The line pinned on Drucker
“What gets measured gets managed” usually carries Peter Drucker’s name. In a 2019 post Danny Buerkli writes that Drucker never said it, and points to the Drucker Institute. We opened the Institute article he links to. It is about a different line, “if you can’t measure it, you can’t manage it”; more on that below. We cannot reread everything Drucker wrote and prove the first line is not in it. What we can show is where its full version comes from.
The short form gets quoted as advice: start counting and management will follow. The full version says something else. In February 2008 Simon Caulkin, management editor of The Observer, gave it a whole column. He does not dispute that the line is true. His point is that it is often misunderstood, and he gives it in full: “What gets measured gets managed – even when it’s pointless to measure and manage it, and even if it harms the purpose of the organisation to do so.” Drucker’s name does not appear in the column.
Caulkin goes back to 1956. That year V. F. Ridgway published “Dysfunctional Consequences of Performance Measurements” in Administrative Science Quarterly. In Caulkin’s summary, Ridgway described what comes of managers reducing as many of their concerns as possible to numbers.
The column’s example sits close to sales. A call centre counts rings, how long calls take, what they cost and how many calls a person makes an hour, and the figures can be on a manager’s desk the same evening. But they describe the level of activity. They do not show whether the call centre is doing its job: a low cost per call means nothing if most of the work is handling complaints about poor service.
Goodhart’s law is not in Goodhart’s words
The second line: “When a measure becomes a target, it ceases to be a good measure.” It goes by the name of Goodhart’s law.
Charles Goodhart is an economist. In 1975 he wrote about monetary management in the UK, in a volume published by the Reserve Bank of Australia. His wording is drier: “Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.” We quote it from a 2021 article in the Journal of Graduate Medical Education that cites that volume.
The famous wording is Marilyn Strathern’s, from a 1997 paper on audit in British universities (European Review, vol. 5, p. 308). Two sentences later she notes that Keith Hoskin calls this Goodhart’s law, after Goodhart’s observation on instruments for monetary control.
The two versions differ in more than style. Goodhart describes a statistical regularity that held while nobody leaned on it. Once it is used as a lever of control, it comes apart. Strathern retold this as a rule about targets and measures. In that form it fits on a slide, and it is easy to forget where it came from.
Deming said the opposite
“If you can’t measure it, you can’t manage it” is credited to Drucker or to W. Edwards Deming, depending on who is quoting. On Drucker, the answer came from the institute that bears his name. In 2013 the Drucker Institute’s website ran a piece by the neuroeconomist Paul Zak. Zak writes that Drucker never said it, and that someone from the Institute had pointed out his mistake. The piece now survives only in a web archive. Deming, in The New Economics, says the reverse: “It is wrong to suppose that if you can’t measure it, you can’t manage it – a costly myth.” The sentence is quoted by the Deming Institute.
Another line travels with Deming’s name on it: “Every system is perfectly designed to get the results it gets.” In a 2015 piece published by the Institute for Healthcare Improvement (IHI), Earl Conway and Paul Batalden explain where it came from. The original belongs to Arthur Jones of Procter & Gamble, and it was about organisations. Batalden swapped “organisations” for “system”, and that is the version that spread.
Of all the lines here, this one is the most practical. If a qualifier delivers junk leads, it is probably not broken. It is set up to deliver junk leads.
Not everything that can be counted counts
“Not everything that can be counted counts, and not everything that counts can be counted” is usually attributed to Einstein. Quote Investigator found the full wording in the sociologist William Bruce Cameron’s 1963 textbook Informal Sociology. Cameron had written the first half in an article back in 1957.
The earliest link to Einstein that Quote Investigator found dates from 1986. A business book from that year claims he chalked the line on the blackboard in his office at Princeton. The book offers no evidence, and by then Einstein had been dead for 31 years.
For a sales funnel, the second half is the useful one. The number of incoming messages is easy to count. Whether a person is ready to buy this month is not. A system judged only on the first slowly stops noticing the second.
How a metric spoils an AI lead qualifier
Radar is our product: it collects public requests and qualifies them with AI. Any qualifier turns a stream of messages into a number, and that number immediately stands a chance of becoming the target.
Suppose a pilot’s success is written down as “qualified leads per week”. The model does not change because of that. The people tuning it do. The cut-off drops by a couple of points. “Showed interest” is added to the signals. The model’s instructions gain a line: “if in doubt, count it as a lead”. Each change looks reasonable on its own. The number goes up, and there are not necessarily any more meetings.
A bot that answers customers goes the same way. Measure the share of messages that got a reply, and the winning move is to reply to everything: spam, “thanks”, a message meant for someone else. The share creeps towards 100%, and no more customers turn up.
That is Goodhart’s observation in applied form. In our piece on cold outreach in the UAE we published a cut-off of 25 points, checked on a test sample. The link between “score above the cut-off” and “looks like a buyer” holds while the cut-off is an instrument. Once someone has to report on the lead count, nothing protects that link any more.
There is a second cost. In a 1971 talk Herbert Simon put it this way: information “consumes the attention of its recipients. Hence a wealth of information creates a poverty of attention”. A qualifier exists to save a manager’s hour. A bloated lead list eats that hour. The result is what Caulkin described: measurement working against the purpose.
Which pilot metric to write down before the start
Our pilots run for two weeks, and the success metric is written down before the start. For a lead qualifier, everything above comes down to a few rules.
The metric sits further down the funnel than whatever the system produces. A qualifier produces leads, so what gets measured is meetings that actually took place. Two weeks may not yield many meetings. In that case it is more honest to record the nearest step the customer takes: replied to the manager, agreed to a meeting. And to write down, in the same place, the date when we will check whether the meeting happened.
Metrics come in pairs, so that one makes the other hard to game:
- leads per day, next to the share the manager confirmed as genuine after checking;
- share of messages answered, next to the share of conversations where the customer wrote again;
- time to first reply, next to the share of replies the manager did not have to rewrite;
- meetings booked, next to the share of meetings that took place.
Push one number up at the expense of its pair, and it shows. We pair metrics the same way in sales automation: six funnel measures, each of them measured “before” first.
Once a week someone reads a sample by eye. Not only the accepted requests but the rejected ones too: nobody complains about those, so errors live there longer. In the UAE piece we described a scoring bug that quietly kept half the leads the engine should have found out of the digest. We caught it before the first live run. Without going through specific discarded messages, it would have looked like “not many leads in this market”.
One last rule, after Deming: what is hard to count does not get dropped from the pilot. The quality of a conversation cannot be put into a number, but it can be read. That is what the manual sample is for.
What we don’t know
We opened and checked Caulkin’s column in September 2026. We read the Drucker Institute’s 2013 article in a web archive copy: the old address now redirects to druckerplus.org. That article clears Drucker of “if you can’t measure it, you can’t manage it”. It says nothing about “what gets measured gets managed”, and we found no source in which the Institute clears him of that line too. On that point we rely on Buerkli alone. We do not know who first attached the line to Drucker. We have not read Ridgway’s 1956 paper in full and know it only through summaries.
We did not open Goodhart’s 1975 original; his wording is taken from the 2021 article. We have not checked the page number of the Deming quote against the book. 1986 is the earliest Einstein link Quote Investigator found. An earlier one may yet turn up.
We have no measurements of the Goodhart effect on our own scoring. The rules above describe how we set up a pilot. How they affect sales, we have not measured yet.
If you want to see what gets counted in your funnel and which of it is easy to game, describe one process in the quiz on the main page. We will send a breakdown within 48 hours, with no call needed.
