Jon's Job Blog

PPG announced, one year, that it had been recognized among the Best Places to Work in IT by Foundry’s Computerworld. The ranking was number eighteen among large organizations. The company’s CIO provided a quote about excellence in digital IT and the focus on people development, along with the commitment to harnessing the power of AI and the cloud environment. The announcement was issued through Business Wire. It was picked up by the financial press. It did what corporate awards announcements are designed to do, which is to produce a piece of durable third-party validation the company can then cite in recruiting materials, press releases, and internal communications for at least the next twelve months.

The following year, PPG announced that it had been recognized among the Best Places to Work in IT by Foundry’s Computerworld. The ranking was number thirty-eight among large organizations. The company’s CIO provided a quote about excellence in digital IT and the focus on people development, along with the commitment to harnessing the power of AI and the cloud environment. The announcement was issued through Business Wire. It was picked up by the financial press. The language was so similar to the previous year’s that a reader comparing the two side by side could be forgiven for missing the twenty-position drop entirely. The press release acknowledged the appearance on the list. It did not mention the rank.

I am going to spend this essay on the gap between those two press releases, because the gap is the subject the series has been circling since its opening essay, and because the gap is the thing corporate metrics are most useful for concealing. This is not a personal grievance about a specific ranking. The Computerworld list is not important in itself. What is important is that the same organization, using the same language, celebrated two rankings that described substantially different realities — and that the language was designed to function equivalently in both cases, which meant it could not function as measurement in either. When the vocabulary of celebration is identical regardless of the underlying result, the vocabulary is no longer reporting on the result. It is producing a result of its own. This essay is about what happens when organizations lose the ability to tell the difference between being measured and being good.

What We Measure Determines Who We Value

There is a long tradition of intelligent people pointing out that metrics, once they become targets, stop working as measurements. The economist Charles Goodhart, writing in 1975 about monetary policy, observed that any statistical regularity used for control purposes will collapse once pressure is placed on it. The anthropologist Marilyn Strathern restated Goodhart’s insight more sharply two decades later: when a measure becomes a target, it ceases to be a good measure. The historian Jerry Muller, in The Tyranny of Metrics, described the broader pattern as metric fixation — the belief that what cannot be quantified is not real, and that the things that can be quantified should be optimized regardless of whether the optimization distorts the underlying activity.

The intellectual lineage here matters because it predates the contemporary corporate context by half a century. The point Goodhart and Strathern and Muller are making is not a critique of modern management specifically. It is a description of a recurring failure mode of measurement systems under pressure. Any system that rewards a proxy for the thing it actually cares about will, over time, produce optimization of the proxy at the expense of the thing. The proxy becomes the target. The target becomes the goal. The goal becomes the metric. The metric becomes the activity. And the original thing the measurement was supposed to track recedes into the background, eventually forgotten by the people operating the system and invisible to the people reading its reports.

Wells Fargo is the textbook case. Beginning in the early 2000s and accelerating through the early 2010s, the bank set aggressive cross-selling targets for branch employees. The metric was accounts per household. The goal was eight. Eight was chosen because it rhymed with the marketing slogan — eight is great. Employees who missed the target faced discipline. Employees who met it received bonuses. Employees who exceeded it were celebrated in internal communications. Over the course of roughly a decade, employees opened approximately two million accounts customers had not authorized. They forged signatures. They created fake email addresses. They moved customer funds into accounts the customers did not know existed, generating fees the customers then paid for services they had not requested. The bank eventually paid $185 million in fines, fired approximately 5,300 employees, and retired the eight-is-great slogan. The CEO who presided over the period retired with a compensation package that was reduced, though not eliminated, by the board. The targets had been met. The metric had been satisfied. The organization had produced exactly what it measured.

Amazon warehouse injury rates are a different version of the same mechanism. For years, the measured quantities in an Amazon fulfillment center were delivery completion, pick rate, packing speed — the operational metrics tied to the promise of fast shipping. The unmeasured quantity was worker injury. Independent analyses of OSHA data have repeatedly found Amazon injury rates well above the warehousing industry average, in some reporting periods roughly double. The company’s internal response was to produce additional safety programming — wellness initiatives, stretching breaks, ergonomic training — without reducing the operational targets that were producing the injuries. The programs addressed the complaint. They did not address the cause. Because the cause was the measured quantity, and the measured quantity was the one the organization was designed to optimize, addressing the cause would have required changing what the organization understood itself to be doing. The programs were cheaper.

