Engagement Is
Not Capability

Every leadership team we have worked with is trying to answer the same question, whether or not anyone says it out loud.

Are we going to win?

Not: are our people happy? Is our culture healthy? Those things matter, and they matter more than most executives are willing to admit. But the questions that actually keep a chief executive awake are narrower and harder than either: Can this organization move fast enough? Can it make a good decision under pressure? Can it truly differentiate itself in the market from competitors with customers? Can it absorb what is coming next?

Leaders often have plenty of conviction about the answers to these questions. What they lack is evidence. So they reach for what is on the table. And in almost every organization of any size, what is on the table is the engagement survey.

It is the only instrument most companies have that asks the entire workforce anything at all. It arrives once a year. It produces a number. The number moves up or down. And a leadership team sits around it and tries to infer something — anything — about whether the enterprise is in shape to compete.

This essay is about why that inference fails.

Not because the survey is badly built. It usually isn’t. Not because the people who run it are not talented. They usually are. It fails because the engagement survey was designed to measure how people feel, and the question the executive is actually asking is what the organization can do. Those are different things. They are related, and the relation is real, but they are not the same thing, and the first cannot be made to stand in for the second.

For a long time, the cost of that substitution was tolerable. It no longer is: the changed arithmetic of a leveraged deal, and the discovery that returns on an AI investment depend far more on the condition of the organization than on the technology being bought, have made the distance between what we measure and what we need to know expensive.

What follows begins with a disconnect — the instruments the field relies on can no longer agree with one another about whether the workforce is getting better or worse — and ends with what a better instrument would have to do.

Gallup reports that global employee engagement fell to 20% in 2025 — the lowest since 2020, a second consecutive annual decline, with no region of the world improving. Its US series shows engagement at 31%, a ten-year low, down from a 36% peak in 2020.1

ADP Research reports the opposite. Its global measure of fully engaged workers climbed from a pandemic low of 14% to 19% by 2024 — which ADP characterized as a record high across a decade of data, and the third consecutive year of growth — before holding flat in 2025.2

Both are large, methodologically serious, global instruments. Both explicitly measure employee engagement. They agree almost exactly on where we are (20% against 19%) and disagree entirely about how we got here.

The disagreement runs wider still. The Conference Board, whose job satisfaction survey has run continuously since 1987 and is the longest-standing instrument of its kind in the United States, reports US job satisfaction at 68.9% in 2026 — a record high, and the sixteenth consecutive annual gain.3 Glassdoor’s Employee Confidence Index, meanwhile, fell to 43.8% in April 2026, the lowest reading since the index launched in 2016, while unprompted mentions of burnout in employee reviews are running at two and a half times their pre-pandemic level.4

So which one is right?

We are not going to try to settle that — and we want to be clear that this is not a dodge. For the question an executive actually needs answered, it does not matter which one is right.

On that question — whether the organization is positioned to compete and win — all of these instruments are silent — not because they are badly built, but because none of them was designed to answer it. Whether engagement is rising or falling, whether satisfaction is at a record high or workers are quietly burning out, not one of these numbers tells a chief executive whether their enterprise can decide quickly, ship ahead of a competitor, absorb a technology shift, or serve a customer better than the alternative.

The instruments disagree with each other. Either way, they are not doing the job.

The field’s own explanation for this failure is well established, and it is largely a story about execution. Organizations measure and then fail to act. Managers are not equipped to run the follow-through. Findings are presented to the leadership team, an initiative list is generated, and the difficult work of translating insight into changed behavior is left undone.

There is real evidence for this account. Gallup’s research finds that only 8% of employees strongly agree that their organization takes action on survey results.5 It is worth pausing on the provenance of that number: it is reported by the firm whose engagement instrument is the most widely deployed in the world, in its own client-facing materials. Whatever else it may be, it is not a competitor’s talking point. On this evidence, roughly nine in ten employees do not believe the exercise leads anywhere — and they have years of direct experience on which to base the judgment.

The execution diagnosis is therefore not wrong. But it is incomplete, and its incompleteness is demonstrable. That diagnosis has been available, and widely circulated, for well over a decade. The prescription following from it — act on the data, close the loop, train the managers, hold leaders accountable — has been the consensus recommendation of the field throughout that period. The recommendation has been made, adopted at least nominally by a great many organizations, and the picture has not clarified.

When a remedy is prescribed for a long time and the patient does not improve, one possibility is that the remedy is being administered badly. Another is that the diagnosis is wrong. The second possibility deserves a hearing.

Here it is necessary to proceed carefully, because the case against engagement surveys is frequently overstated, and the overstatement is precisely what allows practitioners to dismiss it.

