[{"data":1,"prerenderedAt":18},["ShallowReactive",2],{"post-why-kpi-deserves-serious-skepticism":3},{"slug":4,"title":5,"excerpt":6,"content":7,"tags":8,"readingTime":15,"publishedAt":16,"published":17},"why-kpi-deserves-serious-skepticism","Why KPI deserves serious skepticism","A note on two failure modes I have been watching this year — the version of KPI where the form has no teeth, and the version where it has too many. I do not think either of them is working the way the people who designed them hoped. This is more of a thinking-out-loud than a conclusion; I am still chewing on it.","\n## Two failure modes, one shared logic\n\nI have been watching a quiet thing happen this year.\n\nMy group's quarterly KPI review was last week. Three pages of form. An afternoon of meeting. The numbers were read out. The boxes were filled in. At the end, the HR partner thanked everyone for the rigour of the process, the manager thanked HR for the rigour of the process, and we all left. The numbers had no effect on our performance pay. They had no effect on our ratings. They had no effect on our promotions. The form was a record, the record was a checkbox, the checkbox was a sign that the company had a process.\n\nI have come to think of this as one version of the KPI problem. I have been calling it failure mode A in my head, mostly as a placeholder for the version I actually live in. It is the version most people are not talking about, because it is so easy to dismiss. The form is filled. The boxes are ticked. The system is \"running.\" Nothing happens. The pattern, as I see it, is that the system looks like work, fills the calendar, and produces no creative output. It is *dirty work* — but for the people whose job it is supposed to measure.\n\nI have written about this kind of work elsewhere — [[why-dirty-work-deserves-serious-consideration|a piece I called \"dirty work\"]] — the work that looks like work, that returns every quarter in slightly different forms, that never quite goes away. I framed it as a frontline problem back then: UTM links, waitlist setup, shipping tracking, data sync, analytics pipelines. Looking back, I think that frame was incomplete. I should not have stopped there.\n\nThe current company's quarterly KPI is dirty work for the people who have to fill it in, and dirty work for the people who have to read it. It is dirty work for the manager who has to consolidate the numbers, and dirty work for the HR partner who has to file them. The pattern, I think, tends to close the same way it closes at the frontline — one task at a time, until the role is defined by the form and the actual work is what gets squeezed into the cracks.\n\nI would describe failure mode A this way: the form is hollow. The system has no teeth. Nobody is paying attention, but everybody has to act like they are.\n\n## The other failure mode\n\nFailure mode A is the one I live in. Failure mode B is the one most other people seem to live in, at least in the conversations I have had. It is also the version most of the literature I have read is actually about.\n\nFailure mode B, in my reading, is the version where KPI is welded to everything. It is tied to performance pay. It is tied to the rating. It is tied to the promotion. Every quarter, the numbers are added up, weighted, run through a formula, and the formula decides what you make and how far you go. The system is not hollow. The system has teeth. The teeth are, presumably, the point.\n\nThe thing I keep coming back to, watching this version operate, is that the teeth tend to cut the wrong thing.\n\nHere is what I have watched happen, in companies that run this version of KPI, over three jobs and a lot of conversations with friends who run the same kind of system in theirs. None of this is rigorous. It is closer to field notes.\n\n**Sales floors turn into contested territory.** When the only thing that matters is the number, the number becomes the territory. Two sales leads do not share a customer — they fight over the customer, because the customer's contract decides whose year was better. The manager is supposed to mediate, but the manager is also being measured, so the manager picks a side. The customer can feel it. The customer can feel that the sales lead is no longer on their side. I think the customer is right, and I think the customer leaves. The churn is then attributed to \"sales execution,\" and the next quarter's KPI gets a new line item, and the spiral continues.\n\nThe spiral is what I want to name here. I do not think it is a one-time failure. I think it is a *process*. The first quarter, the line gets crossed once. The second quarter, twice. The third quarter, three times. The numbers tend to look good in the short run, because the team is fighting harder. The numbers tend to look terrible in the long run, because the customers stopped trusting the team. The KPI probably cannot see this — by construction, it only sees the quarterly number.