The scene is familiar in many companies. For a year the same slide goes up every week. Traffic top left, engagement top right, reach below. All three numbers green, all three with an arrow pointing up, and the meeting always opens on the same sentence: we’re doing well. That year the company loses money.
The familiar explanation for this familiar story is that the wrong metric was chosen. But not one number on that dashboard is wrong. Traffic really has risen, engagement really has climbed, reach really has widened. The numbers are true. What has stopped being true is the promise those numbers once made.
Every metric is a proxy. Nobody wants traffic for its own sake; traffic is wanted because for a period it was the cheapest and fastest herald of a sale. Engagement stands in for interest, reach for awareness, retention for satisfaction. The link between proxy and thing is strongest at the moment it is established, and for exactly that reason it becomes invisible: before long nobody speaks of traffic as a proxy for sales but as the target itself. And on the day the link breaks, no warning is issued. The number stays where it is, keeps being measured, keeps being reported, stays green. It has merely stopped saying anything.
This blindness has a name. In 1972 Daniel Yankelovich set out the measuring habit of organisations in four steps: first measure what is easily measured; then disregard what is not easily measured; then presume that what is not easily measured is not important; and finally declare that what is not easily measured does not exist. In the years that followed, the chain came to be named after the defence secretary who reported the course of the war in Vietnam through casualty counts. McNamara’s error was not that he loved data; it was that he mistook the place where the data fell silent for a place where nothing existed. America won very nearly every count in that war.
The expiry date of a metric
Inside companies this is lived not as a catastrophe but as a comfort. The stretch in which the proxy comes loose from the thing is, for most teams, the most peaceful stretch they have: targets are met, dashboards are green, nobody brings bad news. The discomfort surfaces only on the cash side, in a quarter when everyone is looking elsewhere, and at that point the discussion locks onto the same place every time: what do we do to grow the number. But that is not the question. The question is what the number still promises.
The only inexpensive way to see this is to test the proxy against the thing itself, periodically. Did revenue rise while traffic rose? Did purchases rise while engagement climbed? Did customers actually stay while retention improved? When two curves stop moving in the same direction, what you have is not a performance problem but a measurement problem, and the two bear no resemblance to each other. The first is solved by more effort, the second only by a new definition. Choosing the wrong one means running the team into the ground and changing nothing.
The moment of the break has two faces, and both mislead. In the first the number falls and the business does not: a publisher whose audience is eroding can grow its revenue in the same year, because it has already detached that revenue from that number. In the second the number rises and the business does not: a company winning account after account can stay on the same profit, because every new account is a copy of an unrepaired model. Whoever looks at the first picture panics; whoever looks at the second celebrates. Both are products of the same error: looking at the number as though it had taken the place of the thing it represents. And telling which of the two has happened to you is not a matter of measuring the number more often; it is a matter of measuring what the number promised.
A metric is dangerous not because it is wrong, but because nobody notices it has stopped being right. Companies do not sink on invented numbers; they sink on expired ones.
This is not a call to stop measuring. I am saying that measurement is maintenance work, not a system you install and forget. A metric is a photograph of the world at the moment it was set; when the world changes the photograph does not, it merely loses its likeness. Most organisations review their metrics once a year, but the thing needing review is not the metric’s target, it is what the metric is still a proxy for. The difference between those two is everything a company thinks it knows about itself.
The most expensive number is the one nobody questions, because the one thing never questioned inside an organisation is the thing everyone once knew to be true. Look at your dashboard: which number is still making a promise nobody remembers last checking?