Probability in Business: How MrBeast, Bezos and Hormozi Win

Probability in business means judging a decision by its odds and its payoffs, not by how it happened to turn out. The people who make the most money say so in plain words. MrBeast runs his channel on click-through rates and A/B tests, Jeff Bezos takes 10% bets with 100x payoffs, and Alex Hormozi wins by volume. All of it is first-course statistics.

We build Probably, a place to practice exactly this math, so we went looking for what famous operators have actually said about it, on the record. Here are the best examples we could source, most popular first.

MrBeast: a 10% click-through rate is a probability

MrBeast runs the most-subscribed channel on YouTube, past 500 million subscribers. A production memo widely attributed to him, leaked in 2024 and never officially confirmed, tells new hires exactly which numbers matter:

The memo's own example: if 100 million people see a thumbnail and 10 million click, the CTR is 10%. In stats terms that's an estimate of a probability, P(click | the thumbnail was shown). Retention is a conditional probability too. The memo describes losing 21 million of 60 million viewers in the first minute of one video:

P(still watching at 1:00 | clicked) = (60M − 21M) / 60M
                                  = 39 / 60 = 0.65

Once YouTube let creators test thumbnails, he tweeted that he no longer had to guess: "I just test and see what people want." That's an A/B test, and you can check one with a two-proportion z-test. Say thumbnail A gets an 8.0% CTR and thumbnail B 8.6%, on 50,000 impressions each:

p̂ = (0.080 + 0.086) / 2 = 0.083
SE = √( p̂(1 − p̂) · (1/50000 + 1/50000) ) ≈ 0.00174
z = (0.086 − 0.080) / 0.00174 ≈ 3.4

If you want to try a calculation like this yourself, the formula sheet has the standardization and the normal tail you need.

Jeff Bezos: take the 10% bet every time

The cleanest statement of expected value from any founder is in Amazon's 2015 letter to shareholders. Bezos contrasts baseball, where a single swing can score at most four runs, with business, where the outcome distribution has a long tail:

E[payout] = 0.10 × 100 + 0.90 × 0 = 10 times the stake
E[profit] = 10 − 1 = +9 per unit staked
P(at least one hit in 10 bets) = 1 − 0.9¹⁰ ≈ 0.65

Try it below. Each bet stakes 1 and pays 100 times the stake with a 10% chance. The line is your average profit per bet as the bets pile up. It swings wildly at first, then settles on the expected value, the dashed line. Lower the payoff until the expected value turns negative and watch the same luck stop saving you.

A year later, Bezos added a second rule that's really about information and the cost of delay: most decisions "should probably be made with somewhere around 70% of the information you wish you had", because waiting for 90% usually means being slow (2016 letter).

Elon Musk: start when the odds are against you

Musk has been unusually open about expecting to fail. Asked on 60 Minutes in 2012 about SpaceX, he said he "would have to be insane" to think the odds were in his favour (CBS). In 2014 he went further:

He has also said he gave SpaceX roughly a 10% chance of ever reaching orbit (Space.com). Treat each launch as an independent trial with that chance and the probability of at least one success climbs fast:

P(at least one success in n tries) = 1 − (1 − p)ⁿ
p = 0.10, n = 4:  1 − 0.9⁴ = 1 − 0.6561 ≈ 0.34

Alex Hormozi: volume negates luck

Hormozi has an episode of his podcast, twice, titled Volume Negates Luck, about probability, luck and why doing more of the right thing beats both. It's the law of large numbers in four words. If each cold outreach has a 5% chance of a yes:

P(at least one yes in 50) = 1 − 0.95⁵⁰ ≈ 0.92
E[yeses in 50] = 50 × 0.05 = 2.5

Probability also sits inside his best-known framework, the value equation: value goes up with the dream outcome and the "perceived likelihood of achievement", and down with time delay and effort. Explaining it, he says:

Because likelihood multiplies the outcome, raising it from 40% to 80% doubles the perceived value of an offer with nothing else changed. That's expected value again, written as a sales formula.

Warren Buffett and Charlie Munger: decision trees and odds

In his 1988 letter, Buffett lists the questions he asks before an arbitrage deal. The first two are pure probability:

Robert Hagstrom, in The Warren Buffett Portfolio, quotes him putting it as a formula: take the probability of loss times the amount of possible loss from the probability of gain times the amount of possible gain. Here it is on a merger deal. A stock trades at $90, the buyer has agreed to pay $100, the deal has an 85% chance of closing, and the stock falls to $70 if it breaks:

E[price] = 0.85 × $100 + 0.15 × $70 = $95.50
edge     = $95.50 − $90 = +$5.50  (+6.1%, before time)

Munger, in his 1994 talk at USC, said Buffett "automatically thinks in terms of decision trees and the elementary math of permutations and combinations", and warned that without elementary probability you go through life badly outmatched (transcript). His example was the racetrack, where odds of 3 to 2 imply a 40% chance and 100 to 1 about 1%. A bet only has an edge when your probability beats the implied one.

Naval, Annie Duke and Taleb: variance, not just averages

Expected value is only half the story. The other half is how much the result can swing, and whether one bad swing ends the game. Naval Ravikant frames it with parallel universes:

That's the difference between a strategy that works with probability 0.999 and a lottery ticket that works 5% of the time. Annie Duke, a poker champion turned decision strategist, writes in Thinking in Bets that the quality of our decisions and luck are the two things that determine how our lives turn out. Her warning is against "resulting", judging a decision by its outcome. Even an 80% favourite loses one time in five, and the chance of winning ten such bets in a row is only 0.8¹⁰ ≈ 11%.

Nassim Taleb puts the risk of ruin most bluntly: "never cross a river if it is on average four feet deep" (The Logic of Risk Taking). A small risk repeated adds up. A 1% chance of ruin per day becomes 1 − 0.99¹⁰⁰ ≈ 63% over a hundred days.

Ray Dalio makes the matching argument for diversification in Principles: fifteen to twenty good, uncorrelated return streams cut risk dramatically without cutting expected return. You can check the claim yourself. For N equally volatile, uncorrelated streams, risk falls as σ/√N, so 15 streams cut it by about 74% and 25 by 80%. It's the same √n that makes the mean of a sample less spread out than a single observation, which is the topic of our guided lesson on sums and means.

The probability toolkit behind all of it

Every example above uses one of six ideas from an introductory probability and statistics course:

How to get fluent in the math

Reading these examples is the easy part. Being able to set up the expected value, the z-score or the 1 − (1 − p)ⁿ yourself, under exam pressure or in a pitch meeting, takes practice with feedback. That's what we built Probably for: you paste a question and solve it line by line, every line is checked as you write it, and the practice track picks your next question from your weakest skill. The lessons, drills and grading are free.