How good can football predictions actually get?

Short answer: not as good as you'd think. The best models in the world land at ~55% on outcomes. We're at 52.4%. Here's why nobody — not Pinnacle, not Sky's super-genius algorithm, not your mate down the pub — beats that ceiling, and what it means when you're picking between Safe and Spicy.

18+ | Gamble responsibly | Odds may change | No guarantee of profit · More info

The accuracy ceiling

Football has noise baked in. Red cards, deflections, terrible refereeing, a goalkeeper having one in the day, a striker putting it in row Z. None of that is predictable from form, xG, or league position. It's the genuinely random ~45% of every match.

That random core caps how good anyone can be:

Closing-line bookies (Pinnacle etc.) ~55%
Best academic models (DC, Bivariate Poisson, ML) ~52%
Us right now (3319 backtested matches) 52.4%
"Always back the home team" 43.0%
Random guess (Home / Draw / Away) 33%

We're a couple of percentage points off the bookies' closing line. That's about as close as a free-to-read model can realistically get without millions of pounds of liquidity squeezing the price.

Super 6 — why £250,000 and nobody wins

Sky's Super 6 asks for six exact scorelines across one weekend. £250,000 sounds nuts. The maths says it's actually about right.

~8%
Our exact-score hit rate
per single match
~10–12%
Best models in the world
per single match
1 in 5,042,900
All six exact at our level
~5.0M to one
1 in 564,474
All six exact at world-best
~0.6M to one

Now compare to just getting the outcome right (Home / Draw / Away) for six matches in a row:

1 in 48
Six outcomes at our accuracy
52% per match
1 in 36
Six outcomes at bookmaker accuracy
~55% per match

That's the gap between picking winners (hard but doable) and picking scorelines (essentially a lottery with a tiny edge).

Safe vs Spicy — what we actually mean

Every match has a whole distribution of possibilities, not a single answer. The model doesn't say "Liverpool win 2-1." It says "Liverpool win is the most likely outcome at 52%, but a draw is 24%, an away win is 24%, and the most likely scoreline within a Liverpool win is 2-1."

Safe = the centre of that distribution. The model's most likely outcome with the most likely scoreline inside it. This is what we bet our reputation on.

Spicy = the highest-value alternative once we compare the model to live bookmaker odds. We score every non-Safe outcome by edge — model probability × bookie decimal − 1 — and pick the biggest positive number. When the bookies have priced an outcome more generously than its true probability deserves, that's the Spicy bet. If no outcome has a positive edge (or no odds are available for the fixture), Spicy falls back to the next-most-likely outcome from the model.

🎯

Safe

The boring middle. Most likely outcome, most likely scoreline. Hits often, pays small.

🌶️

Spicy

The biggest +EV bet vs the bookies — where the price exceeds the model's fair odds by the largest margin. Lower hit rate, but the price compensates when it lands.

Both come from the same model. Spicy isn't a guess — it's the outcome where bookmakers and our probabilities disagree most. Want to see the full distribution for any fixture? Hit the 🌶 Spicy button on any picks card and drag the slider from safe to spicy.

The Spicy Slider — now per-fixture

Each picks card on the home page now has a 🌶 Spicy button next to the team names. Click it to open the full scoreline distribution for that fixture, drag from safe to spicy, see live odds + plain-English caption for any score the model thinks is possible.

← Try it on the picks page

The boring numbers, if you care

Outcome accuracy52.4%vs 55% bookie ceiling
Brier score (3-class)0.1980lower is better · 0 = perfect
Exact-score hit rate7.6%vs ~4% random reasonable
Always-home baseline43.0%the gut-feel benchmark
Backtested matches3,319walk-forward, no peeking · top-5 European leagues
← Back to picks