What is the SwiftScore Engine?

The SwiftScore Engine is our AI football prediction model. It combines a goals model, a team Power Rating and a machine-learning layer into one calibrated forecast for every match, then summarises its confidence as a single 0–100 SwiftScore.

How is a SwiftScore calculated?

Each part of the engine forms its own view of a match. The engine blends those views — weighting whichever has been most accurate lately — and reads off the result probabilities. The SwiftScore expresses how likely and how clear-cut the leading call is, from 0 to 100.

Are SwiftScore predictions guaranteed?

No. They are statistical projections, not guarantees or betting advice. Football is unpredictable and no model wins every time. We publish our full record — wins and losses — so you can judge the engine on real evidence.

How do you grade your predictions?

Every pick is logged before kick-off and can never be edited afterwards. Once the match finishes it is scored automatically against the real result and added to our public accuracy record, broken down by league and by confidence.

The SwiftScore Engine · how it works

How the SwiftScore Engine works

No crystal balls and no insider tips — just a transparent, AI-driven model that forecasts every match, shows its reasoning, and is scored against the real result in public. Here is exactly how it does that, in plain English.

The ensemble
The SwiftScore
Our metrics
What we do & don't claim
Public grading
Calibration
Data & limitations
Responsible use

One prediction, three points of view

The SwiftScore Engine is an ensemble: rather than trusting a single method, it lets three independent models look at the same fixture and then blends their views. Each one is good at something different, and combining them is more accurate and more stable than any one alone.

A goals model

Estimates how many goals each side is likely to score from their attacking and defensive strength, adjusted for home advantage. It produces a full picture of likely scorelines — not just who wins, but how the goals fall.

A Power Rating

A running measure of each team's underlying strength, updated after every result and weighted by the quality of the opponent. It is slow to be fooled by a lucky win or a freak defeat.

A machine-learning layer

Trained on thousands of past matches, it learns the subtler patterns the first two miss — and helps the engine read situations where the obvious signals disagree.

One calibrated forecast + a SwiftScore

The engine combines the three views into a single set of probabilities, then summarises its confidence as one SwiftScore.

The blend isn't fixed — it learns

The engine continuously tracks how accurate each part has been on recent matches and tilts the blend toward whatever is working. It also leans gently on the market's own pricing as a sanity check. So the model you see today is the model that has been earning its place lately — not a setting we picked once and forgot.

What a SwiftScore actually means

The SwiftScore is our signature 0–100 confidence number, shown on every prediction. It answers a simple question: how strong is this call? A high SwiftScore means the engine's parts agree and the match looks clear-cut; a low SwiftScore is the model being honest that a fixture is a genuine coin-flip.

Crucially, a SwiftScore is not a promise. A SwiftScore of 75 doesn't mean the pick will land — it means that, across many similar calls, picks like this tend to come in around three times in four. That is why we publish our calibration: so a number you see here means the same thing it means in reality.

You'll see the highest-SwiftScore selections surfaced on the homepage and in the screener, where you can filter and sort every upcoming prediction by confidence, market and value.

The proprietary metrics you'll see

Alongside the SwiftScore, the engine surfaces a small set of our own metrics so you can understand why a prediction leans the way it does. Each one carries an in-page explainer wherever it appears.

SwiftScoreOur engine’s overall confidence in this call, 0–100. It blends how likely the pick is with how clear-cut the match looks — higher means a stronger, more decisive call.
Power RatingHow strong a team is right now on our scale, built from results, opponent quality and home advantage. Compare two Power Ratings to see who the model rates higher.
Form IndexHow a team has been performing across its last five matches. Higher means hotter recent form; lower means a team that’s been struggling.
Model ConsensusHow much our underlying models agree on this pick. High consensus means every model points the same way; low consensus flags a genuinely uncertain match.

For the cross-site standards everyone knows — home/draw/away chance, Over/Under 2.5, both teams to score and the most likely correct score — we keep the familiar names so you can compare us to anyone.

What we do — and don't — claim

Trust starts with being clear about the limits. Here is exactly where we stand.

✓ What we do

Forecast every match with a transparent, AI-driven model.
Show the reasoning behind each pick, in plain English.
Log every prediction before kick-off and grade it in public.
Publish our hit rate and calibration — wins and losses alike.
Keep improving the model as new results come in.

✗ What we don't

Promise winners, "sure things" or guaranteed returns.
Offer betting, financial or investment advice.
Suggest predictions are a source of income.
Delete losing picks or quietly edit our record.
Take stakes or pay out — we are not a bookmaker.

Every pick, graded in public

A prediction model is only worth as much as its track record, so we publish ours and let it stand or fall on the evidence. The moment the engine makes a call, it is written down with the confidence and the odds at the time — and that record is immutable. We can't go back and pretend a losing pick never happened.

When a match finishes, the pick is scored automatically against the real result and rolled into our public numbers. So far that's a 52% match-result hit rate across 6,705 settled picks, rising to 69% on our high-confidence calls.

You can inspect all of it — by league, by strategy and pick by pick — on the results page.

Calibration: do our probabilities tell the truth?

Accuracy alone can be gamed by only ever predicting favourites. The harder, more honest test is calibration: when the engine says something is 70% likely, does it actually happen about 70% of the time? Almost no rival publishes this. We do.

Our current Brier score — the standard measure of how honest a probability forecast is — is 0.234, where lower is better and 0.25 is a coin toss. In plain terms: a SwiftScore means roughly what it says it means.

The full calibration table, showing what we predicted versus what actually happened in each confidence band, lives on the results page.

What the engine reads — and what it can't

Data sources

Historical resultsThousands of past matches across the competitions we cover, which train and continually update the model.
Fixtures & team dataUpcoming schedules, teams and venues from established football data providers.
Market pricingBookmaker odds, used purely as a sanity check on the engine's own view — never as the prediction itself.

Honest limitations

No model sees everything. The engine works from data, so it can't know about a last-minute injury, a rotated cup line-up, a managerial bust-up or the weather turning a game into a lottery. It is strongest in well-covered leagues with plenty of history, and least certain in noisy, low-data fixtures — which is exactly why a low SwiftScore exists. Football's unpredictability is a feature of the sport, not a bug we can engineer away.

Using this responsibly

Our predictions are for information and entertainment. They are not betting tips, and they should never be treated as a way to make money. If you choose to gamble, only ever stake what you can comfortably afford to lose, and never chase a loss.

You must be 18 or over to view odds-based content on SwiftScore. Free, confidential support is always available: BeGambleAware.org and GamCare, plus our own responsible gambling page. When the fun stops, stop.

Want the numbers? Head to the results & calibration page, browse today's predictions, or read more about who runs SwiftScore.

Frequently asked

What is the SwiftScore Engine?: The SwiftScore Engine is our AI football prediction model. It combines a goals model, a team Power Rating and a machine-learning layer into one calibrated forecast for every match, then summarises its confidence as a single 0–100 SwiftScore.
How is a SwiftScore calculated?: Each part of the engine forms its own view of a match. The engine blends those views — weighting whichever has been most accurate lately — and reads off the result probabilities. The SwiftScore expresses how likely and how clear-cut the leading call is, from 0 to 100.
Are SwiftScore predictions guaranteed?: No. They are statistical projections, not guarantees or betting advice. Football is unpredictable and no model wins every time. We publish our full record — wins and losses — so you can judge the engine on real evidence.
How do you grade your predictions?: Every pick is logged before kick-off and can never be edited afterwards. Once the match finishes it is scored automatically against the real result and added to our public accuracy record, broken down by league and by confidence.