Engagement is a real construct. It is not a marketing invention. The meta-analytic literature linking employee engagement to business-unit outcomes — profitability, productivity, retention, safety, customer loyalty — is among the more heavily replicated bodies of evidence in industrial and organizational psychology, and it broadly holds.6 Organizations with more engaged workforces do, on average, outperform organizations with less engaged ones. An argument that begins by denying this should be dismissed by anyone who knows the literature.

The problem is not that engagement measures nothing. The problem is what it measures.

Engagement is an affective and attitudinal construct. Every serious definition of it — from Kahn’s original formulation in 1990 to the Utrecht Work Engagement Scale that remains the academic standard today — describes a state of mind: how energized a person feels, how absorbed, how willing to bring their whole self to the work.7 It captures how employees feel about their work and their workplace: whether they find meaning in the job, whether they feel recognized, whether someone at work encourages their development. These are real and consequential things. They are also, categorically, statements about employee experience.

They are not statements about organizational capability.

Whether an organization can reach a high-quality decision in a compressed timeframe; whether it can bring a product to market faster than its competitor; whether its systems and tooling genuinely enable the work rather than obstructing it; whether it can absorb a technology shift; whether the customer is better served here than by the alternative — these are questions about what the enterprise can do. They correlate with how people feel, sometimes strongly. They are not the same thing, and a well-constructed measure of the first is not, by that fact, a measure of the second.

Predicting is not measuring.

It is worth being exact here, because the claim is easy to overstate. The research does say that engagement predicts performance, and we have granted it. What the research has never said is that an engagement score measures what an organization can do. A patient’s answer to “how’s your energy?” genuinely does predict how well he will recover. It is not a reading of his immune system, and a doctor who skipped the blood work because the patient said he felt fine would be making a category error.

The field has largely glossed over this distinction. Engagement was treated as a proxy for organizational health, and over time the proxy came to be treated as the thing itself. The executive who asks whether the organization is positioned to compete, and is handed an engagement score, has not been given a bad answer to their question. They have been given a good answer to a different question.

There is a revealing detail buried in the academic record. Researchers have argued for nearly twenty years about whether engagement is even a distinct idea. One camp holds that it is largely a repackaging of things the field already had names for — job satisfaction, commitment, involvement.

The quantitative version of that case is hard to shrug off. Compare a conventional engagement survey against those same older attitudes, and the two track each other at about 0.77 — on a scale where 1.0 would mean they are measuring the same thing. Much of what the engagement survey captures, in other words, the older measures already did.8,9 The other camp says engagement is distinct, because it carries a charge of energy that satisfaction lacks.10 The dispute is unresolved.

But notice what both camps are fighting about. One says engagement is the same as job satisfaction, commitment, and involvement. The other says it is different from job satisfaction, commitment, and involvement. Both are arguing about which feelings employees have, and whether our instruments are sharp enough to tell those feelings apart. Neither is arguing about whether the company can execute. Twenty years of serious debate, and the question a chief executive actually needs answered has not once been the subject of it.

The field’s own theory points the same way. The dominant explanatory model in engagement research holds that engagement is caused by the conditions of the job — autonomy, useful feedback, decent tools, real support, room to grow — and that engagement then leads to performance.11 Engagement sits in the middle. On the field’s own account, it is a symptom of the organization’s condition, not a reading of it.

The recent divergence between engagement scores and economic performance sharpens the point without settling it. Global engagement has declined since 2022 while the global economy has continued to expand. This does not demonstrate that engagement is unimportant: the relationship is probabilistic, it operates at the level of the business unit rather than the world economy, and the comparison is confounded by any number of macroeconomic variables. But it does put pressure on the tacit assumption underwriting the engagement industry — that the engagement score serves as a reliable shorthand for the organization’s condition, such that moving it is equivalent to improving the enterprise. If that assumption held, the divergence would be harder to account for.

Even a valid instrument must be read by someone, and the people who read organizational data are not positioned neutrally with respect to it. They sit at the top of a hierarchy that filters information on its way up.

Gallup’s data captures this with unusual clarity, because it reports engagement separately for managers and for individual contributors. The two have diverged persistently. In 2022, 31% of managers were engaged, against 20% of individual contributors. Gallup terms this the manager “engagement premium,” and it has been a stable feature of the data for years.12 The premium narrowed sharply in the most recent period, but for an unwelcome reason. Manager engagement collapsed to 22% by 2025 while individual-contributor engagement held roughly flat at 19%. The gap closed not because the front line improved, but because management fell to meet it.

That development is significant in its own right, and we will return to it.

