How to make a decision tree, step by step
A decision tree is a question, the options you have, the things you do not control, and a number on every ending. This guide builds one in six steps, with the maths done at each step, using a lease decision as the running example.
Published 2026-09-27. By the DecisionBranch team.
What a decision tree is
A decision tree is a map of a choice: the options you can take, the outcomes you cannot control, and the value of every ending. Squares are decisions, circles are chances, and reading it back from the ends tells you which option is worth most.
Two kinds of node do all the work. A decision node is a choice you make, with one branch per option. A chance node is something that happens to you, with one branch per outcome and a probability on each. Every path ends in a payoff: what that ending is worth. Costs sit on the branches where you pay them.
That is the whole notation. Drawing tools stop here and leave you with a picture. The point of building one is the arithmetic, which the rest of this guide shows.
Step 1: write the question
Start with one line that states the choice as a question. If you cannot write it in one line, you have two decisions; make two trees.
In DecisionBranch the tree is plain text, so step 1 is literally one line. Keep the timeframe in mind: the payoffs at the end will be measured over it.
Renew the office lease or move?
Good questions name the choice, not the goal. Renew the lease or move? is a decision. How do we cut costs? is a brainstorm, and it will not fit on a tree until you have turned it into options.
Step 2: list the options
Indent one line per option under the question. Include the option of doing nothing, because it is often the runner-up and sometimes the winner.
Renew the office lease or move? Renew for three years Move to the smaller space Go fully remote
The tree already draws. Three branches under the question make it a decision node. Three to five options is the usual range; if you have ten, group them.
Step 3: add what you do not control
Under each option, add the outcomes that could follow and a probability for each. The probabilities under one node must add up to 1. A cost on an option goes in brackets on its line.
Here the uncertainty is whether the team grows over the three years. It affects every option, so it appears under each one, and the same probability can be different under different options if the option changes it. Two of the options cost money to take: moving and going remote.
Renew the office lease or move?
Renew for three years
Team grows (p 0.5)
Team stays the same (p 0.5)
Move to the smaller space (cost 40000)
Team grows (p 0.5)
Team stays the same (p 0.5)
Go fully remote (cost 15000)
Team grows (p 0.5)
Team stays the same (p 0.5)Where do the probabilities come from? Your best estimate, base rates, a colleague's view. Do not stall here: step 6 tells you whether the estimate even matters.
Step 4: put numbers on the ends
Give every final outcome a payoff after a colon. Use one unit for the whole tree, over the timeframe in the question, and write payoffs as the gross outcome; costs on the branches are subtracted for you.
# Three-year view. Payoffs are contribution after rent, in thousands.
Renew the office lease or move?
Renew for three years
Team grows (p 0.5): 900
Team stays the same (p 0.5): 420
Move to the smaller space (cost 40)
Team grows (p 0.5): 740 # cramped, some churn
Team stays the same (p 0.5): 620
Go fully remote (cost 15)
Team grows (p 0.5): 820
Team stays the same (p 0.5): 610Payoffs here are three-year contribution after rent, in thousands. The comment lines after # are notes for the reader; the engine ignores them. Now the tree is complete and the numbers appear on every node.
Step 5: read the answer
The recommendation is Go fully remote, worth 700 on average, 40 ahead of Renew for three years. The lit path shows it and the EV badge on every node shows the working.
The maths is called rollback. Work from the ends: each chance node becomes the probability-weighted average of its branches; each decision node takes its best branch, after subtracting the cost of getting there. For the renew option: 50% × 900 + 50% × 420 = 660. Moving is 50% × 740 + 50% × 620 = 680, minus the 40 cost, so 640. Remote is 50% × 820 + 50% × 610 = 715, minus 15: 700. The expected value guide goes deeper, and the analysis calculator writes the steps out for any tree you paste.
Step 6: check how sure you need to be
Sweep the probability you trust least and find the switch point. Here, if the chance the team grows under renewing rises above 58.3%, the answer changes to Renew for three years. It is 50% now.
This is the step that separates a decision tree from a drawing of one. A recommendation that survives a wide range of your estimates is robust; one that flips a few points away from your guess is telling you what to find out before you decide. In the editor, open Analysis → Sensitivity, pick any probability, payoff or cost, and the switch point is marked on the curve.
Two more panels help here. The risk profile lists every ending you could reach under the recommendation and how likely each is, so the chance of a bad outcome is visible and not hidden in an average. The value of information tells you the most it would be worth paying to learn an uncertain outcome before you choose.
Common mistakes
Most wrong trees have one of five problems: probabilities that do not add up, payoffs netted twice, mixed units, options that are not really different, and a question that hides two decisions.
- Probabilities that do not add to 1. The editor flags the line. If two outcomes overlap, redefine them so they cannot both happen.
- Costs subtracted twice. Write the gross payoff on the ending and the cost on the branch. If you also net the cost out of the payoff, the option is penalised twice.
- Mixed units. One branch in thousands, another in units, one over a year and another over three. Pick one unit and one timeframe for the whole tree and say so in a note.
- Options that are the same choice. If two options lead to identical outcomes with identical numbers, merge them; the tree gets easier to read and the maths does not change.
- Two decisions in one. If an option's outcome leads to another choice, that is a nested decision node, and it belongs in the tree. If the second decision is unrelated, it is a second tree.
Which tool to use
Use a drawing tool if you only need the picture. Use a spreadsheet if you already have the structure and want to type formulas. Use a tool that does the maths if you want the picture and the numbers from the same source, with the switch points.
Whiteboard tools and diagram apps draw a decision tree quickly and calculate nothing. Spreadsheets calculate but cannot show the structure, and nested probabilities in cells are easy to break. DecisionBranch keeps the tree as text, draws it, and does the rollback, sensitivity and value of information from that same text. See the comparison with drawing tools for a feature-by-feature list.