II. Above the Ladder Essay 7
Perfectly Optimized, Pointless
Purpose, and what a system is actually for
The bell in a school rings at eight forty-five. It has rung at eight forty-five every morning for eleven years, with the reliability of a well-made machine. The clock behind it is synchronised to the network. Nobody has to do anything. The bell is a model of a system that works perfectly, and it is worth asking the uncomfortable question: works perfectly at what?
Because the bell does not know what it is for. It does not know that it should mark the end of registration, or the start of the first lesson, or that the second bell is the one that matters because the staff are late out of the staff room. The bell is indifferent. It will ring just as accurately at a time when nothing is happening, and it will ring just as accurately for a purpose that has quietly stopped existing.
A system has no opinion about what it is for, and a loop has no opinion either. Purpose does not live in the machine. It lives outside, in the judgement of somebody who decided what working would mean. That is why systems can be exquisitely tuned and completely pointless, and why a measure can improve steadily while the thing you actually wanted stays exactly where it was.
Take a school that wants its students to arrive on time. It measures lateness. It publishes the number. It puts it on a chart in the staff room, where it is visible, which is exactly the right thing to do with a number you want to move.
For a while, the number improves. Students arrive earlier. Then a second thing happens, which nobody planned. The school now has a number for lateness, and the number has become a target, and a target changes behaviour in a specific way: people start managing the number rather than the thing. A student who would be late by two minutes now arrives six minutes early and waits in the rain, so that the register, which is taken at nine, is not affected. A teacher who would be five minutes late now arrives at eight fifty-nine and takes the register at eight fifty-nine. The bell did not move. The number improved. The students learned to be on time in a way that has nothing to do with learning to be on time.
Now the second condition. Suppose the school stops publishing the number and instead asks students directly why they arrive when they arrive. The answers are about the bus, about the length of the walk, about a door that is hard to open when you are carrying a bag, and about a teacher in a previous year who made a point of marking people down for being late. The actual purpose of the lateness system, the thing it is really doing, is to allocate a small amount of shame to whoever turns up last.
Stated purpose and actual purpose
Every system has two purposes. The stated purpose is the one that appears on the notice, in the mission statement, in the sentence somebody says when asked what the thing is for. The actual purpose is what the system reliably produces, whether or not anybody intended it. These are usually different, and the actual one is usually the more accurate description, because it is the one the system has been optimising all along.
Stated and actual purposes are not the same as good and bad purposes. A system can do exactly what it says and still fail, because what it says is the wrong thing. A school timetable says it exists to give every student the best possible access to a broad curriculum. What it reliably does is produce a set of timetables in which nobody has a free period. A customer service desk says it exists to resolve customer problems. What it reliably does is produce a measurable number of closed tickets. Both are honest descriptions of the system. Only one of them is what anybody wanted.
It is worth noticing that a stated purpose and an actual purpose can drift apart over time without anybody deciding anything. The school bought the register system to give teachers time for marking. Then the school needed attendance data for funding. Then the attendance data became the thing the register was for. The stated purpose did not change, because the stated purpose is a sentence in a document and sentences in documents are very stable.
Measure and goal
A goal is what you want. A measure is a number chosen to stand for the goal, so that you can tell whether you are getting there. The measure is an abstraction, in exactly the sense used earlier in this site. It is a compression of something you want, and, like every compression, it has a shape, and the shape does not match the thing perfectly.
That mismatch is normally harmless. A child who is learning to read is a hard thing to measure directly, so you count words read per minute. The count is not reading, but it moves when reading moves, so it is good enough. The trouble starts when the count becomes the thing people are working towards. And that happens reliably, because the count is visible and the reading is not.
There is a second reason the count becomes the target, which is that it is cheaper. Counting words per minute takes a stopwatch and a marked passage. Working out whether the child is reading with understanding takes a conversation, and a conversation does not fit on a chart, and cannot be compared across a class. The measure is not chosen because it is good. It is chosen because it is easy, and the ease is what gives it power later.
This is usually called Goodhart's law, named after the economist Charles Goodhart, and the usual form of it says that when a measure becomes a target, it ceases to be a good measure. The mechanism is simple. A number that is being watched gets managed. The things that are not in the number get neglected, because neglecting them is free and being in the number is not.
Note that the measure keeps doing exactly what it was built to do. The attendance number rises. The register fills in. The chart improves. Nothing is broken. What has changed is the relationship between the number and the reason for the number, and by then the number is the only part of the system anybody is watching.
What a system is really for
There is a line often attributed to the cybernetician Stafford Beer, and it goes something like this: the purpose of a system is what it does. Whatever else you say about a system, the honest description of its purpose is the behaviour it reliably produces, because that is what it has been shaped into by all the attention it has received.
Read that way, the question "what is this for?" has a slightly uncomfortable answer. The register is for producing a complete, timely record of who is present. The report card is for producing a comparable number that can be filed. The performance review is for producing a ranking. The customer survey is for producing a score. These are not cynical readings. They are the actual functions. The question is whether they are the functions you want, and whether you have ever asked.