General Electric under Jack Welch institutionalized the most influential version of this pattern in the form of stack ranking, the practice of sorting employees into performance tiers and systematically eliminating the bottom of the distribution. Stack ranking was celebrated for a generation as a best practice. By the mid-2000s, GE itself had retired it, acknowledging that it had produced predictable dysfunctions: managers gamed the rankings to protect favored employees, high-performing teams were punished because every team had to produce a bottom tier regardless of actual performance, and collaboration collapsed because cooperation with a colleague was structurally indistinguishable from helping that colleague outrank you. The practice continued at other companies for another decade after GE abandoned it, which is the normal lag between a management fad being discredited at its origin and being retired elsewhere. The metric had been designed to identify underperformance. What it produced was the behavior that would be rewarded by the ranking system, which was not the same thing.

What these three cases have in common is not bad intent. They are not stories about villains. The Wells Fargo executives who set the cross-selling targets did not want their employees to commit fraud. The Amazon leadership that optimized warehouse throughput did not want workers to get hurt. The GE managers running stack ranking did not want to degrade collaboration. The metrics produced those outcomes not because anyone at the top wanted them, but because the measurement systems were structurally indifferent to them. What the system measured, the system got. What the system did not measure, the system gradually destroyed. The intent of the designers is irrelevant to the behavior of the apparatus.

The Limits of Quantification

Robert McNamara is the cautionary figure here, though his fallacy predates him. McNamara, who ran the Defense Department during the early years of the Vietnam War, was an early champion of quantitative management in government. He brought the analytical methods he had used to run Ford — statistical process control, operations research, systematic measurement — into military planning. The system he built counted bombs dropped, sorties flown, enemy fighters killed. The things it did not count, because they resisted quantification, were the political legitimacy of the South Vietnamese government, the will of the Viet Cong, the growing skepticism of the American public. The war was, by the measured metrics, going well until approximately the moment it was obviously lost. The sociologist Daniel Yankelovich later formalized the McNamara fallacy in four steps: measure what can easily be measured, disregard what cannot, presume what cannot be measured is not important, and presume what cannot be measured does not exist. The fallacy describes a cognitive descent, not a single mistake.

Corporate metric systems reproduce the pattern with remarkable consistency. The things that can be measured — headcount, spend, throughput, survey response rates, program enrollment — receive disproportionate attention, not because they are necessarily the most important variables but because they are the legible ones. The things that cannot easily be measured — the quality of the working relationship between a manager and a report, the institutional knowledge an employee carries, the tacit understanding of why a particular system behaves the way it does — recede into the background of organizational awareness. Eventually they are forgotten. Eventually the people making decisions begin to believe that the things they cannot measure are not real, and the things they can measure are the whole picture. The organization develops a species of self-inflicted blindness that correlates exactly with the categories its reporting systems cannot capture.

There is a corollary to this that matters for how one experiences a dysfunctional organization from the inside. When an individual employee reports that something important is broken, and the broken thing is not something the organization’s metric systems can see, the report has no receiving surface. The employee’s complaint gets categorized as a personal issue — morale, fit, attitude, whatever semantic bucket the HR vocabulary provides for problems the company does not want to classify as organizational. The complaint does not escalate. It does not enter any report. It does not change any decision. And then, some time later — often a substantial amount of time later — an aggregate metric the organization does care about shifts in the direction the individual had been describing. A ranking drops. A retention number spikes. A customer-satisfaction score softens. At that point, the organization notices. The complaint the individual had filed was a leading indicator. The indicator was ignored because the reporting system had no column for it. The aggregate metric, when it finally arrived, was a lagging confirmation of what the person inside the system had known months or years earlier and had been told was their personal problem.

The Computerworld ranking is an example of this. The 2025 number was eighteen. The 2026 number was thirty-eight. The drop happened over a single measurement cycle. The things that produced it — whatever mix of attrition, internal survey results, program changes, structural decisions, and relocation of roles across geographies — did not happen in a single cycle. They happened over the preceding years, and they were being experienced, reported, and ignored the entire time. The ranking was the point at which the lagging indicator caught up with the leading ones. It was not news to the people inside. It was news to the people reading the press release.

The Fortune Most Admired Companies list operates on a similar principle but more nakedly. The methodology is explicit: Fortune surveys executives, directors, and financial analysts about which companies they most admire within specific industries. The respondents rate the companies on criteria like innovation, people management, long-term investment value, and social responsibility. The output is a ranking. What the ranking is actually measuring is the opinion of the executive class about other members of the executive class. It is a prestige signal measured by the people whose prestige is being signaled. A respondent who works adjacent to a company, has mutual board members with its executives, or has met its CEO at a conference is asked to rate that company’s social responsibility. The measurement is designed to produce a consensus within a bounded community. What it produces, reliably, is exactly that. Companies that appear high on the list have, generally, the executive credibility the list measures. Whether they are well-run, humane, effective, or serving their customers is a separate question the methodology is not designed to answer.

I am not saying these rankings are fraudulent. They are not. They are doing exactly what their methodologies describe. The problem is that the public reception of them does not match what they measure. A company that ranks eighteenth on Computerworld’s list is not, by that fact, a better place for an IT professional to work than a company that ranked twenty-second. The ranking depends on a questionnaire submitted by the company, reviewed by a panel, and compared against similarly self-reported questionnaires from other companies. The rank reflects the relative quality of the questionnaires as much as the relative quality of the workplaces. When a company drops twenty positions in a year, the most charitable reading is not that the workplace degraded by twenty positions’ worth over twelve months. The most charitable reading is that the gap between the questionnaire and the reality finally became too wide to maintain.

Institutional Credibility Collapses Before It Falls

Not long ago, a man was shot and killed in midtown Manhattan. The victim was the CEO of a major American health-insurance company. The event is a matter of public record. What was not, before it happened, a matter of public record was the scale of the ambient public hostility toward the institution the victim had led. The shooting revealed it. Commentary in the aftermath — not the celebration, which was a small and loud minority, but the grim, tired, shoulder-shrugging commentary of ordinary people who had been denied claims, billed incorrectly, held on hold for hours, forced to choose between procedures — was what surfaced. The public response to the killing showed that a significant share of the population had been quietly writing off the legitimacy of the insurance system for years. The executives of that system had not noticed. Their metrics had not noticed. Their metrics could not have noticed, because the metrics their system optimized for — shareholder value, denial rates, administrative efficiency — were not capable of distinguishing a company its customers trusted from a company its customers loathed but continued paying because there was no alternative. As a business, the system was performing well on every dimension the reporting described. As an institution, it had lost the confidence of the people it nominally served. The metrics could not see the second fact until the first fact was broken open by an event outside the system altogether.

I am not drawing a moral equivalence and will not belabor the example. The event was a homicide. The response to it was, in many quarters, a genuine moral failure — a cheapening of human life dressed as class consciousness. I am using it because it illustrates, with unusual clarity, a pattern the rest of this essay has been describing in gentler cases. Institutional credibility does not collapse on the day the event occurs that makes the collapse legible. Credibility collapses over years, silently, in the gap between what the institution measures about itself and what the people subject to the institution experience. The event is the moment the gap becomes undeniable. By the time the gap is undeniable, the institution’s internal reporting has been telling it a reassuring story for so long that the sudden confrontation with public reality registers as an aberration rather than a reckoning.

Engagement surveys are a smaller version of this. Most Fortune 500 companies run an annual or semi-annual engagement survey administered by a third party. The vendor produces an aggregate score, benchmarks it against peer companies, and delivers a report to leadership. The score is generally treated as a measurement of organizational health. It is not, and cannot be, a measurement of organizational health. It is a measurement of what employees are willing to say on a survey they do not fully trust to be anonymous, filtered through a vendor whose business model depends on producing reports leadership finds actionable. Scores that are too high raise questions. Scores that are too low raise different questions. The ideal score, from the vendor’s perspective, is moderately high with specific categories flagged for attention. The survey is not measuring engagement. It is producing a kind of report leadership has learned to consume. The production of the report is the activity. The measurement is ceremonial.

A Framework for Ethical Metrics

None of this is an argument against measurement. Measurement is necessary. The argument is against the substitution of measurement for judgment, and against the peculiar belief, widespread in modern management, that the existence of a number makes the number true. There are four questions any ethical measurement system has to be able to answer, and most corporate measurement systems cannot answer them.

