- With 100 bets at odds near 2.00, ROI has a margin of error of ±19.6 points (95%).
- To see a real 5% edge you need on the order of 1,500 to 2,500 bets.
- A tipster or a model is judged by its verifiable sample, not by a streak or screenshots.
How many bets do you need to know if you are winning?
More than you want.
That is the honest answer.
And in this article we put it into numbers.
Imagine you are 100 bets in and ahead.
You feel good.
The question is whether that is skill or a streak.
To answer it there is a simple tool: the standard error of ROI.
It tells you how far your measured result can drift from the true result, by chance alone.
The examples with odds of 2.00 and 1.90 and with edges of 2%, 5% or 10% are assumptions to illustrate, labeled as examples.
They are not any bettor's results.
Educational content for adults 18 and over.
What the standard error of ROI is
ROI is what you win divided by what you stake.
If you stake 1 unit on every match, it is the average profit per bet.
That average is noisy.
Each bet is all or nothing.
You win odds minus 1, or you lose 1.
That big swing makes the average take a long time to settle.
The formula: the standard deviation of one bet divided by the square root of N.
At odds of 2.00 and a real 5% edge, the standard deviation of one bet is almost 1 unit (99.9%).
That is why, over 100 bets, the standard error is about 10 ROI points.
Table: the margin of error by sample size
We calculated the standard error for a 1-unit bet with a real 5% edge.
The last column is the margin of the 95% confidence interval: your measured ROI plus or minus that number.
| Bets (odds 2.00) | Standard error | 95% margin |
|---|---|---|
| 50 | 14.1 points | ±27.7 points |
| 100 | 10.0 points | ±19.6 points |
| 200 | 7.1 points | ±13.8 points |
| 500 | 4.5 points | ±8.8 points |
| 1,000 | 3.2 points | ±6.2 points |
| 2,000 | 2.2 points | ±4.4 points |
| 5,000 | 1.4 points | ±2.8 points |
| 10,000 | 1.0 points | ±2.0 points |
At odds of 1.90 the result is similar, a little smaller.
Over 100 bets, the standard error is 9.4 points and the margin ±18.5.
Over 1,000 bets, 3.0 points and ±5.9.
Read the 100-bet row.
A margin of ±19.6 points means a real 5% ROI can be measured as anything from +25% to −15%.
In practice, you cannot tell anything apart.
How many bets you need depending on the real edge
Now the inverse question.
If your real edge is a certain size, how many bets do you need so that chance does not cover it up?
We calculated two numbers.
The minimum: the measured ROI matches the real one and the 95% interval sits just above zero.
And the realistic one: having an 80% probability of detecting the edge (one-tailed test, 5%).
| Real edge | Odds 2.00 minimum | Odds 2.00 with 80% | Odds 1.90 minimum | Odds 1.90 with 80% |
|---|---|---|---|---|
| 10% | 380 | 612 | 338 | 544 |
| 5% | 1,533 | 2,467 | 1,371 | 2,207 |
| 2% | 9,600 | 15,452 | 8,621 | 13,875 |
See how the table drops.
Doubling the edge cuts the sample to a quarter.
A small edge demands an enormous sample.
Think of it in time.
If you place 5 bets a week, 1,500 bets is almost six years.
That is why almost no one can prove their edge from their own results alone.
Chance makes champions too: an example with tipsters
Suppose no tipster has any edge at all.
They all win 50% at odds of 2.00.
Over 100 bets, some will shine anyway.
We calculated the exact probability of winning 60 or more out of 100 with a 50% hit rate.
It comes to 2.84%.
That is equivalent to a +20% ROI at odds of 2.00.
Now imagine 100 tipsters with no skill whatsoever.
The probability that at least one reaches 60 wins is 94.4%.
One of them will show spotless screenshots.
A 55% hit rate over 100 bets (+10% ROI at odds of 2.00) happens by pure chance 18.4% of the time.
