What Remote Work Removes—and What One Office Day Restores
The monthly-office-day experiment studied a precise version of remote work: would a fully remote customer-service team benefit if its members worked together in the office one fixed day each month? The value of the case is not that it settles the office debate. It lets us see, with unusual clarity, what a coordinated day changed while pay, hours, technology, call routing, and formal training stayed the same. That separation helps identify the coordination problem before we interpret the results.
A fully remote team with unusually clear boundaries
The trial took place in the inbound customer-service division of TEMPO BPO, a large business-process-outsourcing firm in Türkiye. Employees handled calls about billing, packages, upgrades, and service problems. The work was standardized: calls were routed to the first available employee, workers used firm-provided equipment and common client protocols, shifts followed a regular five-day schedule, and performance was monitored through the same operational system. An employee could not choose which call to handle or sort into a preferred team.
That structure matters because remote-work effects are difficult to interpret when the work itself changes from person to person. If one employee chooses easy tasks from home while another handles difficult cases in the office, a difference in output may reflect task selection rather than location or contact. Here, central routing and common procedures narrowed that possibility. The firm also recorded calls per hour, call duration, customer ratings, audit scores, break time, employment status, and attendance. The researchers could therefore compare output with measures that might reveal a speed-for-quality trade-off or a retention effect.
The outcome architecture also gives the treatment more than one way to succeed or fail. A higher call rate could reflect shorter conversations, fewer pauses, or a change in the kinds of calls handled. Stable customer ratings and audit scores would make a simple “workers just rushed” explanation less convincing. Employment records add another layer: if the office day changes who stays, a later productivity gap may reflect both what individuals learned and which employees remained. The trial therefore has to compare daily work measures with employee-level retention rather than treating output as a self-contained verdict on the policy.
The study population was not every employee who might have been affected by an office policy. In March 2024, the firm invited 661 employees in the province to consider a program requiring one office day per month. Participation was voluntary; 476 expressed interest. Eligibility then required at least three months of tenure, no submitted notice to leave, and residence within roughly 45 minutes of the office. Those conditions produced 248 eligible volunteers. The people who did not volunteer most often cited family or care responsibilities. This is already an important boundary: the trial tests a feasible office option for people willing and able to attend, not the average effect of imposing a commute on an entire remote workforce.
Among the eligible volunteers, 124 employees were assigned to the treatment and 124 to the control group. The firm used pre-existing company ID parity: odd-numbered IDs entered the office-day group and even-numbered IDs remained fully remote. Because the rule depended on an administrative number rather than a manager’s judgment, the assignment did not allow HR to select favored employees after the sample was defined. Treatment and control employees were balanced on most measured baseline characteristics and handled almost identical numbers of calls per hour before the first visit.
The design also fixed the managerial environment. Teams and team leaders were established before the intervention, employees could not choose their shifts, and both groups were observed under the same team structure during the June–July pre-intervention baseline. These details do not make the workplace artificial. They make the counterfactual more legible: the control group shows what happened to comparable employees who continued with the existing fully remote arrangement while the treatment group received a new opportunity for coordinated contact.
That counterfactual is more useful than a simple before-and-after comparison. Customer-service performance can change over time because of seasonality, new client demands, worker experience, or ordinary adjustment. Both groups are exposed to those common forces. The relevant comparison is whether the treatment group changes differently after the office-day program begins. Since assignment was made within an already defined volunteer sample, a systematic post-assignment difference is less easily explained by one group having been selected for stronger motivation, better managers, or easier work.
The design still does not make selection disappear everywhere. The volunteer pool may differ from the employees who declined to participate, and the eligibility rules remove people for whom monthly travel was not feasible or who were already preparing to leave. That is not a flaw in the random assignment; it identifies the population to which the assignment applies. Randomization answers whether the policy changed outcomes among eligible volunteers. It does not answer what would happen if the same policy were imposed on employees with different care responsibilities, longer commutes, less stable internet access, or a different kind of work.
The treatment changed contact, not the job
The intervention ran from August 2024 through April 2025. On one fixed day each month, treatment-group employees traveled by company shuttle to the office and worked their regular shifts there. They used the same laptops, software, procedures, and performance criteria as on remote days. There was no overtime, no new bonus, no formal curriculum, and no change in pay or workload. Attendance was approximately 95% of assigned office days, so the experiment largely delivered the policy it intended to deliver.
The office day nevertheless changed two interfaces that were difficult to reproduce in the fully remote arrangement. The first was the manager–employee interface. Team leaders could observe work in real time, answer questions in the moment, and provide coaching without arranging a separate interaction. The second was the peer interface. Employees shared a room, breaks, breakfast, coffee, and lunch. They could ask how a colleague had handled a customer problem, notice an unfamiliar difficulty, or develop enough familiarity to ask for help later through remote channels.
