In the late 1990s, Gallup researchers conducting large-scale analysis of workplace engagement survey data found that a single question — "I have a best friend at work" — predicted employee engagement and business outcomes as reliably as any item in their instrument. This was unexpected. The question was considered too personal, too casual, insufficiently aligned with the hard-edged vocabulary of performance management. Several client organizations pushed back when Gallup included it in engagement surveys. A few refused to ask it.
Gallup asked it anyway, and the data held. Across industries, countries, job types, and organizational sizes, employees who agreed they had a best friend at work reported higher engagement, better performance, and greater intention to stay than comparable employees who did not. Business units with high proportions of employees reporting a best friend showed better safety records, higher customer satisfaction, and higher profitability than low-proportion units. The effect size was not trivial.
The finding did not travel smoothly into management practice. The standard objection was that friendship and professionalism are in tension — that organizations should be merit-based, not relational, and that managers who cultivated friendship networks were building fiefdoms rather than teams. A more subtle objection was that the best-friend question was an outcome rather than a cause: maybe high-performing business units naturally produced conditions in which friendship could form, rather than friendship producing high performance. The causal direction was contested.
Gallup's position, developed over two decades of replication and meta-analysis, is that the relationship is bidirectional but that friendship is a genuine driver, not merely an indicator. The mechanism runs through engagement: friendship predicts engagement, engagement predicts performance. An employee with a genuine close relationship at work has a stake in the place that transcends their employment contract. They care whether their friend is supported well, whether the team succeeds, whether the organization's values are honored in practice. That stake produces the behaviors — proactive communication, discretionary effort, accountability, loyalty — that engagement measures capture.
The finding also survived across demographic groups with instructive variation. It was stronger for women than for men in the original Gallup datasets. It was stronger in customer-facing and team-intensive roles than in autonomous individual-contributor roles. It was weaker in highly unionized environments where solidarity norms partially substituted for individual friendship. These variations are consistent with the theoretical mechanism: friendship predicts performance most where performance is most dependent on trust and voluntary coordination.
Twenty-five years after the finding was first published, the "best friend at work" item remains one of the most controversial and most consistently replicated items in the organizational psychology literature. It is controversial because it asks organizations to take seriously something that feels soft — human friendship — as a hard business variable. It is replicated because the data keeps showing the same thing: people do better work when they work with people they genuinely care about.