All lessons

Module 2: Probability for decisions · Lesson 2.3 · about 20 min

Expected value: is the promotion worth it?

Expected value as the long-run average of a chancy decision: coupons, warranties, stock levels and lead scoring. How to compute it, why its standard deviation matters too, and the law of large numbers that makes it work.

Key points, meanings and formulas on one page: choose "Save as PDF" in the print box.

Module 2 · about 20 minutes · school maths. You’ll be able to compute expected value and its standard deviation, use them to compare business options, and explain why the long run matters.

The expected value is what a chancy choice pays per try on average over many tries, so it is the number to compare options with when you will repeat the choice.

In everyday terms: a coin game that pays $3 on heads and nothing on tails is worth $1.50 a go in the long run. You never win $1.50 on a single toss, yet after 1,000 tosses you will have close to $1,500.

1. Deciding under chance

Many business choices are bets: send a coupon or not, stock 50 units or 80, chase this lead or that one. Each option has several possible outcomes with different chances. Expected value (EV) turns each option into one number: what it pays on average, if you made the same choice many times.

2. The formula

Multiply each outcome by its probability and add:

E[X]=∑ixipi

A lead worth $500 that converts 20 % of the time has EV = 0.2 × 500 + 0.8 × 0 = $100. If a sales rep has 50 such leads, expect about 50 × $100 = $5,000.

EV is the mean of a random variable (Lesson 3.1 makes that precise), so it shares the mean’s strength: EVs add up. The EV of a portfolio of leads, campaigns or products is the sum of their EVs.

3. Run it many times

Try it: does the coupon pay?
expected profit, no coupon
expected profit, coupon
customers simulated
Without a coupon, 5 % of customers buy and each sale makes $40 profit (made up). The coupon raises the buy rate but takes its value off every sale. Expected value = chance × payoff, added over outcomes: 0.05 × $40 = $2.00 per customer without it. Early on the lines cross and recross; only after thousands of customers does luck wash out.

With a few hundred customers, the line for the worse option can sit above the better one for a while. Only over thousands of customers does each line settle on its EV. This is the law of large numbers, and it is why EV is the right guide for decisions you repeat often (pricing, promotions, credit limits) and a weaker guide for one-off bets.

4. Spread matters too

Two options can share an EV and carry very different risk. The standard deviation of the outcome measures that:

SD(X)=∑ipi(xi−E[X])2

A $100 sure payment and a 1 % chance of $10,000 both have EV $100, but the second is far riskier. Businesses that cannot survive a bad run (small firms, tight cash flow) should weigh the SD, not just the EV.

Outcomes x
The possible results, in money or units.
Probabilities p
How likely each result is; they add up to 1.
E[X] = Σ p·x
Weight each result by its chance and add up: the long-run average per try.
SD of X
How far a single try usually lands from E[X]: the risk.
Physical meaning
Expected value is the balance point of all the possible outcomes, each weighted by how often it happens. Businesses that repeat a decision thousands of times (coupons, warranties, stock) live on that average; a person making the choice once also cares about the spread.

“The expected value is the most likely result.” A $10 coupon with an EV of $2.40 per customer never pays $2.40 to anyone: each customer gives $30 or $0. EV is a long-run average, not a typical single outcome.

Watch it happen, one step at a time:

The choice. Send a $10-off coupon or not? Each sale makes $40 profit before any discount.
No coupon. 5 % buy. EV per customer = 0.05 × $40 = $2.00.
Coupon. 8 % buy, but each sale now makes $30. EV = 0.08 × $30 = $2.40.
Scale up. To 10,000 customers: $20,000 against $24,000. The coupon wins by $4,000.
Check the assumptions. Would some of the 8 % have bought anyway at full price? That "cannibalisation" is the first thing to test (Module 7).
Maths and principle

Principle: the expected value is the probability-weighted average outcome. Over many independent repetitions, the average result gets closer and closer to it.

Equations: E[X]=∑xipi; Var(X)=∑pi(xi−E[X])2; for n independent repeats, total EV is nE[X] and the SD of the total is nSD(X). The total’s spread grows only like n while the EV grows like n, so the result becomes predictable as a share of the total.

Analogy: an insurer cannot predict whose house will burn, but with a million policies it can predict the total claims within a few percent.

5. Worked example

Made-up numbers. Should a shop send a $10-off coupon? Each sale makes $40 profit before any discount.

Option Buy rate Profit per sale EV per customer
No coupon 5 % $40 0.05 × 40 = $2.00
Coupon 8 % $40 − $10 = $30 0.08 × 30 = $2.40

The coupon wins by $0.40 per customer: on 50,000 customers, about $20,000. But the result is sensitive: at a 6.5 % buy rate with the coupon, EV = 0.065 × 30 = $1.95, and it loses. You would need a test (Module 7) to know the real buy rate before rolling it out.

Hidden costs: this EV ignores customers who would have bought anyway but now use the coupon. That is exactly why “no coupon” sales are counted at full profit: always compare against what would have happened without the change.

6. Practice

  1. A raffle sells 1,000 tickets at $5 for one $2,000 prize. What is a ticket’s EV?
  2. A store can stock 60 units (profit $300 if demand is high, chance 0.3; $200 if normal, 0.5; −$50 if low, 0.2). What is the EV?
  3. Option A: $50 for sure. Option B: 50 % chance of $0 and 50 % of $100. Same EV? Which has the bigger SD?
Answers
  1. 0.001 × 2,000 − 5 = 2 − 5 = −$3 per ticket.
  2. 0.3 × 300 + 0.5 × 200 + 0.2 × (−50) = 90 + 100 − 10 = $180.
  3. Both $50. A has SD 0; B has SD 50 (each outcome is 50 from the mean).

Try it on your own data: the probability tree works out the expected value of a decision with several stages.

New words: expected value, payoff, law of large numbers, risk, standard deviation of an outcome.

  • E[X] = sum of (outcome × probability): the long-run average per try.
  • Compare options by EV when the decision is repeated many times.
  • The SD of the outcome measures risk; two options with equal EV can differ a lot in it.
  • The law of large numbers: the average of many tries settles near the EV.

Check yourself

1A lead converts with probability 0.2 and is worth $500 if it does. What is its expected value?
2Why can a decision with positive expected value still lose money?
3An extended warranty costs $40. The chance of a $300 repair is 8 %. What is the expected value of buying it, for the customer?

Sources

Your ticks stay in this browser. Save them as a file on the course home.

Updated 2026-11-22. Numbers in the examples are made up to show the idea unless a source is given. Simulations run in your browser; nothing you type is sent anywhere.