge (High-Conviction Plays Added)

  • Day 4:
    • Places 5 bets across LoL / CS / Dota
    • Goes 3 and 1
    • Profit: ~$13
    • Bankroll described as just under $800 (range mentioned: ~$200+ down to ~$250 down depending on the paragraph)

Worst Event (Format / Market Mismatch)

  • A late “overnight” loss described:
    • Market priced an LoL match at 70% likelihood
    • Bot predicted 65%
    • The match format was best-of-one (single win decides)
    • Bot placed $60 due to “high-conviction” flagging
    • Resulted in a loss
  • Creator describes the day as “astronomically” down (explicit total not fully quantified at that moment)

Late-Stage Adjustments

  • Add micro tier (60–70% confidence uses small bankroll %)
  • Filtering changed:
    • earlier >10k volume
    • later allowed more games to be traded in a day

Day 6 (Strong Bounce)

  • 19 plays: 15 and 4
  • Four high-conviction plays all win
  • Net profit: ~$21
  • Balance after day: ~$740
    • Still ~$260 down overall

Day 7 (Final Day)

  • Starts with 12 open positions
  • Up $14.32 on the day
  • Current balance: $753
    • Overall framing: still ~$250 down
  • Qualitative ongoing results:
    • winning streaks on Dota and LoL
    • one LoL loss
    • CS beginning to find wins
  • Creator’s framing:
    • If not for day one’s mishap, they’d be “in the green.”

30-Day Plan

  • Creator states they will run the bot for 30 days and publish another 30-day video for longer-run validation.

Explicit Recommendations / Conclusions (Creator’s Take)

Edge Claim (Conditional)

  • Creator believes an edge exists, particularly in esports because of:
    • more markets
    • higher variance
    • more online free data

Where It Worked vs Failed (Qualitative)

  • Good areas:
    • Dota and League of Legends
    • especially on 70%+ confidence plays
  • Bad area:
    • Counter-Strike, described as a source of problematic losses

Method Discipline Emphasized

Sustained profitability (if achieved) requires:

  • proper rule configuration
  • strict elimination of bad markets
  • continued self-learning/calibration

Disclosures / Caveats

  • Creator frames the experiment as a challenge and emphasizes honesty/transparency (no “sugarcoating”).
  • No explicit legal “not financial advice” disclaimer is shown in the provided subtitles.

Tickers / Assets / Instruments Mentioned

  • No traditional financial tickers (stocks/ETFs/bonds).
  • Betting is indirectly on esports teams/markets.
  • Data/products and platforms referenced:
    • PandaScore, bo3.gg, HLTV, OpenDota, Stratz
  • No mention of commodities/FX/crypto.

Presenters & Sources

  • Presenter: Single creator/narrator (name not provided in subtitles)
  • AI system: Claude AI by Anthropic (also references “Claude Code”)
  • Data/model sources: bo3.gg, HLTV, OpenDota, Stratz, PandaScore

Original video