France-Spain, the final as seen by Goldman Sachs and BofA

Every four years the ritual repeats itself: the world's leading investment banks practice forecasting the economic impacts of the World Cup (from advertising expenditure to tourism, from new jobs created to GDP growth), without forgetting to leave room for levity, with the prediction of the nation that will win the trophy.

Typically, the latter process is carried out through the use of internally developed models for assigning probabilities on the results of each individual match, predicting the different crossovers and thus arriving at the final.

For example, Goldman Sachs cross-references the historical performance data of each team to determine the number of goals scored (also using systems originally designed to rank chess players), then simulates a series of probabilities that a particular team will reach a particular round, all the way to victory. These predictions are flexible, because after each game day, the model is run again using the updated performance data to generate new probabilities.

This year, the experience is enriched with the use of artificial intelligence. Bank of America, for example, 'fed' Copilot a kilometre-long prompt that begins: 'You act as an elite football analyst and tournament prediction expert. Your task is to predict the most likely winner of the 2026 Fifa World Cup by evaluating all teams participating in the tournament'. This is followed by specific instructions regarding the evaluation of the teams, the recommendation not to focus on reputation alone as much as on the state of form, technical factors and other elements that are usually not very quantifiable, such as team cohesion or the mental strength of the players.

In football, the statistical power of any probabilistic prediction model remains limited, by the very admission of the analysts who designed them. Will AI be able to do better in predicting the outcomes of a series of football matches, which is about as human, imperfect and random as it can get?