Part II · Chapter 3 · Risk, without the score · lesson 1
Two risks, the same number, opposite decisions
12 min
Part II · Chapter 3 · Risk, without the score · lesson 1
12 min
Two rows out of a register somebody keeps for real. The scale is theirs: five bands of likelihood, six of impact, because whoever built the spreadsheet wanted a little more room in the impact column than in the other one.
Row 14. A contractor keeps their login to the billing console for several weeks after the engagement ends. Likelihood: band four. Impact: band three.
Row 31. A copy of the production customer database is restored into a staging environment with no access control on it, and nobody deletes it. Likelihood: band two. Impact: band six.
The register sorts on one column, and that column reads twelve against both rows. They sit beside each other, indistinguishable, and the document is saying they are the same size of problem.
You have Thursday afternoon.
Predict
A band is an ordering. Band four is more likely than band three, and that is the whole of what the digit carries.
Multiplication needs more than an ordering. It needs equal steps and a real zero: a scale where four is twice two, and where the distance from band two to band three is the same distance as band four to band five.
Read the register's own definitions and the steps are visibly unequal. Once in a few years up to once a year or so is a factor of about three. Several times a year up to routinely is a factor of ten or more.
Nobody has ever claimed those two steps are the same size. The product presumes it anyway, in silence, every time the column recalculates.
Five likelihood bands against six impact bands is thirty positions a risk can occupy. Put every position through the arithmetic and seventeen distinct numbers come out the far side. Thirteen of the thirty lose their identity on the way through, and the column cannot say which thirteen.
Twelve is produced three separate ways on this scale, and the three are not variations on a theme. Four-and-three wants somebody to run a leavers process. Two-and-six wants a rule about where production data is allowed to be copied. Three-and-four is a row this organisation has not written down yet, and when they do it will sort into the middle of the other two.
Something people say
“Multiply likelihood × impact, sort descending, and work down the list from the top.”
Which part of that sentence is the part that fails?
The pair of bands supports a surprising amount on its own.
It sorts within a column: every band-six impact together, however rare each one is. It groups: rare and unsurvivable in one pile, frequent and survivable in another, and the two piles get different kinds of attention from different people. And it gives you a sentence you can say out loud — this one is rare and would end us, that one happens quarterly and costs us a day — which carries the information the digit threw away.
What the pair does not support is one ordering of the whole register. That is not a gap in your method. A register of thirty risks does not have a single correct reading order, and every tool that hands you one has invented it.
Check
The coloured grid started as a communication device, and at that job it is good. Somebody has to tell a board about thirty risks in four minutes, and a grid with a red corner does that in one glance while thirty sentences do not.
The grid never needed the number. Position in the grid is the message and the colour is the grouping. The figure got bolted on afterwards, because a grid cannot be put in a spreadsheet column and a number can, and because a number can be reported as having gone down since last quarter.
That is the whole lineage. A visual aid that worked, exported into a format that could not carry it, and arithmetic invented to fill the gap.