\n\n> When a sales lead's number rewards them for closing alone, the lead who used to help another lead close has, by definition, fewer numbers than the lead who closes alone. The lead who helps gets penalized. The lead who closes alone gets rewarded. The two leads figure this out in a quarter. The collaboration ends. The customers can feel it. The customers leave.\n\nWhen one side's moral bottom line drops, I would expect the other side to drop too. The contest is not symmetrical. The contest, in my reading, is a spiral — a downward spiral — and every quarter the spiral tightens, and every quarter the line moves a little further, and I am not sure anyone inside the system notices, because the system is the thing that produced the line in the first place.\n\n**Operations rejects the work that does not score.** Some content is designed to score. Some content is not. The activity announcement. The brand piece that exists to be in the world, not to convert. The press release that has to go out because the calendar says so, not because it is going to drive a click. These tasks, by construction, do not produce the kind of data the KPI rewards. If the operator's rating is tied to the effectiveness of the content they ship, I would expect the operator, in private, to be relieved every time one of these tasks is canceled. The operator will, in public, do the work to the minimum standard. The brand will, I think, slowly hollow out. The KPI is probably going to say the operator is performing. The KPI may be right, by its own lights. The KPI is, in this case, measuring the wrong thing.\n\n> Some work is destined, from the planning stage, not to bring any data result. If the operator's rating is tied to the data result, I would expect the operator, inevitably, to reject the work.\n\n**Collaborative teams get forced into competition.** A small team is asked to ship a thing together. The team is small enough that the work is genuinely collaborative — the design touches the copy, the copy touches the engineering, the engineering touches the design. The team's output is, by construction, a joint product. But the KPI is set per person. The KPIs are similar. The KPIs are easy to compare. The team, in my experience, tends to become a collection of individuals trying to look better than the other members of the team. The so-called \"赛马\" (horse-racing) instinct that the KPI is designed to produce has nothing to race on, because the team's output is not a horse race. The KPI, I think, is producing the wrong instinct in the wrong place. The team, in private, probably knows this. The KPI, by construction, does not care.\n\nThree examples. The same logic, in my reading. The KPI is a number. The number is a proxy. The proxy is always a simplification. The simplification always misses something. The thing it misses is, more often than not, the part of the work that does not reduce to a number. The KPI then rewards the number, and the work that produces the number, and the work that does not produce the number tends to wither.\n\nI would describe failure mode B this way: the form has teeth. The teeth tend to cut the wrong thing. The system is probably not producing theatre — it is producing a slow, expensive, often invisible kind of decay.\n\n## What KPI is actually for\n\nIf both versions tend to fail, why does KPI exist?\n\nMy current best guess is that KPI exists because the manager cannot see the work.\n\nIn a small team, the manager sits in the same room as the people doing the work. The manager can see who is in flow and who is not. The manager can see who is shipping and who is stuck. The manager can see the meeting, the conversation, the off-hand comment that became the breakthrough. The manager does not need a proxy for the work, because the manager can *watch* the work.\n\nIn a large team — and I mean large in the org-chart sense, not the headcount sense — the manager cannot see the work. The manager has ten, twenty, fifty direct reports. The manager has no way to know which of those reports is producing the work that the team needs. The manager needs a proxy. The proxy, in my reading, tends to be the KPI.\n\n> If I had to put a frame around it, I would say KPI is the system that managers reach for when they can no longer observe directly. It is a workaround for a structural problem. The structural problem is that the manager has too many people to know.\n\nI think the right response to this is not to add a better proxy. I think the right response is to fix the observation problem — to reduce the number of direct reports, or to install rituals that put the manager back in the room with the work, or to trust the team to do the work without the proxy. I am less sure that \"the right response\" is the right framing here. The honest answer is that I am still working this out.