What it means
Most business decisions get made by argument in a meeting room, where the loudest or most senior view tends to prevail. Decision analysis replaces that with an explicit model of four things: what could we do, what might happen afterwards, how likely is each outcome, and what is each outcome worth?
The core tool is the decision tree. Squares mark choices you control, circles mark uncertain events you do not, and every branch carries a probability and a payoff, so you can work backwards from the far ends of the tree to a value for each starting choice.
It matters most when the stakes are high and the decision will not repeat often. Capital projects, product launches, litigation settlements and major contract bids all benefit, largely because the exercise forces a team to state assumptions that would otherwise stay unspoken.
A particularly practical strength is pricing the value of information. The same framework can tell you what it is worth paying for market research, a pilot or an expert opinion, by comparing the expected value of deciding now against the expected value of deciding once you know more.
The obvious weakness is that the probabilities are usually estimates rather than facts. That is manageable if you test whether the conclusion holds across a range of assumptions, and in practice the discipline of writing the numbers down is often worth more than the numbers themselves.
In practice
Real-world examples
Example 1
A pharmaceutical firm decides whether to run an extra trial before committing $80,000,000 to a manufacturing plant. A decision tree shows the trial costs $6,000,000 but changes the plant decision often enough to be worth it, so the trial goes ahead.
Example 2
A construction company facing a contract dispute compares settling now for $1,200,000 against litigating, which offers a 55% chance of paying nothing and a 45% chance of paying $3,000,000 plus costs. The expected cost of litigating exceeds the settlement, so the board settles.
Example 3
An online retailer chooses between two warehouse locations. Mapping demand growth scenarios against each site's capacity and rent shows the smaller, cheaper site wins in every scenario except the most optimistic, which the team judges unlikely enough to accept the risk.
Think of it
Decision analysis is like using a GPS navigator for a road trip. Instead of guessing which turn looks best, it calculates the distance, traffic, and speed limits for every route to show you the fastest way forward.
Formula
Calculation
Expected monetary value (EMV) = Sum of (probability of each outcome x payoff of that outcome)
Choose the option with the highest expected monetary value.
A drinks company must decide how to launch a new product.
Option A is an immediate national launch:
40% chance of strong demand, profit of $5,000,000: 0.40 x $5,000,000 = $2,000,000
35% chance of moderate demand, profit of $1,000,000: 0.35 x $1,000,000 = $350,000
25% chance of weak demand, loss of $3,000,000: 0.25 x -$3,000,000 = -$750,000
EMV of Option A = $2,000,000 + $350,000 - $750,000 = $1,600,000
Option B is a regional pilot costing $400,000, followed by a national roll-out only if the pilot works:
60% chance the pilot succeeds and the roll-out earns $3,500,000: 0.60 x $3,500,000 = $2,100,000
40% chance the pilot fails and the company stops, losing the pilot cost: 0.40 x -$400,000 = -$160,000
EMV of Option B = $2,100,000 - $160,000 = $1,940,000
Option B is worth $1,940,000 - $1,600,000 = $340,000 more than Option A, so the pilot route wins even though it delays revenue by a season. That $340,000 is essentially the value of the information the pilot buys.Case study
Seen in the real world
The company below is fictional and the case is purely illustrative. Brackwell Instruments, a maker of industrial sensors, was invited to bid for a three-year supply contract and its sales and finance teams disagreed sharply about the price. Sales wanted to bid low to win the volume, while finance wanted a higher price to protect margin, and neither side could do more than assert that they were right.
The finance director laid the choice out as a simple decision tree instead. Bidding at $4,000,000 gave an estimated 55% chance of winning and would deliver $600,000 of profit, an expected value of 0.55 x $600,000 = $330,000. Bidding at $4,400,000 gave a 30% chance of winning but $1,000,000 of profit, an expected value of 0.30 x $1,000,000 = $300,000. Preparing either bid would cost $80,000 whether or not it was won, reducing the two figures to $250,000 and $220,000 respectively.
The low bid won on the numbers, but only by $30,000, which was itself the useful finding. The gap was so narrow that it sat well inside the uncertainty around the win probabilities, so the team spent a fortnight talking to the customer to sharpen those estimates rather than treating the model's answer as settled. The value of the analysis was less the recommendation than the shared, written-down view of what the argument actually depended on.
Watch out
Common mistakes
- Believing the model produces the decision. Decision analysis organises evidence and assumptions, but a human still has to weigh factors such as reputation and strategy that resist being turned into numbers.
- Presenting invented probabilities as if they were measured. If a 60% estimate is really a considered guess, say so, and test whether the answer changes at 40% or 80%.
- Using expected value alone for a decision that could sink the company. Expected value assumes you can play the odds repeatedly, which is false when one bad outcome ends the business.
Questions
People also ask
What is a decision tree?
A diagram of choices and uncertain events, with probabilities and payoffs on the branches, used to calculate the expected value of each initial choice.
How do you get the probabilities?
From historical data where it exists, and otherwise from structured expert judgement, ideally gathered independently before people discuss it and converge.
Is this worth doing for small decisions?
Rarely, since the effort only pays off when the stakes are large, the uncertainty is genuine and the decision will not simply be repeated next month.