Signaling theory gives this problem a precise language.13 Employees are the signalers; senior leaders are the receivers, the underlying condition is organizational capability, and the central question is whether the organization has a credible mechanism that allows what employees know to cross the hierarchy intact. Without such a mechanism, leaders are not simply operating with incomplete data. They are operating in what signaling theory calls a pooling equilibrium, in which genuinely strong, merely adequate, and quietly failing parts of the enterprise can look remarkably similar from the top.14

The consequence is that a single organization-wide score is not merely uninformative. It is actively concealing. It averages the view from the top together with the view from the front line, producing a number that describes neither, and it does so in a direction that will systematically flatter the organization’s capability. The most decision-relevant fact available in the data — the size of the gap between how the organization looks to those planning the work and to those performing it — is precisely the fact that aggregation destroys.

The second flaw with the engagement model is subtler, and comes from Marcus Buckingham, who has spent his career building these instruments and knows better than most what they can and cannot bear.

In an ADP Research Institute technical report, Buckingham and his colleagues make a claim that ought to be far better known than it is. What predicts the outcomes organizations actually care about — lower turnover, fewer accidents, higher productivity, customer satisfaction — is not the average response. It is the top box: the respondent who can say, without qualification, strongly agree. And when organizations combine the top two boxes into a “percentage favorable,” they are throwing the signal away.15

Sit with the implication. A four is not a slightly weaker five. It is a hedge. It is the answer of someone who finds the thing acceptable, unobjectionable, adequate. And adequacy predicts nothing. Below the top box, as the report puts it, productive behavior becomes unpredictable — people are present, but what they will actually do on any given day is anyone’s guess.

Now apply this to a mean score.

Two organizations both report 3.8 on a dimension. In the first, the distribution is wide: a substantial cluster of fives, a tail of ones and twos. In the second, virtually everyone has parked on four. The averages are identical. The first organization has genuine, unqualified conviction somewhere inside it — something real to build on, and a visible problem to fix. The second has nothing at all. No conviction, no advantage, no foundation. Just a workforce that finds things tolerable.

The mean cannot tell them apart. The distribution can.

If Buckingham is right — and it is his instrument, and his data — then the average is not a summary of the organization. It is the systematic erasure of the only part of the response that carried information.

This is not a fringe position. The Conference Board, reporting its own record-high satisfaction figure, cautions that rising averages alone are not a signal of success, and its lead recommendation to CHROs is to analyze at the employee-segment level rather than rely on overall averages.16 Its data shows why: that headline 68.9% sits atop an average of just 59% satisfaction across the individual elements of work, and satisfaction by household income ranges from 45.3% at the bottom to 76% at the top. Its own researchers suggest the record high may reflect gratitude for employment amid uncertainty rather than any enthusiasm for the work itself.17 The mean, then, is a lie twice told. It hides who is answering, and it hides how strongly they answered. And a strategic plan is, in the final analysis, a bet on organizational capability — a presumption that the enterprise can execute at some rate, absorb some amount of change, decide at some speed. If the assessment underwriting that presumption has been averaged across a hierarchy with a documented optimistic bias, and averaged again across a distribution in which only the extreme carried any predictive weight, then the plan is not merely built on an uncertain figure.

It is built on a figure that is wrong in a predictable direction, and that has been engineered — with the best of intentions — to look reassuring.

For much of the past fifteen years, the cost of misjudging organizational capability was tolerable, because other levers were available to compensate. That is changing on two fronts at once.

The first is technological. The prevailing corporate instinct is to treat productivity as a tooling problem and solve it through procurement. Yet Gallup’s Q1 2026 US workforce survey found that the strongest predictor of whether employees actually use AI — setting aside technical integration — is whether their direct manager actively champions it.18 The Conference Board’s 2026 data points the same way from a different angle: workers confident about AI’s effect on their careers report substantially higher engagement, belonging, and intent to stay, while a meaningful minority report that AI has reduced their satisfaction — which its researchers read as a warning about deploying the technology without adequate training and support.19 The finding locates the real constraint on AI returns not in the model, the license, or the integration layer, but in the organization: in management capacity, in trust, in whether the people who must change how they work believe the change is real and supported. Whether a technology helps people excel at their jobs is an organizational question. It will be answered badly by firms that cannot see their own condition clearly.

This is also where the collapse in manager engagement becomes more than a curiosity. The population on which AI adoption most depends is the population whose engagement has fallen furthest and fastest.