\n\nMy current read on why most KPI systems fail is that they were built by managers who could observe, or by managers who could not observe but did not notice. The systems that work, I think, are the systems that are honest about being a workaround — \"this is the proxy, the proxy is not the thing, the proxy exists because we cannot see the thing.\" In that mode, the KPI is probably the least-bad option. The systems that fail, in my experience, are the systems that pretend the proxy is the thing — the systems that weld the proxy to the compensation, and then act surprised when the proxy becomes the goal.\n\n## The principle, stated tentatively\n\nHere is the principle, as I am currently holding it. It is the inverse of the observation, with the caveat that I am not sure how much of this holds outside the contexts I have personally seen.\n\n- In a team of five, the manager can watch. The KPI is a tax. The tax consumes the afternoon. The afternoon is, in my experience, the most reliable production block of the day. The team pays the tax. The team probably gets nothing for it.\n- In a team of fifty, the manager cannot watch. The KPI is a workaround. The workaround is honest about being a workaround, and the team treats the workaround as a workaround. In that mode, I think the KPI is the least-bad option.\n- In a team of five hundred, the manager cannot watch and the system has to scale. The KPI is a tool of the system, not the manager. The KPI is a tool of HR, not the team. The KPI, in this regime, is the system. The system has stopped being a proxy. The system is, for practical purposes, the thing. The thing, at that scale, has stopped being a person.\n\nI would guess that KPI makes sense only at a scale I have not worked at — 大厂, the big companies. The bigger the company, the more the proxy looks like the thing, the more the form looks like the work, the more the workaround looks like the system. The damage the KPI does is, I think, still there. It is just less visible at scale. The damage is what a five-person team would notice in a week; the damage is probably what a five-hundred-person team notices in a decade.\n\nFor a small team, I would expect the principle to invert. Quantify every employee's work into a data point. Add up the data points. Weight the data points. Compute a weighted average. Get a rating. I think the rating has, in the process, tended to reduce every employee to a number — and the possibilities that don't reduce to a number tend to disappear. The team, having been told its number, probably stops trying things that do not produce the number. The team, having been told its number, probably stops doing the work that does not reduce to a number. I would expect the team's output to get safer, less interesting, and harder to attribute to any one person. Whether the team's creativity, output, and innovation go down — that I am less sure about. I have not measured it. The effect is probably real, and probably hard to see from inside the team.\n\nThis is the same pattern I think I named in the dirty-work piece, in a different costume. The form, over time, tends to eat the work it was supposed to be a record of. The form tends to eat the manager. The form tends to eat the system. The system that cannot name this is, I think, the system that will keep filling in the form, and the form will keep eating, and the work that the form was supposed to measure will keep disappearing.\n\n## What I am not saying\n\nI am not saying KPI has no use. I am not saying performance pay is wrong. I am not saying ratings and promotions should be removed. I am saying that the proxy is, in my reading, a tool, not an end. I am saying that the system that forgets the proxy is a tool is, more often than not, the system that produces the failure modes I have named. I am saying that I think the response to a broken KPI is to fix the observation, not the form — though I am less certain about that than I was a year ago.\n\nI am also not saying the two failure modes are evenly distributed. They are not, in my experience. Failure mode A is the version most people have the luxury of ignoring, because nothing bad happens in the short run. Failure mode B is the version that produces the visible damage — the sales-floor fight, the operations hollow-out, the team that turned into a horse race. The literature I have read mostly talks about failure mode B. I have lived in failure mode A. Failure mode A, I think, is the silent killer of the manager's afternoon.\n\nI am not saying managers are bad. I am saying the system asks them to do something they cannot do, and then blames them for failing to do it. The blame, I think, is misplaced. The system is the problem, or at least the larger part of it. The system is, I think, the thing that needs to change.\n\nI am not saying every team should remove their KPI. Some teams are big enough that the KPI is probably the least-bad option. Some teams are small enough that the manager can watch, and the KPI is a tax the team should be willing to pay. The answer, I suspect, depends on the context. I am not certain enough to say more than that. The wrong answer, I think, is the form itself — the form that pretends to be the thing.