Decisions.

The framework

1. 01

   Write down every way this ends — including the branches your story omits.

   List the outcomes that are genuinely separate and between them cover the ground: the good case, the ordinary case, the slow grinding case, the failure. Put a payoff against each in one unit — money, hours, customers, months of runway — and hold that unit, because a tree that mixes dollars with reputation cannot be added up. Three to five branches is usually enough; more precision than that is false. One outcome has usually been rehearsed so often that it has quietly stopped being a possibility and started being the plan, and the branches either side of it never make it onto the page. They exist whether or not you write them down.

2. 02

   Force a number where you want to write 'likely'.

   In 1964 the CIA analyst Sherman Kent published Words of Estimative Probability, after finding that colleagues reading the same estimative phrase meant anywhere from roughly twenty per cent to eighty per cent by it. Your board papers have the same defect, and it is load-bearing: verbal odds cannot be argued with, tracked or corrected, which is precisely why people reach for them. Put a percentage against every branch and make the set total a hundred. A badly estimated number beats a well-chosen adjective every time. If a probability feels impossible to state, ask instead what would have to be true to move it, because that question almost always produces the estimate.

3. 03

   Multiply, then subtract the option you would otherwise take.

   Expected value is the sum of each branch's payoff multiplied by its probability, and the sum takes a minute. The comparison takes judgement. A positive expected value is never a reason to act on its own, because the alternative is not nothing: the same money, and the same fortnight of your three best people, could go somewhere else entirely. So calculate the expected value of the honest next-best option and subtract it. What remains is the edge, and the edge is the thing you are actually deciding on. The trap is the zero baseline — grading a bet against doing nothing, when doing nothing was never the real competitor.

4. 04

   Two vetoes before you take the highest number: survivability and repetition.

   Expected value is an average across many trials, and averages only tell the truth about bets you get to run many times. So ask two questions before acting on the arithmetic. Can you absorb the worst branch if it lands first — not comfortably, but at all? And will you get to make this class of bet again, so the average has room to assert itself? A bet failing either test gets declined or sized down whatever the maths says, because a ruinous branch at five per cent still arrives one time in twenty. Then record the expected value you calculated, since that record is what later lets you judge the decision rather than the outcome.