How AI Is Used in Professional Betting

How can AI be used in professional betting?

A: Speaking from personal experience, I first became involved in professional betting in 2018. From then until the arrival of ChatGPT, I relied on traditional programming to turn as much of the knowledge and skill I had gained from manual work into software. This included basic tasks such as acquiring data, cleaning it, and developing and validating strategies. My limited programming ability meant that even a small problem could take a long time to solve. Because I had not yet built an end-to-end workflow, I also spent a great deal of time solving things that were not really problems. One example is the team-name normalization issue I mentioned in another article.

Then ChatGPT arrived. First, it greatly improved my coding efficiency. It also dramatically improved my ability to collect data. Unless a website has particularly strong anti-scraping measures, ChatGPT can usually help obtain the required data. Most of what I collect consists of real-time markets and odds for full-time 1X2, Asian handicap, and totals. I stored large amounts of this tick-level data in the hope of discovering useful patterns.

I did not, however, try to feed all this data directly into AI, because I expected that doing so would consume a huge number of tokens. Instead, I design the strategy and backtesting logic, ask AI to write the code, and run it locally.

One strategy I once considered especially valuable was a lead-lag strategy between bookmakers. For example, after one bookmaker changes its line or odds for a match, another bookmaker may move in the same direction after a certain delay. If that pattern held, it could extend the window for real-time arbitrage and make it possible to place bets using ideas from statistical arbitrage. Unfortunately, the bookmakers and odds in my stored data did not support fitting a profitable strategy of this kind.

That does not mean lead-lag relationships never exist between bookmakers. Rather, I cannot effectively place bets with the bookmakers where I found such relationships.

If readers are interested, I will explain the fitting and backtesting process for this strategy in detail in a future article.

How should you discuss football betting strategies with AI?

A: I have read accounts of users asking AI to analyze a match and then using its prediction of the result. In principle, this is like tossing a coin. Statistically, the long-run hit rate is 50%, whether the prediction comes from AI, a bettor, a bookmaker, a monkey, Einstein, or Messi. One reason football has become the world's most popular sport and its largest betting sport is that live broadcasts and random events on the pitch make it a relatively fair and transparent betting subject.

Consider esports betting, whose market continues to grow. When viewers cannot inspect the game's code, manipulation for sufficient financial gain may be difficult to detect. I therefore predict that organizer misconduct will eventually have a severe impact on esports betting.

When I discuss a match or league with AI, I do not rely on its predictive ability, because I understand the underlying principles of LLMs (large language models). Instead, I usually ask it to analyze a strategy I have already developed so that it can identify gaps in my reasoning or knowledge. This is work AI does well. After an idea has gone through several rounds of discussion between a person and AI, it becomes more robust. Even if it is not correct, it is less likely to be seriously wrong.

Take the "strong home team" strategy as an example. When I learned about this strategy, which some bettors had treated as gospel long ago, I initially assumed that flat staking would be enough to test whether it was profitable. After discussing it with AI, I strengthened the backtest by using the Kelly criterion for position sizing. Although the strategy ultimately produced a negative result, this was also the first time a conversation with AI helped me strengthen one of my strategies.

For the actual prompt, I believe professional football betting probably represents only a very small share of the training corpus of large AI models. A question that is too short may not provide enough direction toward the precise concepts involved. I therefore tend to give AI a highly structured and detailed description of the background, followed by the most detailed possible statement of my question.

Prompt:

I am a professional football bettor. I would like you to act as my strategy adviser and review and advise me on my strategy. It is widely known that strong home teams have an obvious advantage in football matches, but whether this advantage can be used for professional betting is a question I need to study. Based on your knowledge of historical football matches and ideas from quantitative investing, please provide a comprehensive analysis of this strategy.

Answer:

The advantage of a strong home team is real, but it is overestimated and difficult to monetize directly. It needs strict filtering and risk management based on quantitative methods. It is not a simple "buy and hold" strategy, but a marginal edge that must be optimized within a multifactor model.

1. Historical data: home advantage exists, but its strength and stability are limited

  • Overall win rate: Across major leagues worldwide, home teams typically win around 45-55% of matches, with draws at roughly 25-30% and away wins at 20-30%. Large historical datasets, for example, put the home win rate at about 46%, nearly twice the away win rate but far from an overwhelming advantage.
  • Top-league trend: In high-level leagues such as the Premier League, the home win rate has fallen from historical highs above 60% to around 42-47% in recent years, with some seasons below 40%. The sharp decline in home advantage during COVID-era matches behind closed doors shows that crowd support is important, though not the only driver.
  • The xG perspective: Home teams generally create more high-quality chances, with an advantage of around 0.3-0.5 expected goals, but differences in team strength dominate when a strong team plays a weak one. Home advantage appears more strongly in attack than in defense or referee bias, though the latter may have a small effect.

