Oleksandr TverdokhliebA while ago I started treating our office World Cup prediction pool as a small forecasting problem rather than a game. Most people fill in their picks with their gut — the team they like, the scoreline that feels right. I did something duller: for every match I worked out which prediction scored the most points on average, submitted that, and didn't change my mind after a bad round. This year that approach won the pool, and it turned out to be one of the cleaner illustrations of expected value I've run into.
I manage a data team, so my instinct with any game like this is to ask what's actually being scored and then optimise for that, rather than for how the picks feel. It's the same instinct I use at work, and the football version is a good place to see it clearly, because the scoreboard settles every argument at the end.
The game, and why the gut loses
The rules were simple. Before each of the 104 matches you submit one scoreline. You get 3 points for the exact score, 1 point for the right outcome (correct winner or a draw, wrong score), and a one-off 6 points if you name the eventual champion. Nothing else counts. Crucially, the scoring is flat: a boring, near-certain result is worth exactly as much as calling a wild upset. Everyone in the pool sees the same fixtures; the only thing that varies is judgement.
Most people play it as a fan. They back their favourites, they talk themselves into a shock or two, and after a bad weekend they start changing their whole method. None of that is optimising for the points — it's optimising for how the picks feel while you make them. Which is fine for fun, and quietly expensive if you're trying to win.
Expected value, in plain terms
Expected value is just the reward for an outcome multiplied by how likely it is. If a pick can score in more than one way, you add up each way. That's the entire idea, and it's the same arithmetic an insurer, a poker player or an analyst pricing a decision runs — only here the payouts are printed in the rules.
A football scoreline pays two ways at once: 1 point if you merely got the winner right, plus the bigger 3 if you nailed the exact score. So the value of any scoreline is the chance of the right outcome, plus the extra reward when the exact score lands. Written out for this game it becomes a short formula:
To get the probabilities I leaned on the one place that prices football for a living: the bookmakers' correct-score market. Their odds bake in a margin, so you strip that out to recover a fair probability for each scoreline, then pick whichever score has the highest expected points. It rarely feels exciting — it often lands on a tight, sensible scoreline — but excitement was never what was being scored.
Volume is where the edge lives
A small per-match edge doesn't look like much on any single game. Across a hundred of them it compounds into a lead that's hard to catch. The final ledger made that obvious — most of the points came early, while the field was wide:
Two-thirds of the winning total was scored in the group stage — the boring, high-volume part everyone else treated as a warm-up. That's the law of large numbers doing quiet work: when you make the same slightly-better-than-average call again and again, the average is where you get paid. It's exactly why I trust a well-instrumented pipeline that runs five hundred small jobs a day over a heroic one-off analysis. The edge is in the repetition.
Booking losses on purpose
Playing the percentages means accepting guaranteed zeros. A heavy favourite goes out; a shootout swings the wrong way; you score nothing. Two whole knockout rounds scored me a flat zero. None of those were mistakes. A method that could dodge shocks would be a method that predicts them, and that method doesn't exist. You take the blank rounds on purpose because the boring picks around them more than pay for the misses.
What it actually transfers to
The football is a toy, but the shape of the decision is not. Most analytics work is a stream of small bets under uncertainty: which leads to prioritise, which segment to target, where to spend the next hour of engineering. You almost never get certainty. What you can get is a reasonable probability and a clear view of the payoff, and those two things are enough to make the expected-value call instead of the gut one — the same move as picking 2-1 over the scoreline that merely feels lucky.
The discipline is the hard part, not the maths. The maths is a multiplication. The discipline is continuing to make the same slightly-favourable call after a run of bad luck makes it feel stupid, and not chasing a shock because it would feel clever to have called it. Most people can do the arithmetic. Far fewer keep doing it once the scoreboard turns against them for a week.
The honest caveat
None of this guarantees a win, and I'd distrust anyone who told you it did. Going into the semi-finals I was around 65% to take the pool; the moment my projected champion lost, that dropped to roughly 10%, and it took a specific result on the final night to pull it back. A good method tilts the odds firmly in your favour and keeps them there — over enough games you finish ahead far more often than not — but variance always gets a vote in any single tournament. Expected value is a claim about the long run, not a promise about tonight.
That's the part worth carrying into real decisions. The goal isn't to be right every time, which is impossible; it's to be pointed the right way, consistently, so the long run does the work. I wrote up the full system that ran the picks — the automation, the pipeline and the agent that operated it for 39 days — in the machine behind the picks, and the finished result is public at wc26.devapp24.com.