Insights

The question, stated fairly

Every budget cycle, some version of this question lands on a CHRO’s desk: we spend real money developing managers — what do we get for it?

It is a fair question, and it deserves a straight answer rather than a defensive one. Leadership development is an unusual line item. The output is a behavior, not an asset. The payback arrives late, if it arrives at all. And the two parties best positioned to evaluate the program — the vendor who sold it and the HR team who bought it — both have an interest in concluding that it worked.

Most of the evidence offered in response makes the problem worse. Vendor case studies, single-client testimonials, and research published by the firm selling the service are not evidence a finance partner should accept, and a CHRO who brings them into a budget meeting should expect to lose the argument. The honest answer starts by conceding that much of what circulates as proof is marketing.

There is better evidence. It has been in the peer-reviewed literature since 2017, and it does something more useful than settle the yes-or-no question.

What 335 studies show

Lacerenza, Reyes, Marlow, Joseph, and Salas published a meta-analysis in the Journal of Applied Psychology that pooled 335 independent samples covering 26,573 participants (Lacerenza et al., 2017). The source studies reach back to 1951, include both published and unpublished work, and were restricted to programs delivered to working employees rather than MBA or undergraduate students. That last screen matters: it means the sample describes practicing managers, not simulations of them.

The authors measured effects against four criteria — reactions, learning, transfer to the job, and organizational results. Corrected for unreliability in the criterion, leadership training improved reactions (δ = .63), learning (δ = .73), transfer (δ = .82), and results (δ = .72). The overall effect was δ = .76, 95% CI [.64, .89] (Lacerenza et al., 2017).

Two features of that finding matter more than the headline. First, the authors tested for publication bias three ways, including a trim-and-fill analysis that imputed zero missing studies, and found no evidence of it; published and unpublished studies did not differ significantly. Second — and this is the part vendors leave out — the four criterion effects were not statistically different from one another. The paper’s own conclusion is that leadership training programs tend to be equally effective across criteria. Do not let anyone sell you the .82 transfer figure as evidence that behavior change is where the returns concentrate. The data do not support that ranking.

Translating the number without inflating it

An effect size is not a percentage and not a dollar figure. Delta expresses a difference measured in standard deviations. Under Cohen’s widely used convention, .20 is small, .50 is medium, and .80 is large, which places these results in the medium-to-large band (Cohen, 1988).

A more concrete reading, using the standard percentile conversion and assuming roughly normal outcome distributions: an effect of .73 puts the average trained manager at about the 77th percentile of the untrained comparison group. A manager who sat in the middle of the pack before training performs better than roughly three-quarters of their untrained peers afterward. That is a real and visible change. It is not a miracle, and it is not an ROI multiple — the study measures effect, not payback period, and anyone converting these deltas directly into a dollar return is doing arithmetic the paper does not license.

One honest caveat belongs in the same breath. The paper reports a moderator for study design, and the most rigorous designs — those combining independent groups with repeated measures — produced a smaller effect, δ = .48, 95% CI [.25, .71] (Lacerenza et al., 2017). The effect stays clearly positive under the tightest methodology. It also shrinks. A CFO who asks whether the headline is flattered by weaker study designs is asking a good question, and the answer is: partly yes, and the finding survives anyway.

The design conditions that separate the programs that work

The reason to bring this paper into a budget conversation is not the average. An average across 335 programs describes a market, not a purchase. What makes the study useful is that it tests which design choices move the effect — which turns a defense of the category into a specification for buying inside it.

The moderator analyses support six conditions (Lacerenza et al., 2017):

• Needs analysis. Programs built on a documented analysis of the organization’s actual gaps outperform generic curricula.

• Feedback. Programs that give participants structured feedback outperform programs that do not.

• Multiple delivery methods, especially practice. Information alone underperforms information combined with demonstration and practice.

• Spaced sessions. Content distributed across multiple sessions outperforms the same content massed into a single event.

• On-site location. Training delivered in or near the work environment outperforms off-site delivery.

• Face-to-face delivery that is not self-administered. Programs facilitated by an instructor outperform self-administered ones.

A note on precision, because this paper is widely misquoted. The direction of each of these six findings is well supported. The specific contrast figures that circulate on vendor websites — “d = 0.78 with needs assessment versus 0.41 without,” and similar pairings — do not correspond to the values in the published tables, and we will not repeat them. Several of the paper’s own moderator estimates carry very wide confidence intervals, which is exactly why selective quotation of them is so easy and so misleading. Use the direction of the findings. Do not quote the contrast deltas.

Read that list again as a procurement checklist

Now hold the market against those six conditions.

The two-day leadership retreat at a conference center fails spacing, fails on-site location, and usually fails needs analysis. The self-paced e-learning library fails facilitation, fails feedback, and fails practice. The keynote speaker fails all six. The off-the-shelf curriculum delivered identically to every client fails needs analysis by design — that is what “off-the-shelf” means.

This is the finding that should change the budget conversation. The question is not whether manager development works. It does, across every criterion the research measures. The question is whether the specific program in front of you was built the way the effective ones were built. Most programs on the market fail three or more of the six conditions, which means the typical buyer is purchasing something meaningfully weaker than the average the literature reports — and then citing that average to justify it.

That reframing gives a CFO something more useful than a verdict on the category. It gives a basis for cutting the spend that will not work and defending the spend that will. It also gives an L&D function a defensible standard to hold vendors to, which is a better position than arguing about whether development matters in the abstract.

What to ask a vendor — including us

Six questions follow directly from the evidence. Ask them of every provider on your list, and ask them of gothamCulture:

1. What needs analysis will you conduct before designing this, and what document will it produce?

2. What feedback will participants receive, from whom, and how many times?

3. What proportion of contact time is practice rather than presentation?

4. Across how many weeks is this delivered, and in how many separate sessions?

5. Where does it happen, and how close is that to where the work actually happens?

6. Who facilitates, and what is the ratio of facilitated time to self-administered time?

A provider who cannot answer these is not necessarily selling a bad program. They are selling a program whose design they cannot defend against the strongest evidence in their own field, which is a different problem and not a smaller one.

Ask us the same six questions. Where the honest answer is weak for what your organization actually needs, we would rather say so before you buy than explain it afterward.

