Late one evening in my Manchester flat, I was staring at a scatter plot of EuroMillions draws that looked more like a spilled bag of flour than a predictable pattern. It was one of those damp, grey nights where the rain against the window makes you want to solve a puzzle, but the more I looked at the 'hot and cold' charts on the internet, the more my head hurt. As a math teacher, I spend my days explaining to Year 10s why probability isn’t a feeling, yet here I was, surrounded by websites claiming that the number 23 was 'due' because it hadn't appeared in three weeks.
Right then, I decided to stop shouting at my laptop and start a proper experiment. I’m not a professional gambler, and I’m certainly not a data scientist with a supercomputer. I’m just a teacher who believes in being honest about numbers. I know the Jackpot winning odds for the EuroMillions are a staggering 1 in 139,838,160. To put that in perspective for my students, that's like picking one specific second out of about four and a half years. It's astronomical. But if people were going to use AI tools to try and 'beat' those odds, I wanted to see what was actually happening under the hood.
The Notebook in the Locked Drawer
In late February, I bought a simple spiral-bound notebook from the corner shop. It now lives in my locked desk drawer at school, tucked safely under a stack of Year 10 algebra tests. There is something deeply satisfying about the physical record of data. I can still smell the faint scent of old paper and graphite as I flip through the coffee-stained pages, seeing the rows of numbers I’ve meticulously recorded every Tuesday and Friday night.
My students would be absolutely delighted to know their teacher spends her Friday nights debating the Law of Large Numbers with a chatbot, but for now, it’s my little secret. The project was simple: I would run a parallel test for six months. On one side, I’d use my own manual frequency analysis—the kind of 'hot and cold' tracking everyone suggests. On the other side, I’d use three different AI-driven platforms that claimed to use neural networks to optimize number selection. I wanted to see if the machine could find something my spreadsheets couldn't.
The first thing I had to account for was the sheer scale of the game. The EuroMillions main number pool goes from 1 to 50, and the Lucky Star number pool ranges from 1 to 12. When you calculate the total main number combinations—which is a standard combinatorial '50 choose 5'—you get 2,118,760 possible sets of five numbers. Once you factor in those Lucky Stars, the complexity explodes. I’ve written more about the foundational math in my previous post, The Notebook in My Desk Drawer: My Months Comparing AI Patterns to My Own Maths, which covers the start of this journey.
The Fallacy of the 'Cold' Number
One rainy Tuesday evening during the second month of my testing, I realized why most lottery advice is total rubbish. I was looking at a 'cold' number—let's say 42—which hadn't been drawn in ages. The logic online says it's 'overdue.' But here is the thing: the lottery machine has no memory. It doesn't remember that it didn't pick 42 last week. It doesn't feel guilty about it. It doesn't have a plan to 'catch up.'
In probability, we call this the Gambler's Fallacy. If you flip a fair coin and get heads ten times in a row, the probability of getting tails on the eleventh flip is still exactly 50%. The coin doesn't know what it did before. The EuroMillions balls are exactly the same. Tracking cold numbers to predict future draws is a fallacy because the physical randomness of the draw means every event is independent. This was the first major realization I noted in my notebook: my manual frequency charts were essentially documenting the past, but they were mathematically useless for predicting the future.
I’m not a financial advisor or a professional statistician, and you should always treat this as a bit of a laugh rather than a retirement strategy. If you’re ever worried about your spending, please do speak to a professional. But from a purely mathematical standpoint, chasing 'due' numbers is like chasing a shadow—it’s always just out of reach because it doesn’t actually exist.
AI vs. The Birthday Bias
Just before the Easter break, I started noticing something interesting in the AI picks. While the AI wasn't 'predicting' the winning numbers (because, again, 1 in 139 million), it was doing something much smarter than most humans. It was avoiding what I call 'The Birthday Bias.'
Most people pick numbers based on dates—birthdays, anniversaries, or the day their cat was born. This means the numbers 1 through 31 are disproportionately popular. If the winning numbers are all under 31, you might win, but you’ll likely share that jackpot with dozens of other people. The AI tools I was testing didn't care about birthdays. They were consistently suggesting numbers in the 32 to 50 range.
By early July, my notebook showed a clear trend. While neither my manual picks nor the AI picks were hitting the jackpot (no surprise there!), the AI-generated lines were statistically 'unique.' They represented combinations that fewer humans would naturally choose. In a game like EuroMillions, where the prize is split equally among winners, the real 'strategy' isn't about increasing your odds of winning—which are fixed by the math—but about decreasing the odds of having to share the prize if you do win. I've explored this logic further in my article on Probability of EuroMillions Patterns in Historical Draw Data.
What Six Months of Data Taught Me
As I closed the notebook for the summer term, I felt a strange sense of relief. The AI tools hadn't turned me into a millionaire, but they had effectively 'cleaned' my playstyle. They removed the emotional attachment to 'lucky' numbers and replaced it with a data-driven approach to number distribution.
Here is what my school desk notebook ultimately revealed:
- Frequency is a ghost: Both 'hot' and 'cold' numbers are just statistical noise in the short term. Over millions of draws, they even out (the Law of Large Numbers), but over a human lifetime, they mean nothing for the next draw.
- Spread matters: AI is excellent at ensuring your 5/50 and 2/12 selection covers the entire board, rather than clustering around dates.
- The Goal is Uniqueness: Using data tools to avoid common patterns (like consecutive numbers or all evens) is the only way to protect the value of a potential win.
I’ll keep the notebook in my drawer for the next term, mostly because I enjoy the process. There’s a certain beauty in the chaos of randomness, and even if I never beat the 1 in 139,838,160 odds, at least I know I’m not falling for the same old rubbish advice that ignores the basic rules of probability. It’s a good lesson for my students, and an even better one for anyone who thinks the lottery is a game of 'luck' that can be figured out with a simple frequency chart.