You Already Have a Model. It Is Just Untested.
Researched and drafted with AI assistance from published sources; published by LokeNessiSport Editorial Team · Last updated: 8 August 2026
18+ | Gambling involves risk. Only bet what you can afford to lose. If gambling is causing you problems, contact BeGambleAware.org or Hjelpelinjen: 800 800 40.
Disclosure: this page contains affiliate links. If you sign up through them we may earn a commission, at no extra cost to you. It does not change what we write.
18+ only. Betting carries a real risk of losing money and is not an income strategy. Set a deposit limit before you start, never chase losses, and stop if it stops being fun.
You have a model. You just have not written it down, so you cannot test it, so you cannot improve it, so you keep making the same mistake in a new fixture every weekend.
Every time you look at a match and think "they will win", you are running a model. It has inputs. It weights them. It produces an output. The only thing separating you from someone with a spreadsheet is that theirs is legible — and legible things can be fixed.
Step 1: find your real inputs, not your stated ones
Ask most people why they fancy a team and you get "they are in good form" or "they are the better side". Those are conclusions, not inputs.
The real inputs are usually a small set, and several of them are things you would not admit to:
- You say "underlying performance": what often actually drives it is the recent results you happened to watch
- You say "squad quality": often it is brand and reputation
- You say "tactical matchup": often it is a pundit's phrasing from Thursday
- You say "home advantage": often it is whether the fixture is on TV
- You say "fatigue and rotation": often it is rarely considered at all
That is not an insult. It is how everyone starts, including people who go on to be good at this. The fix is simply to write the list down honestly, because an input you cannot name is one you cannot test.
Advertisement · 18+ · Gambling involves risk · BeGambleAware.org
Step 2: state a probability, not a winner
"They will win" is untestable. If they lose, you were unlucky; if they win, you were right. Both readings survive any result, which is exactly the problem.
"They win about 60% of the time from this position" is testable. Over thirty of those, you should be roughly right 60% of the time — and if your 60s land at 40%, you have learned something concrete about yourself rather than about the match.
This is the single biggest step up available to anyone forming opinions about sport, and it costs nothing but honesty. The maths behind it is in expected value betting explained and how to read sports betting odds.
!Everyone in this stadium has a model. Almost nobody has written theirs down.
Step 3: separate what you know from what you assume
Three buckets, and being ruthless here is what makes the model useful:
Known. Team news confirmed, suspensions, fixture congestion, venue. Facts.
Estimated. Underlying performance, tactical fit, motivation. Judgements you can improve.
Assumed. "They always turn up in big games." "They cannot defend set pieces." Narratives you have never checked and probably absorbed from commentary.
Most bad predictions are an assumed input wearing a known input's clothes. Our full analysis method is in how to analyse football matches for betting.
> A model is not a machine that tells you the future. It is a machine that tells you which of your beliefs is doing the work — so you can find out whether that belief is any good.
Step 4: write it down before, in a form you cannot edit
Four fields per match, before kick-off, no exceptions:
The mechanism field matters more than the outcome. Being right for a reason that never happened is not a good prediction — it is a coin landing your way while teaching you something false. That is the whole subject of Part 3.
Advertisement · 18+ · Gambling involves risk · BeGambleAware.org
Where the market fits in
The odds are already a model — a very well-resourced one, with a margin on top. Your number is only interesting where it disagrees with theirs, and only if your disagreement is right more often than the margin costs.
That is a high bar and most disagreements are not real edges. Why the favourite wins and you still lose explains the margin mechanics, and how to find value bets covers the search.
!Your number is only interesting where it disagrees with the market's number
What this looks like after a month
Thirty written predictions is enough to see two things you cannot see any other way: whether your confidence is calibrated, and which kind of match you actually read well. Almost everyone has a cluster — one league, one sport, one situation — where they are genuinely better than average, surrounded by a lot of matches where they are not.
Finding the cluster and ignoring everything else is, for most people, the entire improvement available. Part 3 is the spreadsheet that finds it.
Worked example: scoring your first five written predictions
The probability field is what makes a prediction scoreable. The simplest score is the Brier score: for each call, square the gap between your probability and what happened (1 if it happened, 0 if not), then average. Five written calls:
- Home win at 60%, happened: (1 − 0.60)² = 0.160
- Home win at 60%, did not happen: (0 − 0.60)² = 0.360
- Away win at 70%, happened: (1 − 0.70)² = 0.090
- Draw at 55%, did not happen: (0 − 0.55)² = 0.303
- Over 2.5 at 80%, happened: (1 − 0.80)² = 0.040
Average: 0.953 ÷ 5 = 0.19. Saying 50% every time scores 0.25 on two-way calls, so lower is better and 0.19 is a decent start — on five calls, which proves nothing. The habit is what matters: by thirty calls the score starts to mean something, and by a hundred you can split it by league, market and confidence level to find the cluster where you are genuinely good.
Advertisement · 18+ · Gambling involves risk · BeGambleAware.org
If you want to see how machine-built models are put together and tested, how AI predicts sports results walks through Elo, Poisson and machine learning — and the three tests it uses (calibration, Brier score and closing line value) apply to your written predictions exactly as they apply to any algorithm.
The same exercise transfers to combat sports almost unchanged, where the sample per fighter is far smaller and the temptation to reason from narrative is far stronger — our sister site works through it in reading a fight: what "styles make fights" actually means.
Next: Form is a lie — what actually predicts the next match.
Where the partner links sit. If you want somewhere to record a price against your written number, we hold partner accounts with Thunderpick Sports and Verde Sports. Those are partner links and they are two books, not a comparison set — the whole point of Step 4 is that you should be checking more than one price before you act. Which countries either accepts is set by the operator's own terms and changes; check before you register. 18+, terms apply, your stake is at risk.
---
*Some links on this page are affiliate links and are marked as sponsored — where a partner pays us, our analysis does not change. Betting content is for readers aged 18 or over in jurisdictions where it is legal. Gamble responsibly; if it is causing harm, stop and contact a national support line.*
Frequently Asked Questions
What is a betting model, in simple terms?
A set of inputs you consistently use to estimate how likely each outcome is, plus a rule for how you weight them. Everyone who forms an opinion about a match already does this informally; writing it down is what turns it into something you can test and improve.
Do I need maths or a spreadsheet to build one?
Not to start. The first version is three or four named inputs and a stated probability, written before kick-off. A spreadsheet only becomes necessary when you want to measure calibration over a large sample.
Why does writing predictions down matter so much?
Because memory reliably rewrites confident wrong calls into ones you had doubts about. A written record before the event is the only version that survives contact with hindsight.
How do I score my predictions?
Use the Brier score: for each prediction, square the gap between your probability and the outcome (1 or 0), then average. Five calls at 60%, 60%, 70%, 55% and 80% with three of them coming true can score about 0.19, against 0.25 for always saying 50%. Lower is better, and you need dozens of calls before the number means much.