Sources

Lacerenza, C. N., Reyes, D. L., Marlow, S. L., Joseph, D. L., & Salas, E. (2017). Leadership training design, delivery, and implementation: A meta-analysis. Journal of Applied Psychology, 102(12), 1686–1718. https://doi.org/10.1037/apl0000241

Full text (PDF, Rice University): https://doerr.rice.edu/sites/g/files/bxs5941/files/2026-01/Leadership%20Training%20Design,%20Delivery,%20and%20Implementation.pdf

PubMed record (PMID 28749153): https://pubmed.ncbi.nlm.nih.gov/28749153/

Rice University MET Lab summary of the findings: https://metlab.rice.edu/2017/12/06/meta-analysis-identifies-the-conditions-under-which-leadership-training-is-most-effective/

Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Lawrence Erlbaum Associates. (Source of the .20 / .50 / .80 small–medium–large convention referenced above.)

Everyone is quoting the same number. The rigorous read of it points somewhere entirely different from the restructure it is being used to justify.

The story circulating in HR forums this year is that the American manager’s job has doubled in size. Layers are gone, teams have swollen, and every supervisor is now buried under direct reports. It is a compelling story, and it is being used to justify org redesigns. The data behind it says something narrower and considerably more useful.

Gallup’s January 2026 span-of-control analysis is a rare case where the rigorous read is more actionable than the alarmist one. The average did move. The typical manager did not. The distance between those two sentences is the distance between a restructure and a triage.

The Average Moved. The Median Did Not.

Gallup’s U.S. panel data show the mean number of direct reports rising from 8.2 in 2013 to 12.1 in 2025 — an increase of roughly 47% over twelve years (Gallup, 2026). Gallup describes this as a “nearly 50%” increase, and the most recent single-year move is real: 10.9 in 2024 to 12.1 in 2025. That number is the one being quoted.

The median is the number nobody quotes. Across the same twelve years it has oscillated between five and six, landing at six in 2025 — precisely where it sat in 2019 (Gallup, 2026). The typical American manager’s team has not grown at all. The distribution confirms it: 37% of managers and leaders oversee fewer than five people, and roughly two-thirds (66%) manage fewer than 10. Only 13% oversee 25 or more.

These estimates carry real precision. The trend series draws on manager samples ranging from n=2,402 in 2013 to n=8,795 in 2025, with 95% margins of sampling error on the mean estimates between ±0.48 and ±0.78 (Gallup, 2026). The mean’s movement is well outside that band. So is the median’s refusal to move.

What is actually happening is a distributional shift, not a general one. Gallup attributes the past year’s jump largely to a two-percentage-point increase in teams of 25 or more employees — organizations consolidating middle management while loading the survivors (Gallup, 2026). A minority of very large teams is dragging the average upward while most managers’ spans sit untouched.

This Reframes the Intervention

If the mean and the median had both moved, the problem would be structural and the response would be structural: a genuine redesign of reporting lines across the enterprise. They did not both move. One tail of the distribution moved.

That makes this a triage problem. You are looking for a targetable subset — a defined population of managers whose spans expanded during consolidation — not a company-wide restructure. Triage is dramatically cheaper, faster, and more politically survivable than redesign. It is also more likely to work, because it treats the managers who are actually under strain instead of redistributing headcount among managers who were fine.

The harder question is which managers in that subset are genuinely at risk. Span alone will not tell you. Gallup’s data are unambiguous that no single ratio holds across contexts, which is why the search for an ideal number keeps failing (Gallup, 2026).

The Diagnostic That Actually Discriminates

The more useful variable is one most span-of-control conversations never raise: how much of a manager’s week is not management at all.

In Gallup’s U.S. study, 97% of managers report carrying individual contributor responsibilities alongside leading others, at a median of 40% of their time (Gallup, 2026). This is not an edge case or a startup phenomenon. It is the near-universal condition of the American manager: the store supervisor who also works the register, the research director still shipping code, the practice lead still billing hours. Gallup calls them player-coaches. Almost every manager is one.

That figure reframes the flattening debate. When organizations widen spans without removing individual contributor work, they are not simply redrawing a chart — they are asking someone who already spends two days a week on non-managerial output to absorb more people with the remaining three.

There Is a Threshold, and It Is 40%

Here is the finding that should change how a CHRO screens. Managers who spend 40% or less of their time on individual contributor work hold engagement at 37% across every team-size band Gallup measured — fewer than five reports, five to 10, 10 to 25, and 25 or more. The line is flat. Span does not degrade them (Gallup, 2026).

Managers above the 40% threshold show a different curve entirely: 36% engagement with fewer than five reports, 35% at five to 10, 34% at 10 to 25, and 32% at 25 or more (Gallup, 2026). The penalty is modest at small spans and compounds as the team grows. An extreme player-coach can stay engaged running a small team. The same person running 25 people cannot.

Span of control, in other words, is not an independent variable. It is a multiplier on a condition set somewhere else. Adding reports to a manager below the threshold costs you nothing measurable in that manager’s engagement. Adding reports to a manager above it costs you steadily more with each one.

Manager talent moves the same lever. Among managers carrying more than 40% individual contributor load, engagement runs 28% for low-talent managers, 35% for medium and 42% for high — against 27%, 37% and 46% respectively for those at or below the threshold (Gallup, 2026). Talent raises the whole curve. It does not eliminate the threshold effect.

The Two-Variable Screen to Run This Quarter

Chasing a target ratio is a waste of a quarter. Screening on two variables is not, and the screen is small enough to run without a consulting engagement.

Pull two data points for every manager in the organization: the number of direct reports, and the estimated share of weekly time spent on individual contributor work. The second is a single survey question. Then sort into four cells.

• Low IC load, small span. No action. This is the largest group and it is fine.

• Low IC load, large span. Monitor, do not intervene. Gallup’s data show engagement holding flat here. These managers are absorbing the flattening without damage.

• High IC load, small span. Watchlist. Currently stable, but this cell converts to the danger cell the moment anyone adds headcount to it. Flag it before the next reorg, not after.

• High IC load, large span. This is the intervention population. It is where engagement erosion concentrates and where a targeted fix returns the most.

