How to use expected value to make a decision between two options

What is expected value and how is it calculated?

Expected value is the probability weighted average of the possible payoffs of an option. For each outcome, multiply the payoff by its probability, then add the products. A simple case: a 60 percent chance of gaining 100 and a 40 percent chance of gaining 0 gives 0.60 × 100 + 0.40 × 0 = 60. The probabilities of all outcomes sum to 1.

The number reads as the average result if the same choice were repeated many times. For a single decision it works as a common yardstick, so two options with different shapes can be placed on one line. A sure payoff of 50 sits below the 60 of the example, and the comparison takes one line to write down.

The calculation itself has a compact form. The expected value formula writes each payoff next to its probability and adds the products, and a worked example with three options shows the spread around the average, a probability test, and a discount rate for payoffs that arrive in different periods.

Where do the payoffs and probabilities of an expected value come from?

Payoffs come from the plan: revenue, cost, and timing under each outcome, stated in one currency and one period. Use net figures, after the cost of the action itself, so that two options are compared on the same base. Write the unit beside every number, such as US$ thousands over 24 months.

Probabilities are the weak point, so label their source. A frequency from comparable past cases is stronger than a hunch, and a hunch stated openly is stronger than one hidden in a spreadsheet cell. Give each probability a range, such as 60 to 80 percent, and note who proposed it, so the later review can learn from the gap.

How does expected value compare two options in a worked example?

Take two ways to enter a new market, all figures illustrative and in US$ thousands. Option A uses a local distribution partner: a 70 percent chance of gaining 400 and a 30 percent chance of losing 100. Its expected value is 280 from the gain case minus 30 from the loss case, or 250.

Option B is an own launch: a 40 percent chance of gaining 900 and a 60 percent chance of losing 300. Its expected value is 360 from the gain case minus 180 from the loss case, or 180. Option A leads by 70 points of expected value and carries a smaller worst case, 100 against 300.

Choices that unfold in stages need a picture. A decision tree draws the options, the uncertain events, and the payoffs as branches, then rolls back from right to left to show whether a pilot before a launch is worth its cost and how high that cost can go.

At what probability of success does option B overtake option A?

A break-even probability shows how much the answer depends on one estimate. Let p be the chance of success for option B. Its expected value is 900p minus 300 times (1 minus p), which simplifies to 1,200p minus 300. Setting this equal to 250 gives p = 0.458, or 45.8 percent.

The team now has a precise question: is the chance of success for an own launch above 0.458? The estimate in the example is 0.40, so option A holds the lead. A credible case for 0.55 would flip the ranking and lift option B to 360. Evidence on that single probability is worth collecting before the decision.

How does risk appetite change an expected value decision?

Expected value is neutral to risk, so it treats a coin flip for 1,000 the same as a sure 500. Organisations are not neutral. A firm with a cash buffer of 250 can absorb the loss of option A but could be forced to stop after the loss of option B, which is 300 in the example.

Add a second column next to the average: the worst case and the probability of reaching it. State a loss limit before the analysis, such as the amount the firm can lose without changing its plans. An option that breaks the limit is set aside whatever its average, and the remaining options are ranked by expected value.

When every outcome is a sum of money, the measure takes a specific name and a specific set of rules. Expected monetary value suits repeated bets that are small against capital, and it pairs well with a loss limit tied to the cash reserve and a ceiling on what research is worth.

Which mistakes distort an expected value calculation?

Four mistakes recur. Probabilities that do not sum to 1 leave an outcome unpriced. Payoffs in different periods are added without discounting for time. Outcomes that move together, such as a downturn hitting two markets at once, are treated as independent. Confident probabilities appear without any evidence behind them.

A short checklist catches most of them: do the probabilities sum to 1, are all payoffs on the same base, is each probability sourced, and has a colleague challenged the largest payoff? Ten minutes of review costs far less than a decision built on a miscounted tree, and the checklist can be reused on the next choice.

How does the EQOS Asia decision engine extend expected value?

Expected value prices one choice with a few outcomes. The EQOS Asia decision engine widens the frame: its Prediction pillar projects 3 trajectories, optimistic, median, and adverse, with deterministic probabilities and dated tipping points (EQOS Asia method page, as of October 2026). The inputs come from structured interviews and documents, and the computation is replayable.

A reader can start with the average on a sheet and then try a first score in the simulator on a real choice. The 30-minute pre-diagnostic, scheduled in Singapore time (SGT), reviews the probabilities and payoffs of a live decision (EQOS Asia contact page, as of October 2026).

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile