Systems · Operations
A KPI dashboard that can be argued with
Four numbers on top, and each one carries the thing most scorecards leave out: how far it is from the line that matters. The value is the least useful part of a tile.
Synthetic data Every figure and name on this page is invented.
What was failing
Most scorecards report the value and colour it. On-time delivery of 95.30% against a 95% floor reads as green, so nobody asks about it. It is thirty basis points of cushion on a falling line, which is one bad week.
The second failure is quieter: colouring by sign instead of by favourability. A rising cost drawn in cheerful green trains a room to skim past the one tile that matters.
What I built
A four-tile board where every tile reports its margin to the threshold, not just its value. Each metric declares which direction is good (higher, lower, or closer to zero), and every status is computed from that, never picked by hand. A falling cost renders green; a rising one renders red.
The on-time tile is the case in one picture: above its floor, trending down, marked At risk in amber with the margin printed underneath. The fourth tile is the one I would defend hardest. It is grey, it says Review, and its note reads no defensible band yet: the metric is real and the target is not, so the board admits it instead of guessing.
What changed
The question the room asks. A green value invites nothing; a margin invites "what would it take to lose this?" That is the whole argument for the treatment, and the longer version, about agreeing the arithmetic before arguing the result, is in Pin every denominator.
What you can inspect
The board above, rebuilt on invented data with the layout and devices I actually use. The instruments underneath a scorecard (provenance lines, confidence-tagged money, weighted progress and a board that retires itself honestly) are on the operating instruments page.
What isn’t proven
The bands are proposed defaults, tuned against actuals rather than defended forever, and one tile has no band at all. This page shows a treatment on invented data; it is not evidence that a particular board decided better because of it, and I don't claim that here.
A scorecard is the top of the stack, not the whole of it. If the board and the ledger disagree, the ledger is right, and each tile should link to where its number was computed. The reconstruction doesn't show that link.