For that fourth cell, the fix is usually not fewer reports. It is less individual contributor work — reassigning the billable hours, the queue coverage, the hands-on delivery that quietly accumulated when the layer above disappeared. That is a workload decision a division head can make in a week. Redrawing reporting lines is a nine-month project with a change-management budget attached.

What the Data Do Not Settle

Two caveats belong in any board conversation that uses these numbers.

First, the time series and the threshold analysis are not drawn from the same window. The 2013-to-2025 mean trend runs through 2025, while the manager engagement regressions rest on web surveys fielded from 2022 through 2024 with 16,442 managers and leaders, at a ±1.1 percentage-point margin of error (Gallup, 2026). The 2025 spike in large teams postdates the engagement modeling. The threshold is well evidenced; whether it holds at 2025’s higher spans is not yet tested.

Second, these are self-reported estimates of both span and time allocation, and Gallup’s employee-side and manager-side instruments define team size differently — one asks how many people are on the respondent’s team, the other how many report directly to the manager (Gallup, 2026). Neither undermines the finding. Both argue for validating the IC-load number against actual time or utilization data before you act on any individual manager’s cell assignment.

The Reframe

The question “how many direct reports should a manager have?” has no answer, which is why it keeps generating conference panels and no decisions. The question “which of our managers are carrying more than 40% individual contributor load at a widened span?” has a specific, countable answer, and it names the people to help.

The mean rose 47%. The median did not budge. Everything useful lives in that gap.

Sources

Harter, J. (2026, January 13). Span of Control: What’s the Optimal Team Size for Managers? Gallup. https://www.gallup.com/workplace/700718/span-control-optimal-team-size-managers.aspx

Gallup. (2026). Span of Control Meta-Analysis Executive Summary. https://www.gallup.com/file/workplace/701021/Span%20of%20Control%20Meta%20Analysis%20Exec%20Summary.pdf

Gallup. (2020, March 30). The Ideal Team Size Depends on the Manager. https://www.gallup.com/workplace/286997/ideal-team-size-depends-manager.aspx

U.S. Bureau of Labor Statistics. Employment by Major Occupational Group. https://www.bls.gov/emp/tables/emp-by-major-occupational-group.htm

Methodology and Sample Sizes

Methodology note. Figures are drawn from Gallup’s January 13, 2026 span-of-control analysis. The global meta-analysis covers 92,252 teams (897,971 employees) across 312 research studies, 104 organizations, 26 industries and 46 countries. U.S. manager trend data come from Gallup Panel web surveys: 2013 (n=2,402), 2019 (n=2,933), 2022 (n=7,199), 2023 (n=3,971), 2024 (n=4,355) and 2025 (n=8,795), with 95% margins of sampling error on the mean estimates ranging from ±0.48 to ±0.78. Manager engagement and individual-contributor-load analyses draw on 16,442 U.S. managers and leaders surveyed 2022–2024 (margin of error ±1.1 percentage points). Employee-level analyses draw on 44,025 U.S. adults surveyed 2022–2024 (margin of error ±0.7 percentage points). The 47% mean increase is calculated from Gallup’s reported series values of 8.2 (2013) and 12.1 (2025); Gallup’s own narrative characterizes the same change as “nearly 50%.”

The series argued that psychological safety is organizational infrastructure. The payoff showed up in the one place nobody was looking: whether your people actually use the AI you bought.

For several months this series made a single argument in several forms. Psychological safety is not a culture perk, a morale program, or a soft-skills initiative. It is infrastructure. Teams without it underperform in ways that never surface cleanly on a dashboard — the question nobody asks, the error nobody reports, the idea that dies in someone’s head on the walk to the meeting.

We expected that argument to pay off in the familiar places: quality, retention, innovation velocity. It paid off somewhere else entirely.

The most urgent capability question in front of most executive teams right now is AI adoption — not whether to buy, but why the tools sit unused after they are bought. The 2026 research says the operative variable in that question is the same one we have been writing about all along. That is the payoff nobody expected, and it reframes what the next phase of this work has to be about.

The Finding: It Is Not the Strategy. It Is the Supervisor.

Gallup’s 2026 research set out to identify what actually drives AI use at work. Two variables separated from the rest. The first is unsurprising: whether AI is integrated with the systems people already use. The second deserves a full stop. In Gallup’s words, the strongest predictor of whether employees use AI, aside from technical integration, “is whether an employee’s direct manager actively champions it” (Gallup, “Global Employee Engagement Continues Decline”, 2026).

Not whether the CEO announced an AI strategy. Not whether there is a center of excellence, a governance council, or an internal prompt library. Not the license count. The direct manager — the person who runs the one-on-one, assigns the work, and decides what a good week looks like.

The effect is not marginal. Employees who strongly agree their manager champions AI are roughly eight times more likely than everyone else to say AI has transformed how work gets done in their organization — 33% versus 4% (Gallup, Q1 2026 U.S. Workforce Survey, n = 23,717).

The supply of that behavior is thin. In our own work with executive teams, manager championing of AI is almost never a defined expectation, almost never trained for, and almost never measured. The single highest-leverage variable in the adoption equation is the one almost no organization has deliberately built.

The Mechanism: Safety Is What Managers Actually Supply

Championing is the visible behavior. It is not the mechanism. The mechanism is what championing produces downstream, and here the research names it directly.

Microsoft’s 2026 Work Trend Index reports that when managers created psychological safety around experimentation, employees reported up to 20 points higher AI readiness and perceived value, and were 1.4 times more likely to be high-frequency users of agentic AI. When managers visibly modeled AI use themselves, the report finds a 17-point lift in perceived AI value, a 22-point lift in critical thinking about AI use, and a 30-point lift in trust in agentic tools (Microsoft, 2026 Work Trend Index).

Read that source with your eyes open, because we are not going to pretend it is something it is not. The Work Trend Index is published by Microsoft, a company whose commercial position depends directly on enterprise AI adoption. It is vendor research, not independent research. A finding that says “your people would adopt more if your managers behaved differently” is precisely the finding a vendor benefits from producing, because it relocates the barrier from the product to the customer. We cite it here for two reasons: the directional pattern matches an independent survey with a far larger U.S. sample and nothing to sell, and the mechanism it describes has decades of prior organizational literature behind it. Treat the specific magnitudes as vendor-reported. Treat the direction as corroborated.

