A decision matrix that shows how fragile it is MATRIX

A decision matrix makes a judgement look like a calculation

That is not a criticism. Writing the criteria down, weighting them before you score anything, and scoring each option against each criterion is a genuinely better way to think than arguing about conclusions. It forces the disagreement into the open, where it usually turns out to be a disagreement about weights rather than about facts, and that is a much more productive argument to have.

The trouble is what happens next. A total of 7.4 against 7.1 gets written into a paper as “the analysis showed”, the decimal point does the persuading, and nobody asks the only question that matters: how far would one number have to move for the answer to be different?

Usually the answer is “rather less than the error in every number in the table”. Two options separated by four per cent of the scale are not separated at all. The matrix has told you they are equivalent, and the decision is yours to make on grounds it does not contain, which is a real and useful finding, and the opposite of what the total looks like it is saying.

The sensitivity is the output

So after the totals, this works out how far each weight would have to move before the ranking changes, and says when the winner is inside the noise. Those two numbers are what you take to the meeting.

A weight that would have to double before the answer changed is a robust result: you can stop arguing about whether it should be 15 or 20, because it does not matter. A weight that only has to move two points is where the decision actually lives, and that is the conversation the group should be having instead of the one about the totals.

And when the top two are inside the noise, the honest write-up says so. “These two are equivalent on the criteria we agreed; we chose A because of X, which the matrix does not capture” is a defensible sentence. “The analysis showed A at 7.4” is not, and it falls apart the first time somebody re-scores one row.

Using it without fooling yourself

Set the weights before you see any scores, and preferably before the options are named. Weights chosen after the fact are chosen, consciously or not, to produce the answer somebody already wanted, and that is the single most common way these exercises go wrong.

Keep the criteria few and independent. Ten criteria that overlap double-count whatever they have in common, so an option strong on one theme wins by having it measured three times. Five that genuinely differ beat twelve that mostly do not.

Score every option on one criterion before moving to the next, rather than scoring an option all the way down. It is much harder to be consistent down a column than across a row, and consistency down the column is what the weights assume.

Nothing you type is uploaded. Alongside: DECISION records what was decided and what was known at the time, RISK assesses a risk the same way, and BRIEF turns the result into something that survives being passed on.