How Coordination Becomes Performance and Retention
The monthly-office-day trial does more than report that treated employees eventually handled more calls. Its important evidence is the shape of the change: productivity rose slowly, remained higher after the scheduled visits ended, and was accompanied by communication and retention patterns that fit a cumulative coordination process. This chapter follows that chain while keeping three claims separate: what the randomized assignment establishes, what the mechanism evidence suggests, and what the firm's cost calculation assumes.
The effect built instead of appearing
Before the first office visit, treatment and control employees moved together at roughly 10.8 calls per hour. That common starting point matters because the intervention was not attached to an obvious initial productivity gap. Once monthly office days began, however, the treatment series did not jump immediately above the control series. During the first five months, the estimated effect on calls per hour was close to zero. It rose to 0.67 calls per hour in months six through nine and reached 0.87 calls per hour in the post-intervention window in the specification without employee fixed effects.
The timing makes the simple novelty story less persuasive. If employees worked harder because they had received a new perk, felt selected, or were disappointed that colleagues remained at home, the effect would be expected to appear early. Instead, the divergence grew as employees accumulated repeated encounters. After the final office visit in April 2025, treatment-group employees continued to outperform the control group in the first five months of fully remote work. Their post-period mean was 12.4 calls per hour, compared with 11.5 for the control group, a raw gap of 7.8% relative to control.
That raw gap is informative but not the only number that matters. Because attrition differed between the groups, the remaining employees were not necessarily the same composition as the original assignment. The employee-fixed-effects estimate for the post-intervention period was 0.63 calls per hour. It compares changes within the same employees against changes among control employees, and is therefore closer to a within-worker improvement than the raw post-period difference. The two estimates answer related but different questions: the raw gap describes the observed performance of the groups that remained, while the fixed-effects estimate gives more weight to how individuals changed over time.
This distinction is a general lesson about remote-work metrics. A team-level result can improve because each person works more effectively, because the most productive employees are more likely to stay, or because both things happen together. The trial's fixed-effects specification helps separate the first possibility from the composition of the observed workforce, while the separate attrition analysis keeps the second possibility visible. Neither estimate should be treated as the single “true” percentage. The raw result is relevant to what the firm actually observed after the policy; the within-worker result is relevant to how much improvement can be attributed to changing work experience among the same people.
The event-study pattern reinforces the same interpretation. Before the first visit, treatment and control tracked one another. The effects then accumulated after repeated visits and remained visible after the scheduled contact stopped. This does not identify a single psychological or organizational mechanism. It does make an immediate, temporary response less consistent with the evidence. Learning, better feedback, stronger peer ties, or a greater attachment to the team could all build gradually and continue to matter on remote days.
The productivity margin also matters. Average call duration was essentially unchanged in the first five months, then fell by 13.21 seconds in months six through nine and by 17.60 seconds in the post-intervention window. The combined measure of administrative time, hold time, and breaks did not change significantly. In this operation, the higher call rate appears to have come mainly from shorter calls rather than from a broad increase in paid hours or a detectable reduction in recorded break time.
That pattern is consistent with improved efficiency, but it does not tell us whether employees became better at diagnosing customer needs, found answers more quickly, or received coaching that helped them handle recurring problems. The trial observes the output margin directly; it observes the underlying work practices only indirectly. A mechanism claim must therefore remain more cautious than the treatment-effect claim.
The results also provide a quality boundary. Customer ratings and internal random-audit scores did not show an average decline after the intervention began. In this standardized customer-service setting, treatment employees handled more calls without a detectable deterioration in those measured quality outcomes. That makes a simple speed-for-quality explanation less plausible. It does not prove that every aspect of service improved, or that an entirely different form of knowledge work would display the same relationship. Stable ratings and audit scores are evidence against the measured failure, not evidence that no unmeasured cost existed.
The same restraint applies to employee experience. Treated employees later reported more regular manager feedback, stronger cultural fit, and more effective team communication, but the study did not detect comparable effects on broad life satisfaction or work–life balance. The intervention therefore should not be described as a general well-being improvement. Its most visible consequences were tied to how people worked together and whether they remained with the firm. That is consistent with a coordination intervention: its value may show up in the relationships that support work rather than in every measure of a person's life.