The mechanism itself is not novel, which is exactly why it is credible. Using an unfamiliar tool in front of your manager is a competence risk. The first attempts are visibly worse than doing the task the old way, the output is wrong in embarrassing ways, and the time cost is real while the payoff is speculative. Employees run that calculation before they experiment, and in most organizations it returns a clear answer: not worth it.

A manager who champions AI changes the inputs to that calculation. They signal that a bad first draft from a new tool is a normal cost of learning rather than evidence of poor judgment. They make their own fumbling visible. They absorb the short-term productivity dip instead of charging it to the individual. That is psychological safety operating in a specific, technical domain — and it produces measurable adoption.

A Management Problem Wearing a Technology Costume

This reframes AI strategy in a way most organizations have not absorbed. The dominant model treats adoption as a funnel problem: select the platform, provision the licenses, deliver the training, measure usage, remediate the laggards. Every step in that sequence is an IT or L&D activity. None of them touch the variable the data identifies as decisive.

If the strongest non-technical predictor of adoption is a manager behavior, then the AI adoption gap is a management capability gap. It is wearing a technology costume because the invoice arrives from a software vendor and the initiative reports to a technology function. But the failure mode is not technical. Organizations with identical platforms, identical licenses, and identical training decks produce radically different adoption curves, and the variance sits at the team level, under specific managers.

This also explains a pattern that has frustrated executives for two years: enterprise-wide AI announcements that generate a usage spike and then a collapse. A launch is a leadership act. Sustained adoption is a management act. Leadership can authorize the tool. Only the direct manager can make it safe to be bad at it in public for a few weeks.

Where the Next Dollar of Enablement Budget Goes

Most enablement budgets are currently allocated on the assumption that the constraint is employee skill. The evidence says the constraint is manager behavior. That argues for three specific shifts.

Shift the audience. Move a meaningful share of AI enablement spend from all-employee training to frontline and middle managers, and change what that training covers. Managers do not primarily need advanced prompting technique. They need to know how to run a team through a competence dip — how to set expectations for early-stage failure, how to talk about a bad AI output without making it a performance conversation, and how to model their own use out loud.

Shift the measurement. Stop reporting adoption as a single organizational percentage. Report it by manager. Team-level variance is where the signal lives, and an aggregate number is engineered to hide it. When adoption is measured at the team level, the managers who have built experimentation safety become visible and replicable, and the ones who have quietly signaled that AI is a distraction become addressable.

Shift the accountability. If manager championing is the mechanism, it belongs in the manager’s role expectations and performance conversations — not as a usage quota, which produces compliance theater, but as an expectation that the manager has established a team norm for experimentation. The distinction matters. A quota asks managers to generate activity. A norm asks them to generate safety.

None of this requires new technology spend. It requires redirecting a portion of an existing budget toward the population that determines whether the rest of it returns anything.

The Uncomfortable Next Question

There is a problem with a strategy that runs entirely through managers, and it is worth naming before anyone builds a plan on it.

Gallup’s engagement data shows manager engagement falling from 27% to 22% between 2024 and 2025, down nine points since 2022 — the steepest decline of any worker category, and sharpest among managers under 35 and women managers. The engagement premium managers once held over the people they supervise has compressed from eleven points to three (Gallup, State of the Global Workplace 2026). The cohort we have just identified as the mechanism for the most important capability shift in a decade is also the cohort that has deteriorated fastest.

You cannot ask a depleted manager to create psychological safety for someone else. Safety-making is discretionary effort, and discretionary effort is exactly what disengagement removes first. Loading AI enablement onto managers without addressing their own condition is not a strategy. It is a transfer of an unfunded obligation to the least resourced layer of the organization.

Which is where this series goes next. If the manager is the mechanism, then the state of the manager is the constraint — and that state is worse than most executive teams believe. Next: “Your Managers Are Now the Least Engaged People in Your Company.”

Sources

• Gallup. “AI’s Effect on Workplace Culture” and related Q1 2026 U.S. workforce findings. Q1 2026 U.S. Workforce Survey, n = 23,717 employed U.S. adults, fielded February 4–19, 2026. https://www.gallup.com/workplace/712976/ai-effect-workplace-culture.aspx — Sample size (23,717 employed U.S. adults aged 18 and older), field dates (February 4–19, 2026), and a margin of error of ±0.9 percentage points at the 95% confidence level are all stated on the source page.

• Gallup. “Employee Engagement Remains Flat as AI Adoption Accelerates.” https://www.gallup.com/workplace/712433/employee-engagement-remains-flat-adoption-accelerates.aspx

• Gallup. “Global Employee Engagement Continues Decline.” Jim Harter and Ryan Pendell, 2026. Source of the “strongest predictor” statement. https://www.gallup.com/workplace/708071/global-employee-engagement-continues-decline.aspx

• Gallup. State of the Global Workplace 2026. Manager engagement decline (27% to 22%, 2024–2025; nine points since 2022) and manager–individual contributor engagement gap compression. https://www.gallup.com/workplace/349484/state-of-the-global-workplace.aspx

• Microsoft. 2026 Work Trend Index Annual Report: “Agents, Human Agency, and the Opportunity for Every Organization.” Survey of 20,000 AI users across 10 markets, plus Microsoft 365 telemetry and a separate 1,800-person global manager study. https://www.microsoft.com/en-us/worklab/work-trend-index/agents-human-agency-and-the-opportunity-for-every-organization — Vendor-published research. Microsoft has a direct commercial interest in enterprise AI adoption findings; this is not independent research and is identified as such in the body. Microsoft’s published methodology states that the survey was conducted by an independent research firm, Edelman Data x Intelligence, among 20,000 full-time employed or self-employed knowledge workers who use AI at work, 2,000 in each of 10 markets (Australia, Brazil, France, Germany, India, Italy, Japan, the Netherlands, the United Kingdom and the United States), fielded February 18 to April 7, 2026. Note that commissioned independent fielding does not make the report independent research; the vendor-interest disclosure stands regardless.

You have cited it. You have sat through a vendor deck that cited it at you. Somewhere in your organization, a slide is being built right now that will cite it again: managers account for 70% of the variance in employee engagement. It has become the load-bearing wall of the entire manager-development category, and it has been repeated so often that almost no one goes back to read it.

