What it means
The structure is deliberately plain: a square node marks a point where a person chooses something, a circle node marks a point where chance decides, and each line leaving a node is one possible path. At the far right of every path sits a payoff, which is the cash result if that particular sequence of events happens.
Managers reach for decision trees when a decision cannot wait for certainty, which describes most real business choices. Should you build the bigger factory or the smaller one, settle the lawsuit or fight it, run a pilot or go straight to full launch?
Laying the options side by side forces a team to state its assumptions out loud instead of arguing from instinct. You work a tree from right to left, a process usually called rolling back.
At each chance node you multiply every payoff by its probability and add the results to get an expected value, and at each decision node you keep the branch with the highest expected value and discard the rest. The probabilities are the weak point, because they are estimates rather than facts.
Good practice is to test how far a probability would have to move before the recommended choice flips, which is a form of sensitivity analysis. If the answer is "barely at all", the tree is telling you the decision is close rather than clear.
A common variant discounts payoffs that arrive in later years so the tree compares present values rather than raw cash. Another variant replaces expected value with a risk-adjusted score when the business genuinely could not survive the worst branch, since a small chance of insolvency matters far more than an average figure suggests.
In practice
Real-world examples
Example 1
A speciality drinks maker is deciding whether to spend $600,000 on a national listing push. The tree gives a 45% chance the supermarket chain renews, worth $2,000,000, and a 55% chance it does not, worth $400,000. Expected value is $1,120,000, comfortably above the $600,000 cost, so the board approves the spend.
Example 2
A logistics firm faces a customer claim and builds a two-branch tree: settle now for a certain $250,000, or defend at a legal cost of $180,000 with a 30% chance of paying $900,000. The defend branch has an expected total cost of $450,000, so settling is cheaper on the numbers and the general counsel recommends it.
Example 3
A software company uses a decision tree before a platform rewrite, splitting the outcome into on-time delivery, six-month slip and abandonment. Quantifying the abandonment branch at a $3,000,000 write-off changes the conversation from "how exciting is this" to "how do we cut the chance of that branch".
Think of it
Think of a decision tree like using a satnav for a road trip. It shows you the main route, potential traffic jams, and alternative turn-offs so you can pick the fastest way to reach your destination.
Formula
Calculation
Expected value of a chance node = (probability of outcome A x payoff A) + (probability of outcome B x payoff B). Net expected value = expected value - upfront cost.
A packaging business is choosing between a large new production line and a small one. The large line costs $2,000,000 upfront. Demand is judged 60% likely to be high, returning $5,000,000, and 40% likely to be low, returning $1,000,000.
Expected value = (0.60 x $5,000,000) + (0.40 x $1,000,000) = $3,000,000 + $400,000 = $3,400,000
Net expected value = $3,400,000 - $2,000,000 = $1,400,000
The small line costs $800,000 and returns $2,200,000 in high demand or $1,200,000 in low demand, with the same probabilities.
Expected value = (0.60 x $2,200,000) + (0.40 x $1,200,000) = $1,320,000 + $480,000 = $1,800,000
Net expected value = $1,800,000 - $800,000 = $1,000,000
The large line wins on expected value by $400,000. The tree also shows the price of that advantage: in low demand the large line loses $1,000,000 while the small line still earns $400,000.Case study
Seen in the real world
Harbourline Ferries is an illustrative, fictional coastal operator running six vessels between three island towns. Its board was split over whether to buy a second high-speed catamaran for $4,500,000 or refit two ageing vessels for $1,200,000, and the argument had run for three meetings without moving.
The finance director built a decision tree instead of another slide deck. Tourism growth was given a 55% chance of continuing, in which case the catamaran returned $8,000,000 of lifetime cash and the refit returned $3,000,000, and a 45% chance of stalling, in which case the catamaran returned $2,500,000 and the refit returned $2,200,000. The catamaran's expected value came to $5,525,000 against a $4,500,000 cost, a net of $1,025,000, while the refit came to $2,640,000 against $1,200,000, a net of $1,440,000.
The refit won, and it won for a reason nobody had articulated before: the catamaran only paid off in one branch. Harbourline's board then asked the more useful question of what evidence would raise the tourism probability above 64%, which is the point at which the catamaran overtakes the refit.
Watch out
Common mistakes
- Treating the probabilities as if they were measured facts. They are almost always informed guesses, and a tree built on invented precision is still a guess with better graphics.
- Choosing the highest expected value without looking at the worst branch. Expected value is an average across futures the business will only live through once, so a branch that would be fatal deserves separate attention.
- Leaving out the option of doing nothing. Every tree should include a "carry on as we are" branch, otherwise the analysis quietly assumes action is compulsory.
Questions
People also ask
How many branches should a decision tree have?
Few enough that a non-specialist can read it in a minute, which in practice usually means two or three options and two or three outcomes each.
Do decision trees work when the payoffs are not financial?
Yes, as long as you can score the outcomes consistently, though translating outcomes into money keeps the comparison honest and easier to challenge.
Should payoffs be discounted for time?
If the branches pay out in different years, yes, use present values so that a distant $1,000,000 is not compared directly with cash arriving next quarter.