112
...the next pick to the people who saw you pick the "winner". Now half of those people see one team, the other half see you pick the other team, and whoever saw you pick the winner thinks you've got a 100% accuracy rate over two games. You could do that for a while and then offer to sell your pick for the Superbowl. Starting with a big enough group in the beginning, this might be really lucrative.
But is it legal?
Well it's not mathematically possible
The formula is p/(2^n)
P would be the number of people you start with, and n is the number of games.
If you start with the population of the US, 350 million people, you can only do this for about 28 matches before you run out of people.
If it makes any difference, American football seasons are only like 16 games.
"only"
I'm pretty sure people would give you money after 10 correct predictions in a row. At that point there are 350k remaining.
Except for the fact that the entire rest of the population would have gotten the emails. This relies on literally nobody talking about it.
How many coincidences do you need until you believe something to be true?
Science is usually fine with a one in twenty chance (p<0.05, 5 emails) or one in one hundred (p<0.01, 7 emails). Physics is the most strict discipline and requires up to one in three hundred (p<0.003, 9 emails), or even one in a 3.5 million chance (5 sigma, p<0.0000003, 22 emails).
Sure, most mails would be caught in the spam filter anyway and you're not gonna get emails for every single person. And if you have two mail addresses for the same person they'd immediately catch on, once the two addresses get sent different predictions.
But the point is, we are dealing with big numbers here and it is very much reasonable to expect some level of success from such a strategy.
I doubt it'd be any amount of successful. And yes it'd be caught in the spam filter with the other 95% of total emails sent every day.