Offer someone a choice: $100 for sure, or a 1% chance at $5,000. The $5,000 grabs attention, but imagine making this choice 100 times. The sure option pays $100 each time, $10,000 in all. The gamble wins about once, $5,000 in all. Per choice that is 0.01×5000=50 dollars, the gamble's expected value. On average the boring option is twice as good.
Expected value is a long-run average, so it is the right guide for choices repeated many times, like a store setting prices across thousands of sales. For a choice made once, also look at risk: how spread out the outcomes are and how bad the worst one is. Option A pays $500 for sure; option B pays $1,200 with probability 0.5 and loses $100 otherwise. B's expected value is 0.5(1200)+0.5(−100)=550 dollars, more than A's 500, yet someone who cannot afford to lose $100 this month may sensibly take A.
Insurance shows both ideas. Suppose a phone plan costs $8 a month and 3% of phones need a $200 repair each month. The insurer's expected cost per customer is 0.03×200=6 dollars, so it gains about $2 a month per customer. For the customer, buying has a lower expected value than not buying, yet it can still be sensible if a sudden $200 bill would be a real hardship.
Computing one expected value is not yet a decision. "What is this ticket's expected value?" has a single numeric answer; "Should you buy it?" needs that number for every option, plus a judgment about risk.
Think of it like this. Choosing a supermarket checkout line is a small version of this. You can't know which line will move fastest, so you weigh what you see (how many shoppers, how full their carts) and pick the best average bet, avoiding the line where one price check could stall everything. Many real decisions happen only once, though, so no long run averages away bad luck.
Core idea. Decision under uncertainty means choosing the option with the best probability-weighted outcome, not the one with the flashiest single result.