\n\n## Where this connects\n\nHonestly, not to as many things as I would usually try to force. I have a habit of treating every new essay as a node in a network diagram — \"this connects to that, and to this, and also to that other thing over there.\" I am trying to stop doing that. The honest answer here is that this piece sits next to one earlier piece and not much else, and that is probably the right size for it.\n\n[[why-dirty-work-deserves-serious-consideration|The dirty-work piece]] is, in my reading, the direct precondition. The KPI form is dirty work for the manager and the HR partner — the same pattern, just on the other side of the org chart. The form looks like work, fills the calendar, produces no output. The response, in my reading, is the same response: remove the form, give the manager the conditions to actually see the work, and let the people who are supposed to be doing the work do the work.\n\nThat is about it, as far as direct connections go. The other systems I have written about — the bio link surface, the agent pipeline, the 16:00 break — they are connected to this piece in the way that everything in a working life is connected to everything else, which is to say, not very specifically. I have stopped trying to make every new essay fit a network diagram. Sometimes an essay is just an essay.\n\n## What the literature already says\n\nI did not invent any of this. The framing here is borrowed, openly, from work that has been around much longer than I have.\n\n**Eliyahu Goldratt, *The Goal* — and the Theory of Constraints behind it.** Every system has exactly one bottleneck at a time, and the right response to a bottleneck is, in Goldratt's reading, never to add capacity at the bottleneck itself. The right response is to identify the bottleneck, relieve it, and let the rest of the system catch up. I think the KPI system, in failure mode B, is a useful example of a system that adds capacity at the wrong place. The KPI system does not relieve the manager's inability to observe. The KPI system adds a proxy. The proxy does not observe. The proxy counts. The system, in trying to solve the observation problem, has, in my reading, created a counting problem — and the counting problem is now the bottleneck. I am less sure that counting harder will solve it. I am fairly sure that the team that tries will end up with a more accurate number and, possibly, a worse company.\n\nMihaly Csikszentmihalyi's flow research and Daniel Pink's *Drive* point at the same thing from different angles — the form is hostile to flow, and the proxy replaces the purpose it was meant to measure. I am not going to re-explain them here. The point is, I think, the same point.\n\nThese are not new ideas. They are old ideas, applied to a problem most teams are still working through, one column at a time.\n\n## A postscript, on the dirty work for managers\n\nThe piece I wrote on dirty work, a few months ago — [[why-dirty-work-deserves-serious-consideration|why dirty work deserves serious consideration]] — named the pattern from the frontline. The pattern was: the dirty work consumes the hours of the people who are supposed to be doing the work that the team actually needs. The pattern was: the role is defined by the dirty work and the creative work is the side project.\n\nI am starting to think the pattern is the same on the other side of the org chart. The KPI form, in failure mode A, probably consumes the hours of the people who are supposed to be managing. The KPI consolidation, in failure mode B, probably consumes the hours of the people who are supposed to be leading. The dirty work for managers is, I think, the form. The dirty work for managers is, I think, the rating. The dirty work for managers is, I think, the meeting that has to happen because the form has to be filed.\n\nThe two halves of the org chart are, in my reading, the same argument. The form, over time, tends to eat the work it was supposed to be a record of. The form tends to eat the manager. The form tends to eat the system. The system that cannot name this is, I think, the system that will keep filling in the form, and the form will keep eating, and the work that the form was supposed to measure will keep disappearing.\n\nI am still working out the last part. I think the hardest part, for me at least, is admitting that the form was probably never the answer — though I would not have said this a few years ago. And then, instead of fixing the form, fixing the thing the form was a proxy for. I am less sure how to do that in practice. I am fairly sure it is the right direction.\n    ",[9,10,11,12,13,14],"Process","Org Design","Team Design","Management","KPI","Dirty work",8,"2026-08-05",true,1786114461224]