Communication, retention, and the value of a narrow intervention
The strongest mechanism clues concern the relationships that the office day made easier. In the endline survey, treated employees reported spending 36 more minutes communicating with their three most frequent colleagues during the relevant week than control employees. Within the treatment group, pairs assigned as desk neighbors were 11 percentage points more likely to communicate in the following week, relative to a comparison mean of 9%. The random seating assignment is valuable here because it creates a small piece of exogenous proximity: two employees did not become neighbors because they had already chosen to communicate.
Other interaction measures point in the same direction but carry less causal force. Pairs who reported interacting during the morning, lunch, or afternoon breaks had increases of roughly 18–19 percentage points in subsequent communication. Those encounters were not randomly assigned. People who chose to speak may have been more sociable, more experienced, or more likely to need one another. Their association with later communication supports the idea that shared time can create continuing ties, but it should not be read as an independent treatment effect.
The endline survey also found that treated employees were more likely to report very regular feedback from their team leader, stronger cultural fit, and more effective team communication. These results fit the proposed manager and peer channels. Yet the trial cannot tell us how much of the productivity gain came from feedback, peer learning, social connection, or selective retention. The study did not observe the content of informal messages directly, and communication measures came partly from surveys. The correct interpretation is cumulative: several pieces of evidence point toward richer interaction, but no single piece closes the causal mechanism.
Retention supplies a second route through which coordination can affect measured performance. By the end of the observation window, cumulative attrition was 21.0% in the fully remote control group and 13.7% in the monthly-office-day group. The gap remained visible after the last office visit, suggesting that the policy did more than delay employees' decisions to leave. A worker who feels more connected, receives more useful feedback, or learns more easily from colleagues may have more reason to stay.
Retention also changes what the productivity numbers mean. Descriptive comparisons of employees observed after the intervention are consistent with selective retention of more productive workers in the treatment group. In the treatment group, stayers were more productive than quitters; the pattern differed in the control group. This does not replace the randomized attrition estimate with a new causal claim about individual quality. It shows why the raw productivity gap can contain two components: people may improve, and the policy may change which people remain available to be measured.
The retention result also clarifies why connection is an economic mechanism rather than just a morale story. In a job with standardized tasks and limited variable pay, an occasional chance to receive feedback or ask a peer for help may change the everyday cost of solving problems. If that lowers frustration or makes progress more visible, it can affect the decision to stay even without changing wages. The source does not directly observe those individual decisions, so this remains an interpretation supported by the communication and survey patterns, not a separately identified pathway. What the randomized comparison establishes is the difference in exits; the proposed reasons explain why that difference is plausible.
The firm-level cost calculation illustrates why a small coordination intervention might be attractive even if its mechanism is not fully decomposed. Over nine months, transport, meals, refreshments, and office space cost about $13,000 for the treated employees. The seven-percentage-point attrition reduction implied about nine fewer separations; using the firm's estimate of $714 for recruitment, initial training, and the productivity ramp-up of a replacement, the turnover saving was approximately $6,400. The paper then valued the raw 7.8% post-period productivity gap at the employer's labor cost of about $4 per hour, assuming 1,560 hours over the operating horizon. That produced an estimated $60,300 productivity benefit and total monetized benefits of about $66,700—more than five times the program cost.
The ratio is a decision input, not a universal welfare result. It is calculated from the firm's perspective and assumes that the raw post-period productivity gap is a steady-state gain that allows the same call volume to be handled with fewer paid labor hours. It does not price the employee's travel time, care responsibilities, fatigue, or alternative uses of the office. Nor does it establish that every organization would face the same replacement cost or obtain the same productivity margin. A manager using this evidence responsibly would ask whether the local coordination problem resembles the trial, whether the necessary attendance supports are affordable, and whether the relevant quality and retention measures are available.
The broader lesson is narrower than “office days work.” The randomized trial supports the claim that this assigned pattern of coordinated contact improved measured outcomes for eligible volunteers in this customer-service operation. Its time profile and communication results make accumulated learning, feedback, and ties plausible explanations. Its design does not tell us the optimal frequency, the best format for other work, or the relative contribution of each channel. What it does provide is a way to think about remote coordination as an infrastructure problem: if a team loses informal contact, the remedy should be judged by whether it restores a specific function and whether the result persists after the intervention itself is gone.
That conclusion completes the internal side of the question. Remote work can be made less isolating through deliberate coordination, but the trial says little about whether remote arrangements also change who can reach opportunity outside the firm. The next checkpoint connects these two dimensions and preserves the difference between a causal firm-level intervention and broader evidence about career mobility.