Why Being Right 70% of the Time Can Still Lose You Money (And Being Right 26% Can Make a Fortune)
Our AFL model is right 71% of the time and loses money. Our NRL try scorer model is right 26% of the time and prints profit. Here is why.
Daniel Pham
Quantitative Strategy Lead
7 min read·Published 10 Aug 2026
So I spent the last few weeks backtesting a bunch of betting models across AFL, NRL, tennis, and baseball. Thousands of historical games, tens of thousands of bets. And I found something that kind of broke my brain a little.
Being right most of the time can still lose you money. And being right barely any of the time can make you a fortune. I know that sounds like clickbait but stick with me because the numbers are genuinely wild.
OK so check this out
I built an AFL model that picks the right winner 71% of the time. That sounds insane right? Like if someone told you they tip 7 out of 10 correctly you would think they are printing money.
Nope. It loses money. -14.3% ROI to be exact.
Meanwhile our NRL anytime try scorer model is right about 1 in 4. Twenty-six percent. That sounds terrible. But it makes +67% ROI. Sixty-seven percent return on investment.
The AFL model is right 7 out of 10 times and still loses money. The NRL try scorer model is right 1 in 4 and makes a killing. How the hell does that work?
Here is the thing nobody tells you
Being accurate and being profitable are two completely different things. And I know that sounds obvious when you say it out loud but most punters (myself included before this) think in terms of "am I picking the winner?" when they should be thinking "am I getting the right price?"
The bookies are really good at their job. When a team wins 70% of the time, the odds already reflect that. You are paying full price for something everyone already knows. There is no edge in backing a $1.40 favourite that wins 70% of the time — you need them to win more than 71.4% just to break even after margin.
But say you find a bloke who scores a try 26% of the time, and the bookmaker is paying $5.79 like he only does it 17% of the time. That is a genuine mispricing. You do not need him to score every week. You just need him to score more often than the odds imply. And at $5.79 he only needs to score 1 in 4.3 times for you to break even. He scores 1 in 3.8. That gap is where the money is.
The bit that actually matters
I tested what happens if you just always back the favourite across 233,000+ races. Strike rate: 42%. ROI: -26%. You win less than half the time and the odds do not make up for it. It is a slow bleed.
So what actually works? I ran a threshold sweep — basically asking "if we only bet when the model thinks it has at least X% edge over the market, what happens?" Here is the answer:
EV Threshold
Bets
Win Rate
ROI
No filter
106
63.2%
+47.5%
≥ 3%
83
66.3%
+58.1%
≥ 7%
61
68.8%
+75.3%
≥ 10%
49
69.4%
+69.8%
≥ 15%
28
67.9%
+60.7%
The sweet spot is about 7%. Below that you are taking noise bets. Above 15% you are passing up good opportunities because the sample gets too small. At 7% you are making 75 cents on every dollar and still getting enough bets to make it real.
What I took away from this
Look I am not going to pretend I have all the answers. But after staring at these numbers for a few weeks here is what I think matters:
Win rate is a vanity metric. A 55% model that finds value beats a 70% model that does not.
When the model says 7%+ edge, bet it — even when it feels like a longshot.
One good bet does not make you rich. Hundreds of +EV bets does.
You are not trying to pick winners. You are trying to find prices the market got wrong.
The bit I keep coming back to
Our tennis model is 63% accurate. Sounds average right? It is +19% ROI across 28,500 matches. Because it knows when to bet, not just who wins.
The best punters I know are not the ones who pick the most winners. They are the ones who find the most value. And honestly that distinction is the whole game.
If you want to dig deeper into the math we have a guide on positive EV betting that walks through the concept properly. But honestly the table above says most of what you need to know.
All data from our own backtesting. Historical data, no look-ahead bias. Past performance does not guarantee future results. Gamble responsibly.
About the author
Daniel Pham
Quantitative Strategy Lead
Daniel writes about the maths underneath advantage betting — expected value, Kelly sizing, closing line value, bankroll theory. Translates the theoretical side into practical decisions AU punters can actually apply.
Analysis14 min
Prediction Markets vs Bookmakers: Where's the Edge in AU?
Prediction markets like Polymarket and Kalshi are growing fast. But do they offer better odds than Australian bookmakers? We measured 100+ markets to find out.