We went back and read it. The number is real, the research behind it is serious, and the way our industry cites it is sloppy in a way that changes what you would do about it. This piece is our attempt to cite it precisely — including the parts that get dropped, and the parts that no one, including us, can currently verify.

Where the Number Actually Comes From

The statistic traces to a single article: Randall Beck and Jim Harter, “Managers Account for 70% of Variance in Employee Engagement,” published in the Gallup Business Journal on April 21, 2015. That article is not itself the study. It reports a finding from a separate Gallup publication, State of the American Manager: Analytics and Advice for Leaders, released in April 2015.

What the Claim Actually Says

Here is the sentence, as Gallup wrote it. Beck and Harter report that managers account for “at least 70% of the variance in employee engagement scores across business units” — and they close the sentence by attributing it to what Gallup estimates in State of the American Manager.

Read that again with an eye on the qualifiers, because there are four of them and the industry routinely discards three:

“At least.” This is a floor, not a point estimate. “70%” and “at least 70%” are different claims, and only one of them is Gallup's.

“Estimates.” Gallup hedged its own finding. When a downstream citation renders this as “Gallup found” or “research proves,” the citation has upgraded the claim past what its author asserted.

“Engagement scores.” The dependent variable is a survey instrument — Gallup's Q12 — not productivity, retention, or profit. Those linkages exist in Gallup's work, but they are a separate argument.

“Across business units.” This is the one that actually changes the meaning, and it deserves its own section.

The Three Words Everyone Drops

“Across business units” specifies what kind of variance is being explained. The claim is about the differences between teams, not about the absolute level of engagement in any of them. It answers the question: when two units inside the same company score differently, what accounts for the gap? Gallup's answer is that most of it tracks to the manager.

That is a meaningfully narrower statement than the one in general circulation. Consider a company where every single unit scores identically and mediocre. Between-unit variance is near zero. There is nothing for the 70% to explain — and yet engagement is still poor, and the causes are still real. The statistic is structurally silent about that company. It describes the spread, not the level.

The practical translation: this number tells you how much of the gap between your best and worst teams is attributable to who is running them. It does not tell you how much of your engagement ceiling is attributable to managers. Those are different problems with different owners and different budgets.

What Is Genuinely Behind It

The evidentiary base is not thin. In the 2015 article, Gallup describes having studied performance at hundreds of companies and measured the engagement of 27 million employees and more than 2.5 million work units over the preceding two decades. Gallup's own landing page for State of the American Manager describes the report as a study of 2.5 million manager-led teams across 195 countries, drawing on more than four decades of talent research.

No competitor in this space has anything comparable. Gallup has also published in the peer-reviewed literature: Harter, Schmidt, and Hayes (2002), in the Journal of Applied Psychology, analyzed 7,939 business units across 36 companies and established business-unit-level relationships between engagement and outcomes including productivity, profitability, turnover, and safety incidents. That paper is real, it is peer-reviewed, and it is frequently the substance people are gesturing at when they invoke Gallup's credibility.

It is also not this finding. The 2002 paper addresses engagement-to-outcomes; the 70% claim is a variance decomposition attributing between-unit engagement differences to the manager. Those are different analyses. The 2015 Business Journal article does not report the model specification, the confidence interval, the covariates, or the decomposition itself — it reports the conclusion and points to the services report.

So we went looking for the analysis itself. The 2015 article does not report it — no model specification, no covariates, no confidence interval, no method note — and we could not locate any peer-reviewed publication that presents it. That is a statement about what is publicly inspectable, not an allegation that the work was not done. State of the American Manager, where the estimate originates, reaches readers through a registration form rather than a journal, and the decomposition behind its headline number has never been put where anyone outside Gallup can examine it.

The honest summary: a very large proprietary dataset, a plausible and widely replicated-in-practice finding, and a methodology that is not publicly inspectable. Reasonable people can accept the estimate and still want the model.

Why the Imprecision Matters Practically

Loose citation is not a pedantic concern. It produces a specific, expensive misallocation, and we see it in the field regularly.

When “at least 70% of the variance across business units” becomes “managers drive 70% of engagement,” the natural conclusion is that engagement is a manager-quality problem, and the natural response is to buy manager training. Budget flows to the individual. Meanwhile the system variables that determine whether a competent manager can succeed at all go unexamined: span of control, role clarity, staffing levels, conflicting priorities cascaded from above, the promotion criteria that put this person in the role, and whether the organization has given them any real authority over the things their people care about.

Gallup's own article, notably, does not make the training argument. It makes a selection argument. Beck and Harter report that roughly one in 10 people possess high talent to manage, that about 18% of people currently in management roles demonstrate that high talent, and that companies miss on high managerial talent in 82% of their hiring decisions. Their recommendation is better identification and placement — a systems intervention. The industry took a systems finding and sold it back as a workshop.

There is a second consequence. Because the statistic is about spread rather than level, an organization that successfully improves manager quality should expect its weakest units to converge upward toward its strongest. That is genuine value. But if the ceiling itself is low — if the best-managed unit in the company scores poorly — then compressing variance will not fix engagement, and a CHRO who promised the board a 70% effect will have a difficult meeting.

The Sharper Questions to Ask Instead

The statistic is most useful as a prompt for diagnostics on your own data. Five questions we would ask before spending anything:

How much of our engagement variance is actually between units versus within them? Gallup's estimate is a population finding across thousands of companies. Yours is an empirical question with an answerable form. Run the decomposition on your own survey data.

How wide is the spread? If your units cluster tightly, manager quality is not your primary lever regardless of what the population statistic says — there is little between-unit variance to attack.

What does our best-managed unit actually score? That number is the practical ceiling your system currently permits. If it is unacceptable, the constraint is structural, and manager development will not reach it.

Which manager-level differences are selection and which are development? These have different costs, different timelines, and different owners. Gallup's evidence leans harder on selection than most citations of Gallup acknowledge.

What are we asking managers to absorb that no manager could? Span of control, resource asymmetry between units, and unclear or competing goals all show up in engagement scores as manager effects. The attribution is real. The cause may sit two levels above them.

Why We Wrote This

Gallup did the work, published the finding with appropriate hedges, and named the report it came from. The imprecision is ours — the industry's. Every time the qualifiers get sanded off in a deck, a real and useful finding gets converted into a blunter claim that supports a worse decision.