What if the school replaced the lateness count with a measure of something closer to the goal? Say, the proportion of students who say they know where to go when they arrive. The number is harder to collect, less reliable, and much more expensive, because it involves asking people. It will also move more slowly, and it will be harder to put on a chart. But the moment a teacher sees it, there is a question to answer: which students do not know where to go, and what would help them. The measure names the thing. The number of late arrivals did not name anything, and left to itself the school has built a system for shaming the last person through the door.
Now change the condition. What if the school kept the late count but stopped attaching a judgement to it? No chart, no comment, no use in any meeting. Just a record. A measure that is never a target stays a rough and honest approximation, because nothing in the system is rewarded for improving it. Cost is the only cost, and the cost of collecting a number nobody looks at is a great deal lower than the cost of a number everybody looks at.
"A system with a clear purpose cannot go wrong." Purpose statements are usually the least examined part of any organisation. They are written once, in good weather, by people who are certain, and then they are carried for years while the system underneath changes completely. A slogan is not a purpose. A slogan is a sentence that has stopped being checked against anything.
The other misconception: "if the measure is accurate, the goal is served." An accurate measure of the wrong thing is worse than a rough measure of the right thing, because it buys false confidence. The count of late arrivals is accurate. It just is not what anybody cares about, and its accuracy makes it harder to notice.
Who it serves
Every system serves somebody, and the somebody is usually not who thinks they are being served. A school timetable serves the constraints of the building first, the preferences of the senior staff second, and the students third. A delivery service serves the cost per drop, and the customer is the residue. A customer service department serves the reporting requirement, and the customer is what is left after the reporting requirement is met.
Asking who a system serves is not a cynical question. It is a way of finding out whether the answer matches the stated purpose, and the answer is often surprising in a way that changes what you do. If the timetable serves the building, then asking for a change means asking about rooms and about timetablers, not about the students. If the service serves the cost per drop, then a complaint about waiting is a complaint about a design decision someone made on purpose, somewhere, with a reason.
Purpose as slogan
When a stated purpose stops being connected to any decision, it becomes a slogan, and slogans are not harmless. They cost attention, they make disagreement look like a failure of values, and they make it harder to have the useful conversation about what the system is actually for and who it actually serves.
The test of a live purpose is simple: it changes what somebody does on an ordinary Tuesday. If changing the purpose would change nothing about what happens in the building on an ordinary Tuesday, the purpose is a poster. A live purpose is expensive, because it constrains the options. It says no to some things that would otherwise be attractive, and those refusals are the proof that it is real.
There is a way to make this test sharper without a single meeting. Ask the person who does the work what they would do differently if the stated purpose were true. If the answer is nothing, the purpose is decorative. If the answer is a list of things they would have to stop doing, the purpose is alive, and it is also costing something, and the cost is where you should look if you want to know whether it is sustainable.
Limits
Purpose is the least tractable thing on this ladder, because it cannot be derived from the situation. It comes from outside. That is not a weakness, it is the definition. A system cannot tell you what it is for, because what it is for is a decision about which outcomes matter, and a decision about values is not something a system can read off its own structure. This is the one level where the ladder runs out and you need something else, which is people who are willing to say what they want.
Second limit: there is no view from nowhere. Every purpose is somebody's purpose. A system optimised for one group's goals will, in some way, cost another group something, and the cost is often invisible precisely because it is not in the system. A timetable that is perfectly efficient for the school may be very expensive for one student who now has four lessons in a row and no way to eat.
Third limit: stated purposes are hard to change, and they are hardest to change when they are most wrong, because a purpose that has been in place for a long time has an constituency. It has shaped jobs, timetables, buildings, and expectations, and asking what else it could be for is asking people to give up something that has become theirs.And a fourth limit, which is the plainest one. Stating a purpose does not produce it. A mission statement is a description, not a mechanism. Somewhere between the intention and the ordinary Tuesday there is a chain of decisions, each one made by somebody who was not in the room when the purpose was agreed, and each one pulling slightly. The purpose survives only where somebody keeps doing the work of keeping it true.
Stuck
You are stuck at the purpose level when you are arguing about what something should be for and nobody is able to say what anybody actually wants. The conversation has become a matter of values, which means it has become a matter of preference, which means it cannot be settled by evidence, which is why it goes on.
The way out is to make the question factual. Not "what should this be for", but "what does this reliably produce, and who does that help, and who does it cost". Those three questions can be answered by looking. And once they are answered, the purpose conversation gets much shorter, because you are no longer discussing abstractions. You are discussing a specific arrangement with specific costs, and those can be traded against each other.
Sources
- Goodhart's law is named after the economist Charles Goodhart. The wording used here is the common paraphrase rather than a quotation, and the underlying point about measures becoming targets is widely discussed in economics and in public administration.
- The line that the purpose of a system is what it does is generally attributed to Stafford Beer, the British cybernetician. The exact wording varies between sources and should be checked before quoting.
- All examples are hypothetical or ordinary operational situations. No specific studies or incidents are cited.