The second front is financial, and for private equity it is acute. Bain’s 2026 Global Private Equity Report lays out the arithmetic under the heading “12 is the new 5.” In a typical 2015 buyout, roughly half the purchase price was borrowed at 6–7%, asset prices were climbing, and multiple expansion did much of the work — a deal required approximately 5% annual EBITDA growth to deliver a 2.5x multiple on invested capital over five years. Today, borrowing costs sit in the 8–9% range, leverage has fallen to 30–40%, and purchase multiples remain at record levels without expanding. The same 2.5x return now requires something closer to 10–12% annual EBITDA growth.20 Financial engineering and multiple expansion — which for a decade could carry a transaction even where operational improvement was marginal — are exhausted as primary levers. The return must now come out of the operating business, which is to say it must come out of the organization’s actual ability to execute. The premium on assessing that ability accurately, both at diligence and through the hold, has roughly doubled alongside the required growth rate.

An operating partner underwriting 10–12% annual EBITDA growth is making a very large bet on organizational capability. It is worth asking what instrument is being used to price it.

If the foregoing is right, the remedy is not a better-executed engagement survey. It is a different instrument, aimed at a different construct, and read differently. Five requirements follow.

These requirements are not hypothetical, and we should be transparent that we hold a stake in them. We have built an instrument intended to meet them.

The Competitive Index™ draws on work used in the field for over a decade at GE and Rockefeller Capital Management. It comprises thirty-two questions across eight dimensions — and reports results segmented by business, function, level, and geography.

How can thirty-two perception items measure capability rather than feeling? By changing what respondents are asked to report. An engagement item asks how a person feels — whether they find meaning in the work, whether they feel recognized. A capability item asks what they observe — whether decisions that need three functions in the room actually get them, whether a competitor consistently reaches the customer first, whether the tools in daily use speed the work or slow it. Employees are not asked to rate their morale; they are asked to report conditions they are positioned to see, like whether decisions are made quickly enough or they are encouraged to take risks. At GE, Jack and the team used the Competitive Index to support a transformation from a traditional to a digital organization. The CI was able to pinpoint groups within the organization that were struggling to transform and those with successes that could be shared. The ethos of transparency and prioritization that must go alongside the process made sure the results were shared company-wide and that the areas of improvement that would have the greatest impact were targeted first. At Rockefeller, the diagnostic was more focused on whether we were building the culture and capabilities that we aspired to in creating a unique and elevated client experience.

In addition to our practical experience in the field, a recent reliability study of 241 respondents established internal-consistency coefficients (Cronbach’s alpha) between 0.74 and 0.88 across the eight dimensions.22 Internal consistency of this kind is a necessary foundation for validity, not a substitute for it; confirming that the instrument measures what it claims — through expert review of the item-to-dimension mapping, and in time through retest stability — is the next step. Two design decisions follow directly from the argument above.

We do not report a mean without its distribution, and we flag the dimensions where the fives cluster. The reasoning extends Buckingham’s, and we think it applies with even greater force to a measure of competitiveness than to a measure of engagement — because competition is relative. An organization that is adequate at decision-making does not beat a rival that is exceptional at it.

The Competitive Index is better understood not as another employee survey but as an answer to the question this essay began with: whether the organization is positioned to compete and win. It converts what the people closest to the work already know into a credible read on capability — the kind of read a leader can use to underwrite strategy, price a deal, target investment, and hold the organization accountable for change that can be measured. In a market where growth must increasingly be earned through execution rather than leverage or multiple expansion, seeing the organization clearly — its real capacity to decide, adapt, and outperform a rival — is itself the advantage.

The economic case for a capability diagnostic begins before any intervention is designed. Better visibility changes the quality of the bet. It allows a CEO, board, investor, or operating partner to distinguish a strategy that is demanding but executable from one whose assumptions exceed the organization’s present capacity. That distinction has value even before a single process is redesigned.

The value also compounds. A first diagnostic surfaces the signal. Leadership action becomes a countersignal: employees observe whether the response is real, precise, and connected to what they reported.21 A follow-up CI then measures both the change in capability and the credibility of the leadership response. Where leaders act visibly, subsequent signals become more specific and useful. Where nothing changes, signal quality deteriorates because employees learn that candor carries no consequence.

What we are looking for is not the absence of complaint. It is the presence of conviction — and it has to be a five, or it is nothing.

We report the seniority gap rather than averaging it away. Our reliability study surfaced a finding we had not gone looking for: senior leaders rated their organizations 0.3 to 0.5 points higher than the individual contributors doing the work, consistently, across dimensions. This is consistent with Jack’s experience at both GE and Rockefeller. We take this as corroboration of the vantage-point problem rather than as a proprietary discovery — Gallup has documented a parallel gap for years, and The Conference Board is now urging its own members to stop reading averages for much the same reason. But it is a useful reminder that the phenomenon appears wherever one looks for it, and that instruments which average it away are concealing the most important thing they know.