We think how a firm handles a statistic is diagnostic. A consultancy that rounds a source in its marketing will round an analysis in your engagement study. We would rather show you the sentence, show you what it does and does not support, and tell you plainly where the methodology is not publicly available. That last part is uncomfortable to write. It is also the part that should make the rest of this credible.

Sources

Beck, R., & Harter, J. (2015, April 21). “Managers Account for 70% of Variance in Employee Engagement.” Gallup Business Journal. https://news.gallup.com/businessjournal/182792/managers-account-variance-employee-engagement.aspx — Source of the exact claim wording (“at least 70% … across business units,” “Gallup estimates”); the 27 million employees / 2.5 million work units / hundreds of companies / two decades figures; and the managerial talent figures (1 in 10, 18%, 82%).

Gallup. (2015, April). State of the American Manager: Analytics and Advice for Leaders. https://www.gallup.com/services/182138/state-american-manager.aspx — The report the 70% estimate is drawn from. Describes itself as a study of 2.5 million manager-led teams in 195 countries measuring the engagement of 27 million employees, based on more than four decades of talent research. Available by registration.

Harter, J. K., Schmidt, F. L., & Hayes, T. L. (2002). “Business-Unit-Level Relationship Between Employee Satisfaction, Employee Engagement, and Business Outcomes: A Meta-Analysis.” Journal of Applied Psychology, 87(2), 268–279. https://doi.org/10.1037/0021-9010.87.2.268 — Peer-reviewed Gallup research on engagement and business outcomes across 7,939 business units in 36 companies. Cited here to establish that Gallup's engagement research is represented in the peer-reviewed literature; this paper does not contain the 70% variance decomposition.

Gallup. Q12 Meta-Analysis Report. https://www.gallup.com/workplace/321725/gallup-q12-meta-analysis-report.aspx — Gallup's ongoing meta-analytic program linking business-unit engagement to performance outcomes, referenced within the 2015 Business Journal article.

Editorial note: all Gallup figures in this draft were verified against the primary Gallup pages listed above on 2026-09-04. The one previously open item — whether a peer-reviewed publication presents the variance decomposition behind the 70% figure — was searched independently and none was found; the passage now states only what is verifiable, and no open items remain. No statistic in this draft is sourced from a secondary blog or aggregator; several such sources circulate materially different figures (e.g., different employee and work-unit counts) and were deliberately excluded.

Nearly every people strategy written in the last two decades rests on an assumption that no longer holds. The assumption is that managers are the engaged ones — that they arrive at the team meeting with more conviction than the people sitting across from them, and that strategy, culture, and change therefore travel outward and downward through them. Gallup's latest global data show that assumption has quietly expired. The first link in the cascade is now the weakest one.

A Nine-Point Drop in Three Years

Gallup's State of the Global Workplace: 2026 Report tracks engagement separately for managers and individual contributors. Between 2022 and 2025, manager engagement fell from 31% to 22% — a nine-point decline, five points of which landed in a single year between 2024 and 2025 (Harter & Pendell, Gallup, April 7, 2026). Over the same period, individual contributors moved from 20% to 19%, dipping to 18% in 2023 and 2024 before recovering a point in 2025.

Read those two lines against each other. One population lost nearly a third of its engaged share in three years. The other held roughly flat. Global engagement overall fell from 23% to 20%, its lowest level since 2020, and Gallup attributes much of that decline specifically to managers. Each percentage point represents approximately 21 million fewer engaged employees worldwide. Low engagement cost the global economy roughly $10 trillion in lost productivity last year — about 9% of GDP (Gallup, 2026).

The Premium Is Gone

Name the change plainly: the manager engagement premium has collapsed. In 2022, managers led individual contributors by 11 percentage points. In 2025, they lead by three. Gallup states the mechanism directly — managers “typically experience an ‘engagement premium’ of consistently higher levels of engagement compared to individual contributors. But increasingly, managers’ engagement levels are getting much closer to those they lead.”

Precision matters, so be precise: managers are not yet less engaged than their teams. Three points still separate them. But a three-point premium is not a premium a strategy can rest on — it sits inside the range where the distinction stops being operationally meaningful. The trend line, not the current level, is the planning input. Treating managers as a reservoir of surplus discretionary effort is now a bet against the direction of the data rather than a reading of it.

Every Cascade Now Has a Broken First Link

Consider what a standard people strategy actually asks managers to do. Cascade the strategy. Deliver the engagement survey debrief. Model the values. Own the change conversation. Champion the new tools. Each of these is a transfer of energy, and each assumes the manager has surplus energy to transfer. At 22% engagement, roughly four in five managers do not.

This is not an abstract concern. Gallup's Q1 2026 U.S. workforce survey found that, aside from technical integration, the strongest predictor of whether an employee uses AI is whether their direct manager actively champions it. Organizations are routing the largest technology transition in a generation through the population whose engagement is falling fastest. The same structural dependency applies to every culture initiative, every retention effort, and every restructuring on the current roadmap.

A Second Dataset, Arriving Independently

Gallup measures an outcome. McKinsey measured the mechanism — a different instrument, a different sample, and a different set of questions, drawn from a global survey of 706 middle managers.

Middle managers spend less than 25% of their time actually managing people and nurturing talent, and 43% say they are burned out (McKinsey, 2023). Asked what most negatively affected their experience in the role, 44% named organizational bureaucracy — the most-cited factor in every region surveyed (Field, Hancock & Quintero, McKinsey, 2024). Only 20% strongly agreed that their organizations help them succeed as people managers; 42% either disagreed or were unsure.

That last cluster is the tell. Bureaucracy, time allocation, and role design are not personal attributes. They are properties of the system the person is working inside.

Two Sources, One Structural Diagnosis

State this explicitly, because it determines what you do next: two independent research programs, using different instruments on different populations in different years, converge on the same diagnosis — and that diagnosis is structural, not individual.

An individual explanation would predict variance. Some managers thriving, others struggling, sorted by selection quality, capability, or personal resilience. That is not the pattern in the data. The decline appears across regions simultaneously, no region of the world increased engagement in the past year, and bureaucracy is the top-cited complaint everywhere McKinsey surveyed. Gallup further links the sharpest regional drop — an eight-point fall in South Asia — to organizational flattening: fewer management roles, wider spans of control, and the same underlying responsibilities.

