Skip to Content Skip to Menu Skip to Footer

By Hyunyoung Choi, Jeongwook Choi, and Ada Lee


Historically, the office has been a space where people can conduct work-related tasks, interact, ideate and collaborate to develop their careers.

By using the frequency of online communication between co-workers as a proxy for collaboration, we developed a hybrid office/remote model that allows employees to safely return to the office by minimizing office occupancy rate while maximizing collaboration opportunities and in-person interactions.

This model achieves an approximate 50% improvement on 3 days in-person interaction compared to random scheduling when at the office and offers 80% of the critical in-person interactions throughout the week compared to the pre-COVID era’s office-only model.


What is the office for? In the wake of the global coronavirus pandemic, corporations have begun to question their real estate footprint. At the same time, few issues are more immediately polarizing for employees. Some emphasize the benefits of collaboration and proximity in the office, while others are reluctant to return to lengthy commutes when remote work has been a time of increased productivity for many white collar workers. Corporations and employees alike are calling into question the office’s necessity as a space for collaboration and advancement. Businesses are realizing the extent to which they can achieve their commercial goals while employees work remotely. But what are these companies and employees missing in the absence of propinquity?

Originally, the office was conceived to be a place where proximity to others would promote productivity through in-person collaboration and communication. Companies fought hard for prime locations around the globe, selecting for large urban centers favored by so-called “knowledge workers”. However, during the pandemic, many office workers were forced to work from home. Many people were surprised by their ability to focus while working remotely and some moved to cities with a lower cost of living to take advantage of the flexibility remote environment offered. Partly, this was the result of technology platforms that predate the pandemic. Meetings and conferences continued using online applications and the chat function in applications like Microsoft Teams and Slack replaced the walk over to a neighboring cubicle for a quick discussion. The absence of daily commutes allowed office-workers to spend more time with their families. Employees are now challenging the role of the workplace, demanding greater flexibility.

Work has become a thing, not a place. The concept of a workplaces is no longer confined to physical offices, but extended to online ones where people can conduct work-related tasks and develop their careers. But not all businesses and employees are alike. Some businesses and jobs cannot be fulfilled virtually. Some are reporting “Zoom Fatigue” from having video conferences for hours. Physical and mental boundaries around work and home have become murky.

Many firms such as Amazon, Twitter and Facebook are discussing possibilities to permanently switch over to a remote environment whereas others need people on site to run the business. S&P Global has taken a methodologically rigorous approach to reimaging the workplace. Our in-house data science team took a deep dive into communication patterns across the firm and developed a new approach that will explore the right balance between on-site and remote by analyzing the tradeoffs between flexibility and productivity. Our analysis has yielded a hybrid model in which employees spend some days in the office promoting collaboration through physical proximity, while giving employees the freedom to work from home based on employee needs and concerns on the public health.

Our goal was to create a hybrid schedule to maximize collaboration among teams while minimizing office occupancy to observe the social distance measure for the safety of employees. We began by analyzing two main data sources – employees’ data and electronic communications data. Employees’ data included basic information such as employee id, location, division, and manager. To understand the extent to which individuals and teams collaborated, we leveraged communication data, mapping each employees communication network using frequency of communication as a proxy for productivity when working together on-site.

The Model

To reimagine the workplace, we are proposing a hybrid model of on-site and virtual.

We approached the problem using network analysis. First, we respected the organizational design and assumed a new schedule would need to have team leads and team members come to work on the same day. Each team is defined by their reporting hierarchy collected from organizational data. While doing so, we put a cap on the maximum size of the team. This cap will vary depending on office size and utilization rates. By limiting the size of the teams, we hoped to promote collaboration across teams rather than just having each team alone on-site on their “assigned days”.

We used electronic communications data (email and messaging) as a proxy for level of collaboration. Based upon the volume and intensity of communication between teams we identified those teams that were strongly linked and likely to benefit from in-person interaction. Defining each node as an individual or a team, we generated a picture of the network of S&P Global employees. 

From this complete picture, we dove in further to detect communities or groups that are loosely collaborative due to  a few inter-group connections. Working a bit at a time, we broke down the full network into communities by finding connections and relationships that are most likely to act as a bridge between groups [Figure 1]. Because the network was created based on communications data, we believe the communities we uncovered will be highly effective when physically working together. This allows S&P Global to create a hybrid schedule that allows each community member to come to work on the same day.

Figure 1.

We assumed that teams with more frequent communications can achieve higher productivity when working together in person at the office. So we used linear programming to maximize productivity by scheduling closely linked teams to come to the office on the same day. Figure 2 is an example of how a schedule of approximately 110 employees would look like if each of them were to be on-premise 3 days a week. The scheduling is color-coded by the department.


Figure 2.


We simulated different work schedules based upon different office utilization rates. We conducted separate studies on 1, 2, 3 day scheduling with office capacity caps at 25%, 45% and 65% respectively. Then we compared the results to different baselines to measure the effectiveness of the recommendation system.

By distinguishing highly collaborative teams from the full pack and efficiently scheduling to maximize those teams’ time together on-site, we achieved an approximate 50% improvement on 3 days in-person interaction compared to random scheduling when at the office [Table 1]. When comparing in-person interaction, we were comparing the daily average number of interactions each person gets with the list of people they communicate with the most.


1 day in the office

2 day in the office

3 day in the office


Group 1

Group 2

Group 1

Group 2

Group 1

Group 2

Office 1







Office 2







Office 3







Office 4