The first is who asked for this to be measured. Every metric reflects a choice by someone, somewhere, to make a particular quantity visible. That choice has consequences. Metrics that are easy to produce from existing data tend to survive; metrics that would require new data collection or new systems tend to die in committee. The result is a measurement landscape biased toward what is already legible, which is to say, biased toward the interests of whoever built the original systems. A metric about customer wait times is a choice. A metric about employee dignity is also a choice, and it is almost never made.

The second is who benefits when the number goes up. Every metric has a beneficiary. The beneficiary is rarely the same as the nominal subject of the measurement. Employee-engagement surveys benefit HR vendors and the leadership team that gets to cite them. Training-program participation metrics benefit the training department. Safety-incident rates benefit the risk-management function and the insurance carriers setting premiums. When a metric moves, the beneficiary’s standing changes regardless of whether the underlying condition has changed. Asking who benefits is a way of seeing which interests the measurement system is actually serving, as opposed to which interests it claims to serve.

The third is what happens to the person when the number is wrong about them. This is the question most corporate metrics are structurally incapable of asking. A performance ranking that places an employee in the bottom tier has consequences — compensation, assignment, eventually termination. If the ranking is wrong about the employee — if the ranking reflects stack-ranking distortion, or the manager’s blind spots, or the hidden cost of asking hard questions in a culture that rewards ease — the person absorbs the cost. The metric does not. The organization, from its own vantage point, is functioning as designed. The person inside the number is the variable the system cannot see, and the fact that it cannot see them is the system’s comfort.

The fourth is what we are systematically choosing not to measure. This is the question the McNamara fallacy forbids. If what cannot be measured is assumed not to exist, then the set of unmeasured things becomes invisible to decision-making. But the unmeasured set is not empty. It contains most of what actually matters: relationships, institutional memory, quiet contributions, problems being flagged by people whose reports never leave the meeting where they were raised. An organization that cannot describe its own unmeasured set is an organization that has lost the ability to be surprised by itself, which is another way of saying it has lost the ability to correct.

None of this is unique to any one company, which is the point. The measurement regime of modern corporate work is a shared inheritance with shared pathologies, distributed across industries and enterprises of every size because the incentives that produce them are widely distributed. Dashboards are easier than judgment. Surveys are cheaper than conversations. Rankings are more defensible than opinions. The measurement systems we have are the ones we built, and we built them because they were convenient, and the cost of the convenience is the thing the measurement systems were designed to hide.

Closing

The Computerworld ranking drop is a small example of a large pattern, and it is the one I keep coming back to because it is documented, public, and legible. A company that had ranked eighteenth ranked thirty-eighth a year later. The language celebrating both rankings was nearly identical. The CIO quoted in both press releases was the same person, using the same phrases about people development and the potential of AI. The drop was, in aggregate, a twenty-position slide over a single year, during a period when the organization was publicly celebrating its recognition on the list. The public could read both press releases. The public was not told about the delta. The delta was not the point of the releases. The releases were the point of the releases.

Metrics are moral instruments. They encode what the organization considers worth knowing, which means they encode what the organization considers worth doing, which means they encode what the organization considers worth being. An organization that celebrates a thirty-eighth-place ranking in the same language it used to celebrate an eighteenth-place ranking is an organization that has forgotten the difference, which is a particular kind of forgetting. It is not absent-mindedness. It is a byproduct of a measurement culture that has produced, over time, a workforce of people who have learned to read the number, produce the quote, and issue the release regardless of what the number actually says. The practice of issuing the release becomes the measured activity. The number in the release becomes a detail. The reader who notices the detail is the one the system did not design for.

Measuring what matters means redefining what success looks like, which in practice means admitting that most of what matters cannot be captured in a dashboard. The people who build the dashboards are often the first to know this — you cannot write a DAX measure for a year without discovering, intimately, the difference between what the data can show and what the data is being asked to stand in for. The problem is not that dashboards exist or that some people’s work is to produce them. The problem is that the organization learns to trust the dashboard more than it trusts the people inside it, including the people who made the dashboard in the first place. This is not a scandal. It is the architecture. The architecture produces the press releases, and the press releases produce the appearance of legitimacy, and the appearance of legitimacy is what the organization has been selling to itself and to its customers and its investors and its employees for long enough that it can no longer tell the appearance from the thing.

Sometimes the dashboard is the last to find out.

← Essay 16: Purpose After Burnout Essay 18: How ‘Professionalism’ Silences the Neurodivergent → View Full Archive