The historical record sharpens the point. Gallup found in 2018 that managers already reported frequent or constant burnout at a slightly higher rate than individual contributors — 26% versus 24% — during a period when managers still held a double-digit engagement premium (Wigert & Agrawal, Gallup, 2018). The role has been demanding for a long time. What changed between 2022 and 2025 is not the demand. It is that the compensating conditions — autonomy, time to do the actual work of managing, and organizational support — eroded until the premium could no longer hold.

What a CHRO Should Stop Doing First

The instinct is to launch a manager development program. Resist it. Training a manager to be more resilient inside a role that structurally allocates less than a quarter of their time to managing is a category error — it treats a design defect as a skill gap, and it adds one more obligation to the calendar that created the problem. The first move is subtraction, not addition.

• Stop treating manager engagement as a lagging indicator. Most engagement reporting either folds managers into the aggregate or reads their scores as a by-product of team health. Break the population out, trend it separately, and report it to the board on its own line. A divergence you do not measure is a divergence you cannot manage.
• Stop adding to the manager role without subtracting from it. Every initiative that HR, finance, legal, and IT launch this year lands on the same calendar. Institute a subtraction requirement: no new manager obligation ships without naming what it replaces.
• Stop flattening without redesigning the role. Removing a layer redistributes its work rather than eliminating it. Gallup's own analysis ties wider spans of control to lower manager engagement. If the structure changes, the role definition and the decision rights must change with it — in the same project, not a later one.
• Stop routing every program through managers by default. Some communications, some compliance, and much administration can go direct. For each planned cascade, ask whether the manager is genuinely the right channel or merely the available one.

The Gap Is Design, Not People

None of this is destiny. Gallup notes that some organizations engage their managers at four times the global average. McKinsey's analysis of Organizational Health Index data found that organizations whose managers demonstrate strong managerial behaviors realized from three to 21 times greater total shareholder return over five years — a correlation drawn from a sample of 50 companies rather than a causal claim, but a substantial signal (Field, Hancock, Smallets & Weddle, McKinsey, 2023).

The distance between those organizations and the global average is not explained by better people. It is explained by better-designed roles. The CHROs who move first will be the ones who stop asking their managers to carry more, and start asking what the role was built to hold.

Sources

1. Harter, J., & Pendell, R. (2026, April 7). “Global Employee Engagement Continues Decline.” Gallup. https://www.gallup.com/workplace/708071/global-employee-engagement-continues-decline.aspx
2. Gallup. (2026). “State of the Global Workplace: 2026 Report.” https://www.gallup.com/workplace/349484/state-of-the-global-workplace.aspx
3. Gallup. (2026, January 13). “Span of Control: What’s the Optimal Team Size for Managers?” https://www.gallup.com/workplace/700718/span-control-optimal-team-size-managers.aspx
4. Wigert, B., & Agrawal, S. (2018, July 16). “Employee Burnout, Part 2: What Managers Can Do.” Gallup. https://www.gallup.com/workplace/237119/employee-burnout-part-2-managers.aspx
5. Field, E., Hancock, B., Mugayar-Baldocchi, M., & Schaninger, B. (2023, March 10). “Stop wasting your most precious resource: Middle managers.” McKinsey & Company. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/stop-wasting-your-most-precious-resource-middle-managers
6. McKinsey & Company. (2023, July). “Unleashing the power of the middle manager” [McKinsey Live webinar, drawing on Power to the Middle]. Source of the “less than 25% of time” and “43% burned out” figures. https://www.mckinsey.com/featured-insights/mckinsey-live/webinars/unleashing-the-power-of-the-middle-manager
7. Field, E., Hancock, B., & Quintero, E. (2024, March 20). “Middle managers can succeed by simplifying the role.” McKinsey & Company. Source of the “44% bureaucracy” figure. https://www.mckinsey.com/capabilities/people-and-organization/our-insights/the-organization-blog/middle-managers-can-succeed-by-simplifying-the-role
8. Field, E., Hancock, B., Smallets, S., & Weddle, B. (2023, June 26). “Investing in middle managers pays off—literally.” McKinsey & Company. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/investing-in-middle-managers-pays-off-literally

I get this question all the time, usually from HR leaders or senior executives trying to figure out where to spend their L&D budget.

"Should we invest in executive coaching or leadership development?"

My honest answer is that it's the wrong question. They're not competing options. They do different things. The more useful question is: what does this leader — or this team — actually need right now?

Let me break down how I think about it.

Leadership Development Is About Building Capability Across a Population

Leadership development programs are designed to move a group of people. You're typically trying to build specific competencies — strategic thinking, communication, change leadership, whatever your organization has identified as critical — across a cohort of managers or leaders.

The output is a more capable leadership population. Done well, it shifts norms. It creates shared language. It builds a bench. It's particularly powerful when you're trying to change how leadership shows up as a whole — not just fix one person.

The tradeoff is that programs are, by definition, designed around shared needs. They're built for what most participants need, not for what any individual needs. Which is why they sometimes feel generic.

Executive Coaching Is About Unlocking a Specific Person

Executive coaching is one-on-one. It's confidential. It's focused on the specific person in front of you — their context, their challenges, their patterns, their goals.

Good executive coaching helps a leader see themselves more clearly, make sense of the dynamics around them, and change specific behaviors that are limiting their effectiveness. It creates conditions for real self-awareness in a way that classroom programs can't.

In my experience, coaching is most valuable when:

• A high-potential leader is getting in their own way and doesn't know it
• Someone is in a new role and the transition isn't going smoothly
• A senior leader is dealing with a specific behavioral pattern that's creating friction — and they're ready to work on it
• Leadership is asking a lot of someone and they need a confidential thinking partner

Coaching doesn't work as well when the leader isn't genuinely committed to the process. It's not remediation. It's development — and development requires the person to want it.

The Overlap People Miss

Here's where it gets interesting: the best leadership development programs use coaching methods. And the best executive coaches draw on the same research that informs good leadership development.

Coaching skills — asking better questions, listening more carefully, creating space for reflection — are also what distinguishes great leaders from adequate ones. When we build leadership development programs at gothamCulture, coaching techniques are usually embedded in how the learning is delivered, not just what's taught.

