Decision Tree Analysis

Risk technique · See it on the map

Runnable here

Mapping an uncertain decision as branching choices and chances, then weighing each branch's likely value.

When to use it

For a discrete decision with a small number of real alternatives, where at least one branch involves a chance event with an estimable probability and payoff — build in-house versus buy a component with an uncertain defect rate, for instance, or invest in a fix now versus risk a penalty later.

When to avoid it

When the decision has continuous or many interacting uncertain factors rather than a handful of discrete branches — the tree gets unreadable, and simulation is the better tool. Also avoid it when the probabilities plugged into the tree are pure guesses dressed up as numbers; the arithmetic looks rigorous regardless of input quality, which can mislead a decision-maker into more confidence than the numbers deserve.

Steps

1. Draw the decision as a tree: a decision node for each choice available, a chance node for each uncertain event, and an end value for each resulting outcome.
2. Assign a probability to each branch out of a chance node; they must sum to 1 at each node.
3. Assign a monetary value (cost or payoff) to each end-of-branch outcome.
4. Working from the tree's ends back toward its root, compute the expected monetary value (EMV) at each chance node: sum of (probability x outcome value) across its branches.
5. At each decision node, choose the branch with the better EMV — the tree makes that comparison explicit instead of leaving it to intuition.
6. Report the chosen path and its EMV, along with the tree itself so the reasoning is checkable by someone else.

What it produces

• A decision tree diagram showing the choices, chance events, probabilities, and payoffs.
• An expected monetary value for each option under consideration.
• A recommended choice, with the reasoning behind it laid out rather than asserted.

Common pitfalls

• Presenting EMV as a guaranteed outcome rather than an average across many hypothetical repeats of the same decision — a project only happens once, so the EMV-best choice can still lose on this particular run, and a risk-averse stakeholder may reasonably prefer the lower-EMV, lower-variance option.
• Assigning probabilities without a stated basis, so the EMV inherits false precision from numbers nobody could defend under questioning.
• Omitting a real alternative from the tree because it complicates the diagram, which quietly biases the analysis toward whichever options were easiest to draw.

Worked example

A manufacturer deciding whether to tool a part in-house or outsource it draws a tree: tooling in-house costs $80,000 up front, with a 20% chance of a $40,000 rework if the first run fails tolerance (EMV = 80,000 + 0.2 x 40,000 = $88,000). Outsourcing costs a flat $95,000 with no rework risk. The tree makes the comparison explicit: $88,000 expected versus $95,000 certain, and the PM chooses in-house, accepting the wider possible range in exchange for the better average.

Source

• PMBOK-6 §11.4.2.5

Where it comes from: this technique is named by the PMBOK Guide, 6th edition.