Ask someone why they left a job and you will hear about the company. The culture shifted. Leadership lost the plot. Ask what their last six months actually felt like and the story shrinks to about eight people: one manager, a handful of peers, and whoever made every standup heavier than it needed to be.
That smaller unit is where the evidence lives. Inside a single company, on the same pay bands and the same benefits package, turnover swings hard from one team to the next. Company-wide culture is held roughly constant in that comparison. Team culture is the variable that moves.
The Variance Sits Between Teams Inside the Same Company
Gallup's most-repeated finding on this is that managers account for at least 70% of the variance in employee engagement scores across business units (Gallup, "Why Great Managers Are So Rare"). The phrase doing the work there is across business units. Those units sit inside the same organizations, sharing a brand, a compensation philosophy, and a values statement. Engagement still swings wildly between them.
Turnover follows the same pattern. Comparing top-quartile to bottom-quartile engaged business units, Gallup reports an 18% difference in turnover at high-turnover organizations and a 43% difference at low-turnover ones, alongside an 81% difference in absenteeism (Gallup on culture versus benefits and perks). Two teams, one payroll system, very different odds that any given person is still there next year.
A Raise Does Not Buy Back the Manager Relationship
Gallup puts a price on a good manager in plain numbers: it takes more than a 20% pay raise to lure most employees away from a manager who engages them, and next to nothing to poach most disengaged workers (Gallup, "The Great Resignation Is Really the Great Discontent").
In that same Gallup analysis, job-search behavior scales directly with engagement level. As of March 2021, 74% of actively disengaged employees were searching for or watching for other jobs. That drops to 55% among not-engaged employees and 30% among engaged ones. Same labor market, same posted salaries elsewhere, a 44-point spread in who is looking.
This is where retention budgets break. Compensation is the lever executives reach for because it is centrally controllable. The engagement level that decides whether a raise lands is set locally, by a manager and a group of peers nobody on the comp committee has met.
Peers Carry as Much of the Load as the Manager
Collapsing team culture into manager quality misses half of the picture. Gallup finds only 2 in 10 U.S. employees report having a best friend at work, and the ones who do get more done, work more safely, and innovate more. Since the pandemic, Gallup describes an even stronger relationship between having a best friend at work and outcomes including intent to leave, overall satisfaction, and likelihood to recommend the workplace (Gallup on workplace friendships, Gallup Panel surveys run annually from 2019 to 2022, samples of 13,594 to 16,586 U.S. adults). Distributed work stripped out ambient connection and made the surviving ones more load-bearing, which is one reason building trust on a remote team takes deliberate structure.
The strongest academic version of this tracks departures instead of sentiment. Mossholder, Settoon and Henagan followed 176 healthcare employees at a large public medical center for five years and used survival analysis to model who actually left ("A Relational Perspective on Turnover," Academy of Management Journal, 48(4), 607-618). The study modeled four relational variables. Two predicted departures: network centrality, how embedded a person is in the coworker network, and interpersonal citizenship behavior, how much they help and support colleagues. The two attitudinal ones, perceived coworker support and felt obligation toward coworkers, did not. What people did and where they sat in the network beat what they reported feeling.
Read that as a warning about who leaves first. The person on the edge of the team's network, never pulled into the side conversations, is measurably more likely to go, and how supported they say they feel will not tell you in advance.
What the Research Establishes, and Where It Stops
The "people leave managers" line deserves the precision its own evidence carries. Gerstner and Day's meta-analysis of leader-member exchange research found significant relationships between the quality of the individual employee-manager relationship and turnover intentions, job performance, satisfaction with supervision, and commitment. The relationship to actual turnover was not significant in that analysis (Journal of Applied Psychology, 82(6), 827-844). Intention and departure are separate outcomes, and the meta-analytic evidence is stronger for the first one.
