Decision Tree Tool for High-Stakes Decisions

A decision tree tool does one thing well: it forces the room to draw every branch before it picks one. I have used trees for decades, in boardrooms, in project teams, and in regulatory reviews where the options carried real consequences. Consulting firms recommend them. Government agencies require them for cost-benefit submissions. The part that fails is not the diagram. It is the number written beside each fork. A probability that nobody tested is an assumption wearing a percentage sign.

I built the Walk to catch what the tree skips. A decision tree maps the branches; the method tests whether the number at each fork can bear the weight of the commitment. One structures options. The other tests the assumption underneath them.

A decision tree tool maps branches from a starting choice and assigns a probability to each fork, producing an expected value for the preferred path.

What a decision tree covers

The tree is strongest when the options are bounded and the probabilities are grounded in evidence. A product manager choosing between three launch windows, each with a conversion estimate drawn from last quarter's data, gets real value from a decision tree. The expected value at each branch is calculable because the numbers feeding it come from measurement, not from the room's best guess.

Trees also make the structure of a choice visible. A board that has been arguing in circles can see, on one page, where the paths diverge and what each path assumes about cost, timing, and outcome. That visual discipline is genuinely useful, as long as nobody mistakes the diagram for the decision.

Kamiński, Jakubczyk and Szufel showed in their 2018 framework that decision makers often cannot uniquely assign probabilities to possible events. Due to what they call non-stochastic distributional uncertainty, there is no single expected value. The number the tree produces depends on which probability estimate you feed it. Most practitioners skip sensitivity analysis entirely, which means the tree gives one answer when it should give a range. That gap between the single number and the real range is where decisions under uncertainty go wrong: the tree creates confidence the evidence does not support.

When a decision tree is enough

A decision tree is enough when the choice is bounded and the downside is containable. If the wrong branch costs the team a week or a modest budget line, the tree does the job. The options are known, the probabilities are drawn from data rather than from the room's comfort level, and the consequence of a wrong fork is recoverable.

Consider a logistics team choosing between two shipping routes. The cost data comes from contracts, the delay probabilities from the last 200 shipments. No probability at any fork is a guess. The tree produces a credible expected value because every input is grounded. That is the threshold for decision quality: the numbers feeding the structure must come from evidence, not from comfort. The data is clean, the stakes are bounded, and the tree adds genuine structure by forcing the team to calculate rather than assert.

The line moves when the assumption underneath the preferred branch reaches people, capital, or the organisation's licence to operate. At that point, the probability written beside the fork is no longer a technical estimate. It is a judgement about what the room believes will happen. The method matters there, because the tree accepts whatever number you write; it tests whether that number can bear the decision.

In my experience, the room usually knows when it has crossed that line. What it lacks is the discipline to stop drawing branches and start asking whether the numbers beside them are facts or hopes. No decision making framework will install that discipline for you; it has to come from the room itself.

Name the probability your decision tree depends on and ask whether it is drawn from evidence or from the room. Start the Walk →

What a decision tree skips

An insurer once built a decision tree for a wrongful-death claim and assigned a 70 per cent probability to a defence verdict. The case went to trial. Mock juries, evidence rulings, and witness credibility all shifted during preparation, every one of them a signal that the 70 per cent figure had become an untested assumption. Nobody updated the tree. The jury awarded $40 million. The insurer was later hit with a $7.2 million judgement for relying on an analysis it never revisited.

The problem was not the tree's structure. The branches were correctly drawn. The problem was the number at the first fork: it was a guess that hardened into a fact the moment someone wrote it down, and nobody asked the question that would have exposed it. What must be true for this probability to hold? That question never entered the room. The 70 per cent figure sat unchallenged through months of preparation while every signal around it was changing.

John DeGroote, a mediator who has written extensively on probability misuse in decision trees, calls this the garbage-in problem: if the probabilities lack an evidentiary foundation, the expected value is meaningless regardless of how sophisticated the tree. A court in Yassin v. Certified Grocers of Illinois rejected a decision tree analysis because the probabilities were assigned "from thin air." The method forces that separation, between what the room can support with evidence and what it is assuming because the tree demanded a number.

That is the step every decision tree tool in the decision making tool guide skips. Trees ask "what are the options?" and "what is the probability?" They never ask "what are we assuming about those probabilities?" The tree structures the choice. It does not test the evidence underneath it.

What the Walk produces

The Walk takes a team through the full method and produces a Decision Document at the end. The document names the assumption that matters, records the evidence that supports or undermines it, and writes the monitoring signal, the trigger that would reopen the call before the assumption fails silently.

That monitoring signal is the part most rooms never write. A decision tree declares a winner and closes. The Decision Document stays open: it names the condition under which the preferred branch would no longer be preferred, and it assigns someone to watch for it. Without that signal written down, the decision simply ages. Six months later the conditions that made it right may have shifted, but nobody has a trigger to revisit. The tree that produced the original recommendation sits in a folder, its probabilities frozen, its branches untested against what actually happened.

A decision tree produces a diagram with a recommended branch. The Decision Document produces accountability: who owns the assumption, what would change their mind, and when they would act. The tree tells the room which option scores highest. The document tells the room what must remain true for that option to hold.

I built it because most rooms will not produce that document without structure. Roger Estall and I wrote Deciding after sitting through too many meetings where the tree was drawn, the preferred branch was obvious, and nobody was willing to say what they were assuming about the number that made it obvious.

Grant Purdy is the co-author, with Roger Estall, of Deciding (2020), and the architect of the Universal Decision-Making Method.