One in every five or six sequences with no skill at all looks like that.
That is why one good month proves nothing.
How to judge a tipster with these numbers
With the above you now have a yardstick.
It works for any tipster, channel or group that shows you results.
- Count the bets: fewer than 500 is anecdotal. Fewer than 100 says nothing.
- Ask for the odds: without the odds posted before the match there is no way to calculate ROI.
- Ask for the full record: tipsters who delete their bad bets look much better.
- Compare them with chance: does their ROI fall outside the margin of error in the table above?
- Distrust screenshots: a screenshot shows one day, not a sample.
In the article on how to spot a fake tipster you will find concrete warning signs.
If you want to compare tipsters with models, see AI vs tipster.
And Gambeta's record: what the numbers say
If we demand rigor from others, we apply it here.
We calculated the ROI of the public record with code, and we explain how.
Data: the public record file, cut off on October 28, 2026.
There are 1,594 picks since March 21, 2026.
We removed 33 pending and 22 voided.
That leaves 1,539 settled picks.
They hit 55.6%, with average published odds of 1.83.
The calculation stakes a flat 1 unit per pick, at the published odds, with no commissions.
| Record data (cutoff Oct 28, 2026) | Value |
|---|---|
| Settled picks (won or lost) | 1,539 |
| Hit rate | 55.6% |
| ROI with a flat 1 unit per pick | −0.02% |
| ROI standard error | 2.4 points |
| 95% confidence interval | −4.7% to +4.7% |
The reading is clear.
With this sample you cannot claim that the model either wins or loses.
Its ROI is indistinguishable from zero, and also from +4%.
This is not a final conclusion.
The record keeps growing, and the calculation assumes a flat stake and the published odds, not the best available odds.
You can check it yourself in the public record.
What to do if you cannot wait for thousands of bets
There are ways to speed up the signal.
They are not foolproof, but they help.
- Measure against the close: if your odds are usually better than the closing line of an efficient market, that is a faster signal than the result. We explain it in odds movement.
- Use probabilities, not just results: compare your estimated probability with the fair price.
- Keep a clean record: a tracking spreadsheet with odds, stake and result for every bet.
- Bet small stakes in the meantime: the sample builds up without risking your bankroll.
To calculate the expected value of each bet, use the EV calculator.
To understand the streaks that show up along the way, read losing streaks: how many are normal.
What to read next
Now you know a small sample misleads.
The next step is understanding what information the market carries and why the close is a useful reference.
Continue with odds movement: what it means.
If you want the full picture, the AI for betting guide brings it all together.
Bet only what you can afford to lose, and if you stop enjoying it, see the responsible gambling guide.
Frequently asked questions
How many bets do I need to know if my method works?
It depends on the edge you want to detect.
For a real 5% edge at odds near 2.00, between 1,500 and 2,500 bets.
For 2%, between 9,600 and 15,500.
These are calculations from this article, with their assumptions.
Are 100 bets enough?
No.
At odds of 2.00, the ROI margin of error over 100 bets is ±19.6 points with 95% confidence.
You could be earning 15% or losing 15% and not realize it.
What is the standard error of ROI?
It is how much your measured ROI typically varies from the true one by pure chance.
It is calculated as the standard deviation of one bet divided by the square root of the number of bets.
The more you bet, the smaller it gets.
How do I know if a tipster is good?
Look at how many verifiable bets they have, at what odds, and whether they posted them before the match.
A good percentage over 50 bets says almost nothing.
The article on how to spot a fake tipster has the full checklist.
Does Gambeta's record prove the model wins?
Not yet enough to claim it.
With 1,539 settled picks, the measured ROI is practically zero and the confidence interval runs from −4.7% to +4.7%.
We publish the figure as it is: the sample keeps growing.
What if my real edge is small, 1% or 2%?
It is almost impossible to verify from results alone.
You would need tens of thousands of bets.
That is why other signals matter more, such as comparing your odds with the close of an efficient market.