Quantitative perspective: Home advantage is equivalent to an invisible handicap. It may be worth 0.3-0.6 goals between evenly matched teams, but for strong home teams this advantage is fully reflected in the odds. It also declines according to the league, era, and specific teams involved.

2. Betting feasibility: a blind strategy will almost certainly lose; an edge requires careful research

  • ROI from blindly backing home favorites: Historical backtests show that simply backing home favorites, especially at short odds, is unlikely to be profitable over the long run and may produce negative ROI. Bookmakers already incorporate home advantage into their prices, while the vigorish or margin is usually 5-10%, and favorite prices are compressed. Studies covering tens of thousands of matches show that blindly backing either favorites or underdogs does not generate sustainable positive returns.
  • Why this is difficult:
  • Market efficiency: Sharp bookmakers such as Pinnacle price markets very efficiently, and the consensus has already absorbed home advantage.
  • Variance and sample size: Individual football results contain substantial noise from luck, red cards, and injuries. Thousands of matches may be needed to validate an edge, and even strong home teams unexpectedly draw or lose.
  • Short-odds trap: Odds of 1.30-1.50 require a win rate above 70-77% just to cover the margin, which the actual win rate may not reach.
  • Declining trend: Better travel, VAR, tactical development, and player mobility have weakened home advantage in modern football. Continued reliance on historical data will overestimate it.

Analogy with quantitative investing: This resembles a well-known factor in value or momentum investing whose alpha has already been arbitraged away. Pure beta exposure from blindly backing home teams does not produce excess returns.

3. Strategy advice: build a multifactor quantitative framework

Do not use "strong home team" as an independent strategy. Integrate it into the system as a filter or weighted factor. The goal is to find home value that the market has underestimated: positive-EV bets.

Core framework, in the style of quantitative investing:

  1. Basic filters
  2. Consider only top-tier leagues and strong teams at home to reduce noise.
  3. Require a large difference in quality, such as a significantly better Elo rating or league position for the home team.

  4. Avoid very short odds below 1.40 and look for possible value in the 1.50-2.20 range.

  5. Combine multiple factors to improve win rate and value

  6. Recent form and rest: A home team with a winning run or strong xG trend, combined with an away team suffering from fatigue or travel.
  7. Injuries and lineup: The absence of key players can strengthen or weaken home advantage.
  8. Head-to-head and venue-specific history: Some teams are exceptionally strong at particular home grounds.
  9. Motivation: A title race or relegation battle versus a team with little left to play for.
  10. Contrary factors: Avoid bets when the away team is strong, has a good away record, or the home team has a history of poor home performances.

  11. Advanced methods: Use an xG model and a Poisson distribution to simulate the score distribution and compare the resulting probability with implied probability, or use a machine-learning model to predict residuals.

  12. Risk management, including the Kelly criterion

  13. Position sizing: Adjust the stake dynamically according to the edge, defined as your probability minus the implied probability. Never go all in. The Kelly formula is f = (bp - q) / b, where b = odds - 1, p is your estimated win probability, and q = 1 - p.
  14. Bankroll: A professional bankroll should contain at least 100-200 units, with each bet below 2-5%.
  15. Diversification: Spread bets across leagues and markets such as match result, handicap, and totals, and do not chase losses.

  16. Stop-loss and tracking: Record the EV of every bet, backtest the model regularly, and adjust its parameters.

  17. Data-driven iteration

  18. Sources include Opta or StatsBomb xG data, Elo ratings, and Betfair exchange prices for market depth.
  19. Backtest on at least five to ten seasons of data and use out-of-sample testing to prevent overfitting.
  20. Review how home advantage changes in each league every year.

Realistic expectations: Excellent quantitative bettors may achieve annual ROI of 4-10%, which is already exceptional, through discipline rather than a single factor. Strong home teams can form part of a portfolio, but using this factor alone is unlikely to sustain positive returns.

4. Possible improvements and warnings

  • Market selection: Home advantage can be stronger in lower-level leagues, though liquidity is lower, information asymmetry is greater, and risk is higher.
  • Other markets: Asian handicaps and secondary markets for corners or cards can sometimes be less efficiently priced.
  • Psychology and execution: Avoid confirmation bias, such as remembering only decisive home wins, and follow the model strictly.
  • Legal and sustainable practice: Ensure compliance and treat betting as a long-term business rather than gambling.

Summary: The strategy has a sound basis, but it cannot stand alone. Embed it in a comprehensive quantitative system, focus on positive-EV situations, and combine it with strict bankroll management and continuous improvement. The core of professional betting is edge + execution + time.

This answer raised my understanding of a simple four-word strategy to a level I could not previously have reached. There are countless other examples like it. I believe this is AI's greatest contribution to professional betting.