Beyond the odds: what the 2026 World Cup taught us about bettors, AI and prediction markets

The 2026 World Cup delivered 308 goals across 104 matches, the highest scoring rate since 1970, and results that challenged supporters and analysts. Cape Verde reached the knockout stage unbeaten in regular time, while Norway eliminated Brazil.

Yet the semi-finalists were Spain, France, England and Argentina – the four highest-ranked teams in the world. That apparent contradiction captures an important truth about sports betting: individual matches are unpredictable, but strong models still reveal the broader pattern.

For GGBET UA, the tournament was both an infrastructure stress test and the backdrop to our major communications campaign, ‘Big Game Summer’. We were among Ukraine’s most active bookmakers in terms of events, betting markets and promotional activity. Despite Ukraine’s absence and late kick-offs in Kyiv, our busiest matchdays were surpassed in daily activity only by Oleksandr Usyk fights.

An upset is not a broken model

As I told Ukrainian business outlet LIGA.net in a recent interview, if every prediction came true, we would not watch football. An underdog victory does not necessarily mean the model was wrong. If a team has a 12%-15% chance of winning and then wins, it has landed within the range the model identified.

What mathematics cannot predict is when a red card, rebound or exceptional performance changes a match. Models understand probability, but they cannot know how a key player will perform that day.

Reputation can also distort customer judgement. A team’s initial rating reflects its history, brand and past performance, but current form moves faster than reputation. That was visible with Brazil and Germany, which bettors tended to overrate early in the tournament, while Cape Verde and Norway were initially underestimated.

Analytics made the greatest difference when there was a clear gap between the teams. During the group stage and round of 32, 56%-57% of bets were winners. That fell to around half during the round of 16 and quarter-finals, 38% in the semi-finals and only one in four for the final. The more evenly matched the teams became, the less decisive the available data was.

More football changes the rhythm of betting

The expanded format produced 104 matches, compared with 64 in 2022. The group stage featured 72 fixtures in 17 days, creating concentrated peaks almost daily.

Compared with Euro 2024, which had 51 matches, the number of bets on our platform increased 234%, while bets per player rose 186%. The challenge was not simply processing more activity, but managing repeated spikes in demand. GGBET UA offered an average of 104 active markets per match, rising to 256 for the final.

The importance of a match mattered as much as the teams involved. Betting activity was 14% higher in the round of 32 than in the group stage, rising to 49% in the round of 16, 61% in the quarter-finals and 180% in the final.

By contrast, the third-place play-off between France and England attracted activity comparable with the round of 32. Switzerland versus Colombia ranked among our six most-bet-on matches because elimination was on the line.

Local context can overturn global patterns

Late kick-offs created one of the clearest behavioural shifts. Pre-match wagers accounted for 62% of tournament betting volume in Ukraine. Under normal conditions, they represent 30% to 35%, with live betting generating the majority.

Many customers placed their bets in the evening, went to sleep and checked the result the next morning. For the final, however, they stayed awake and pre-match and live betting were evenly split. Product and trading strategies therefore cannot rely solely on global tournament trends; scheduling and local habits can materially change the customer journey.

AI works best when it solves a defined problem

Computer vision can extract player positions, speed and other data directly from broadcasts, supporting trading without physical sensors. Recommendation models help customers navigate thousands of daily events, while AI also contributes to customer support, anti-fraud and software development.

Autonomous coaching, reliable injury prediction and real-time tactical advice remain closer to promise than widespread practice. The practical lesson for operators is to judge AI by measurable utility, not by the size of the claim.

Prediction markets are becoming real competitors

Prediction platforms can no longer be treated as a niche. In the US, sports reportedly account for around 80% of activity on Kalshi and 39% on Polymarket. Their exchange model offers flexibility because users set the price and provide liquidity.

Traditional sportsbooks still hold important advantages: broader market depth, familiar payment journeys and established promotional mechanics. Regulation is another constraint, as many jurisdictions treat these services as gambling products requiring local licences.

The World Cup showed that technology does not remove uncertainty from sport. It helps operators price it, manage it and understand how customers respond to it. The winners will be those that combine robust models and infrastructure with an equally strong understanding of local behaviour, while recognising when a new competitor is becoming too significant to ignore.