The effect also moves with national culture. A separate meta-analysis by Rockstuhl, Dulebohn, Ang and Shore, covering 282 independent samples from 23 countries, found the link between manager-relationship quality and turnover intentions runs stronger in horizontal-individualistic (largely Western) cultures than in vertical-collectivistic (largely Asian) ones (Journal of Applied Psychology, 97(6), 1097-1130). A globally distributed organization should not assume one effect size across regions.
Two limits are structural. First, no published study partitions team-level cohesion, company-wide culture, and pay into a single model as competing predictors. Griffeth, Hom and Gaertner's 2000 meta-analysis (Journal of Management, 26(3), 463-488), described on publication as the most wide-ranging quantitative review of turnover antecedents to date, does not run that three-way comparison either. The causal picture gets built by triangulating separate lines of research. Second, no credible figure isolates the cost of turnover caused specifically by team dysfunction. The general figures: Gallup estimates U.S. businesses lose roughly $1 trillion a year to voluntary turnover, with replacement cost per employee running 0.5x to 2x annual salary (Gallup on the $1 trillion problem). Anyone quoting you a team-dysfunction-specific dollar amount is extrapolating from that.
Psychological Safety Is the Team-Level Property You Can Measure
Team culture sounds unmeasurable until you pick a construct validated at the team level. Amy Edmondson's study of 51 work teams inside a single manufacturing company, spanning sales, production, product development and staff functions, found psychological safety behaves as a shared team belief, with a group-level intraclass correlation of .39 (F(50,427)=6.98, p<.001) (Administrative Science Quarterly, 44(2), 350-383). Statistically, the team is the unit of analysis: people inside a team converge on how safe it feels to speak up, and teams inside one company diverge.
Gallup models the retention consequence directly. Moving the share of U.S. workers who strongly agree that "at work, my opinions seem to count" from 3 in 10 to 6 in 10 is associated with a 27% reduction in turnover, 40% fewer safety incidents, and 12% higher productivity (Gallup on psychological safety). That is a team-level construct with a retention number attached to it, which puts it ahead of most culture language.
Half of Departures Were Preventable and Half Got No Conversation
52% of voluntarily exiting employees say their manager or organization could have done something to prevent them from leaving. 51% say that in the three months before they left, neither their manager nor any other leader spoke with them about their job satisfaction or their future with the organization (Gallup). The intervention window was open and went unused.
What closes it is unglamorous and team-sized:
- Run stay conversations on a schedule. Once a quarter, per person, on how the job is going and where it is headed. Half of leavers never had that conversation in their final three months.
- Treat speaking up as an observable behavior. Psychological safety shows up in who challenges a plan in a meeting and what happens to them afterward. Notice who has stopped speaking up, and when it started.
- Engineer peer connection deliberately. Pairing, rotating collaborators, and shared problems build network centrality. Two in ten people having a close friend at work is an outcome of how the work is structured, and the structure is yours to change.
- Break your turnover data down by team. An org-level attrition number averages away exactly the variance that predicts where the next resignation comes from. Layer that onto the broader set of employee retention strategies that work.
Annual Engagement Surveys Report at the Wrong Altitude
A company-wide annual survey aggregates to the organization, arrives once a year, and asks people to self-report on relationships they have every incentive to describe charitably. The signal a manager can act on is team-level, current, and behavioral.
QuestWorks was built for that altitude. It runs on its own cinematic, voice-controlled platform, with Slack or Microsoft Teams serving as the integration layer. Once a week the whole team plays a 25-minute session at the same time, seated dynamically into groups of 3-6, and the behavior that surfaces under narrative pressure becomes a Team Intelligence Score: 0-100, weekly, with an 8-week trend line, named sub-scores including Role Clarity and Decision Velocity, and a "Do this week" list of concrete actions. Participation is voluntary and opt-in, and the data is never tied to performance reviews. Pricing is $199/month per team with 10 seats included, or $1,990/year, and the trial runs 30 days and four game days with no credit card required.
Whether someone is still there in a year gets decided inside the eight people they work with every day. That unit is the one a manager can influence directly, and the one most retention programs never measure.