The distinction is subtle but important. The treatment did not add instruction in the conventional sense. It added a common setting in which instruction, feedback, and informal exchange could occur. It also did not require employees to remain in that setting. After each visit, treatment-group employees returned to fully remote work until the next scheduled day. The hypothesized benefit was therefore not the productivity of a permanent office routine. It was the possibility that a small amount of shared time could alter the quality of remote interaction that followed.
The office day also synchronized people who would otherwise encounter one another asynchronously or not at all. A worker could see which problems occupied a colleague's attention, while a team leader could spot a question before it became a recurring mistake. Shared breaks created time that was not assigned to a ticket or a call. Such contact is easy to describe as atmosphere, but the relevant design claim is more concrete: a common setting can lower the cost of asking for help and make a previously unknown colleague legible as a possible source of help.
This is why the intervention's coordination was collective rather than merely individual. A firm could tell each worker to come in once a month on a personal schedule, but that would not create the same shared interface. The treatment brought members of a team together on a fixed day, so the potential benefits came from overlap: a question could reach a peer who was present, a manager could respond while the issue was visible, and a later remote exchange could begin with some shared context. The study does not prove that every one of these events occurred, but it defines the mechanism that the policy made possible.
The seating arrangement made one part of this interface more informative. Within treatment groups, employees were assigned seats at random. The arrangement did not determine whether two people were friendly or whether a manager expected them to collaborate. It created exogenous proximity during the visit. If desk neighbors were more likely to communicate in the following week, that would be evidence that a brief physical connection could persist after the room disappeared. Interactions during breaks were also recorded, but those were not randomized: people who chose to talk may already have been more sociable or more likely to need one another.
That distinction keeps the mechanism analysis honest. The main treatment comparison asks whether being assigned to monthly office days changed outcomes relative to remaining fully remote. The seating analysis asks a narrower question inside the treatment group: did an externally assigned local connection predict later communication? It cannot by itself show that office days raised productivity, and it cannot compare a desk neighbor with a remote control employee as though the seating assignment were the whole treatment. It is useful because it links a specific feature of the office day—proximity—to a later behavior without claiming to explain every path from presence to performance.
The firm also covered direct participation costs such as transport and meals, which made attendance feasible for the assigned sample. That choice is part of the treatment's practical meaning. The tested policy was not simply “require employees to appear”; it included a shuttle, a usable office, common timing, and a day that did not disrupt ordinary operations. If a different employer supplied none of those supports, a lower attendance rate or a larger employee burden could change both the size and the direction of the result.
This is why the experiment can be read as a design test rather than a culture test. It did not compare people who preferred offices with people who preferred homes. It compared assigned exposure to a coordinated day among eligible volunteers. It did not ask whether all work should happen face to face. It asked whether a limited, shared interruption to remote isolation could create useful connections while leaving the remote operating model intact.
The comparison is strongest for the treatment as assigned. Random assignment supports an intent-to-treat estimate, and the standardized task reduces some forms of interference. Repeated administrative observations also let the researchers trace effects before, during, and after the intervention. But repeated observations do not turn 248 employees into thousands of independent experiments. Nor do they remove the limits created by the setting. Customer service is more standardized than many forms of knowledge work; the office was accessible by shuttle; and employees who could not or did not want to attend were not part of the randomized sample.
The time structure is especially valuable because it separates exposure from the moment of assignment. The two groups had a common pre-intervention period, the treatment group then received nine scheduled opportunities for contact, and the post-intervention period tested whether anything remained after the scheduled visits ended. A policy that changes only the day on which people are physically present should not be expected to produce an immediate transformation in every outcome. The design instead allows the question to be asked at the level of accumulation: can repeated low-intensity contact alter later remote work?
High compliance makes the assigned policy easier to interpret, but it does not turn the estimate into a universal treatment-on-everyone effect. Most assigned office days were attended, so the study is not primarily a test of a policy that employees routinely ignored. Still, attendance itself depended on the office being accessible and the participants being willing. In another workforce, the same assignment could generate a different experience because the burden of reaching the office would be higher or because the task would provide fewer opportunities for peer learning.
Those boundaries are part of the result, not an afterthought. A monthly office day may work differently when the main problem is creative collaboration, when teams are distributed across countries, or when attendance imposes substantial care and travel costs. The trial also cannot tell us that physical proximity is the only way to create feedback or peer learning. It tests one low-intensity coordination design. Its purpose is to identify whether the design had consequences, and to provide clues about the interfaces through which those consequences might arise.
The next step is to follow those consequences through time. If the office day merely created a short-lived morale boost, the treatment effect should appear immediately and fade. If it helped people learn from one another, receive better feedback, or build ties that remained useful remotely, the pattern should accumulate across visits and persist after the last one. That time profile is where the coordination hypothesis becomes testable.