And executive coaches who are good at their work don't just reflect things back at you. They bring frameworks. They bring data. They bring perspective from working with other leaders in similar positions. That's not so different from what a good L&D program does.

How to Decide

Here's my rough decision rule.

Use leadership development when: you're trying to build capability across a group, align your leadership population around shared principles, or create a cultural shift in how leadership shows up in your organization.

Use executive coaching when: a specific individual needs focused, personalized support — especially when their development needs are too personal, too sensitive, or too situationally complex for a group setting.

Use both when: you're running an enterprise leadership development program and want to maximize individual transfer. Coaching paired with cohort learning dramatically increases the likelihood that what people learn in the program actually changes what they do.

What We See Most Often

Organizations that invest only in programs often end up frustrated that behavior doesn't change. The training happens. People rate it highly. Nothing shifts. Part of why is that programs without individual accountability or follow-through let people stay comfortable.

Organizations that invest only in coaching can build strong individual leaders but miss the culture-level change that only happens when leadership development works at scale.

The combination is what actually moves organizations.

One More Thing

If you're evaluating executive coaching vendors or leadership development partners, the framing matters. Look for people who are honest about what each approach can and can't do. Be skeptical of anyone who tells you one size fits all.

Culture, leadership, and development are context-dependent. What works in a fast-scaling tech company won't work the same way in a century-old healthcare system. The best practitioners start by understanding your situation — not by pitching their favorite solution.

If you're trying to figure out what your organization actually needs, we're happy to think through it with you. That conversation doesn't cost anything and usually clarifies more than people expect.

I'm a fan of data. I've spent the better part of my career trying to help organizations move away from gut-feel decisions about their people and toward something more rigorous. So what I'm about to say isn't an argument against people analytics.

It's an argument for using it correctly.

Organizations are investing heavily in people analytics right now — HR dashboards, engagement platforms, turnover predictors, workforce planning tools. The technology has gotten genuinely good. And yet I talk to leaders regularly who have access to more people data than ever before and feel less confident about what's actually happening in their organizations, not more.

The problem isn't the data. It's what we're actually asking the data to answer.

What People Analytics Is Good At

Let me start with the genuine strengths, because they're real.

People analytics is excellent at quantifying outcomes. Turnover rates, time-to-hire, absenteeism, internal mobility, pay equity, succession pipeline depth. These are things that can be measured accurately, tracked over time, and compared across teams, departments, and peers. If you want to know whether your engineering team turns over faster than your sales team, people analytics will tell you. If you want to know whether a particular manager's team has significantly higher absenteeism than the rest of the organization, the data will surface it.

It's also good at identifying correlations that prompt useful investigation. When people analytics shows that employees who complete a particular onboarding experience have 30% better retention at the 18-month mark, that's worth paying attention to. It's not proof of causation, but it's a signal worth following.

And it's good at making the case for investment. When HR wants to make the business case for a leadership development program or a culture initiative, having data that connects people practices to business outcomes is enormously useful. This is probably where people analytics has delivered its most durable value.

Where People Analytics Falls Short

Here's where I push back on the field a little.

People analytics is good at measuring what happened. It's much weaker at explaining why. And when it comes to organizational culture, the why is almost everything.

Your engagement survey tells you that 40% of employees are disengaged. It does not tell you what is driving the disengagement. Is it poor management in specific teams? Unclear career paths? A leadership team that says one thing and does another? Compensation that's out of step with the market? A culture of overwork that nobody names out loud but everyone experiences? The survey score is real. But the number, by itself, tells you almost nothing about where to intervene.

Culture is fundamentally a set of shared beliefs, norms, and behaviors. Those things are embedded in stories, in how people talk about what happened at last week's all-hands, in the unwritten rules about what gets you rewarded or punished here. You can't fully capture that in a Likert scale. The most important things about culture often exist in the spaces between the questions you thought to ask.

The Specific Mistake I See Most Often

Organizations conflate measurement with insight. They run the survey. They get the scores. They present the scores to leadership. And then — because the data exists and looks rigorous — they treat it as a diagnosis when it's actually just a description.

"Our culture scores are down in the innovation category" is a description. It is not a diagnosis. Treating it as one leads to interventions that address the symptom (the score) rather than the underlying dynamic.

I've watched organizations pour resources into "innovation culture" programs — design thinking workshops, hackathons, failure-celebration events — while completely ignoring the performance management system that was actively penalizing people for taking risks. The analytics said innovation was low. The analytics could not see that the reward system was the problem. So nobody looked there.

What to Do With the Data Instead

People analytics is most useful when it's paired with qualitative sense-making. The numbers tell you where to look. The conversations tell you what you're actually seeing.

When engagement scores drop in a particular division, the right response isn't to immediately launch an intervention. It's to go find out what's happening. Talk to the people. Talk to their managers. Understand the context. The data has pointed you toward a location — now do the diagnostic work to understand what's actually going on there.

Similarly, when people analytics surfaces a correlation — say, a link between a specific manager's style and higher team performance — treat it as a hypothesis to investigate, not a conclusion to act on. Maybe the manager is exceptional. Maybe they're inheriting better talent. Maybe the team's performance metrics favor their type of work. The data is telling you something interesting. It is not yet telling you what to do.

The Culture Measurement Question Worth Asking

If your organization is trying to use data to understand its culture, the question I'd suggest starting with isn't "what are our engagement scores?" It's "what are the behaviors we actually want to see in this organization, and are we measuring whether they're happening?"

That's a harder question to operationalize. It requires having a clear point of view about what your culture should be — not just survey questions about whether employees feel valued or understood.

Our Culture Mosaic Survey is built around this distinction. It measures specific behaviors and conditions that drive culture, not just sentiment. The goal is to give organizations data that actually points toward action — not just a score to report.

The Honest Summary

People analytics is a genuinely useful tool when it's used for what it's actually good at: quantifying outcomes, surfacing patterns worth investigating, and making the case for investment in people.

It is not a substitute for leadership judgment. It is not a diagnostic tool for culture. And it will not tell you what to do — only where to look.

The organizations using people analytics most effectively are the ones that haven't outsourced their thinking to the dashboard. They use the data to ask better questions, then they go ask them.