BetNumbers, tipster leaderboards and the survivorship problem
Tracking tipsters publicly and showing their records is a better idea than most of this industry manages. It also has a statistical flaw that no amount of honest bookkeeping can fix.
Updated · The Natybet Desk
BetNumbers publishes free football predictions across a very wide range of competitions and markets, combining algorithmic output with a tipster model where contributors post publicly and have their records tracked and displayed. Showing tipster records rather than hiding them is the right instinct, and it is more than most of this market offers.
The flaw is in the leaderboard itself, and honest bookkeeping cannot fix it. When many tipsters each make many predictions, some finish with outstanding records by chance alone. A leaderboard shows you whoever is on top and cannot tell you whether that is skill or luck, because the people who did badly are not on the screen. Quoted success rates around 65% mean nothing until you know the markets, the average odds and the number of settled picks.
Natybet is not affiliated with BetNumbers. This is an independent explainer written by us. We do not run BetNumbers, we are not paid by them, we cannot publish or change anything on their site, and we have no access to their model or their data. Everything below is our own reading of what they publish.
The maths of a leaderboard with no skill in it
This is the most useful thing on this page, and it applies to every tipster leaderboard, WhatsApp group and Telegram channel you will ever see.
Imagine a hundred tipsters. Give none of them any ability whatsoever: each one picks at random on twenty events that are genuine coin flips. On average each finishes on ten correct out of twenty, which looks exactly like the nothing it is.
But that is the average. The spread is what matters. With a hundred people making twenty random picks, the best performer will typically land somewhere around fifteen or sixteen correct. That is a strike rate of 75 to 80%, produced entirely by chance, by someone with no ability at all. Somebody has to come top, and coming top is what luck looks like from the front.
Now put that person at the head of a leaderboard, and understand what the reader sees: a public, verifiable, honestly maintained record showing 80%. Every number on it is true. The record is not fabricated. And it still carries almost no information about what that tipster will do next, because the process that put them at the top was selection, not skill.
Why this is not an accusation
Nothing above requires anyone to be dishonest. That is precisely what makes it worth understanding. A site can track every tipster scrupulously, publish every result, delete nothing, and still present a leaderboard whose top entries are mostly noise. The bias is in looking at the winners, and it arrives before anybody has done anything wrong.
It is the same mechanism as the fixed-match numbers trick, where a seller messages a large group with different predictions and then markets to whichever subgroup happened to receive the correct one. The difference is intent. In the scam version the selection is deliberate; on a leaderboard it happens by itself. The statistics are identical, which is why the same defence works against both: ask how many predictions, made by how many people, over what period, before reading any strike rate as evidence.
What would make a tipster record meaningful
- Volume. A few hundred settled picks starts to be informative. Twenty is noise, fifty is barely better. The exact number depends on the edge you are trying to detect, and the smaller the true edge the more picks you need to see it.
- Prices attached. Hit rate alone cannot be judged. A tipster on 65% at average odds of 1.35 is losing money; one on 45% at 2.60 is making it. Without odds recorded, the leaderboard is sorting on the wrong column entirely.
- Consistency across periods. Skill persists, luck does not. A record that was strong last quarter and strong this quarter means considerably more than one strong run, and a leaderboard sorted on all-time totals hides exactly that distinction.
- A denominator you can see. How many tipsters are on the platform? The more there are, the more impressive the top record will look for purely mechanical reasons, and the less it means.
Apply all four to our own record as well. It is currently very short, which means it is currently weak evidence, and we would rather say so than let a small sample of early results stand in for a claim we have not earned yet.
Reading the quoted success rate
Figures in the region of 65% are commonly quoted for services in this category. The number is not implausible, and that is exactly why it needs unpacking rather than dismissing.
A 65% strike rate is easy to manufacture without any forecasting ability by choosing safe markets. Double chance covers two of three outcomes, so a competent selection there lands far more often than not. Over 0.5 goals in a match lands about nine times in ten. Neither produces money, because the prices reflect exactly how likely they are. Meanwhile a genuinely skilled forecaster working on 1X2 in competitive leagues might sit closer to 50% and be far more valuable.
So the strike rate on its own tells you about market selection rather than about ability. Ask which markets it was earned in, and what the average price was, and the figure either becomes readable or reveals that it was never meant to be.
What the site is genuinely good for
Wide free coverage and a large set of markets, presented without a paywall in front of the basics. If you want a second opinion on a fixture, or a quick read on where a market sits before you look at prices, that is a real service.
Use the tipster records as a starting point rather than a ranking. Look for volume and duration rather than headline percentage, discount anybody whose record is short however good it looks, and convert everything into implied probability before you decide anything. The leaderboard is showing you the survivors of a process it cannot fully see, and so are you.
Frequently asked questions
What is BetNumbers?
A free football prediction site covering a very large number of leagues, with tips across 1X2, both teams to score, correct score, double chance, half-time and full-time and over or under 2.5 goals. It combines algorithmic output with a tipster model in which contributors post predictions publicly and have their records tracked and displayed.
Is a tipster leaderboard a good way to pick tips?
It is much better than nothing and much weaker than it appears, for a reason that has nothing to do with honesty. When many people each make many predictions, some will finish with excellent records through luck alone. A leaderboard sorts for whoever is currently at the top, and it cannot distinguish the genuinely skilled from the temporarily fortunate.
How much of a top record is luck?
More than most people expect. If a hundred tipsters each pick twenty coin-flip events, the best of them will typically land around fifteen or sixteen right, which looks like a 75 to 80% strike rate and is pure chance. The leaderboard will show that person at the top with a record that looks like evidence of skill, because the ninety-nine others are not what you are looking at.
Is the quoted success rate reliable?
Success rates in the region of 65% get quoted for this and many similar services. Any such figure needs three things attached before it can be read at all: which markets it covers, what the average odds were, and over how many settled picks. A 65% rate on short-priced double chance selections is unremarkable and can lose money steadily.
Should I pay for the VIP tier?
Our general position on paid tips is that the arithmetic does not work: a paid selection must beat the bookmaker's margin and then also recover its own cost before you are ahead. On top of that, the same survivorship problem applies to whichever tipster the subscription is built around.
How would I know if a tipster is genuinely good?
Sample size and prices. A few hundred settled picks with the odds recorded, showing a return above stake, is meaningful. Twenty or fifty picks is not, however impressive it looks, and a percentage with no odds attached cannot be assessed at all. Judge us on the same basis.