To make better, data-driven decisions around how we allocate physical space, Microsoft CSEO has created a platform to acquire and visualize spatial data at all Microsoft facilities. Learn how Microsoft Azure Machine Learning helps us dynamically allocate resources for a rapidly changing workforce.

EXPLORE RELATED CONTENT

Work is changing, and our requirements within the workplace are changing, too. More than ever, we must understand how we use office space and resources, make changes to keep everyone safe, and allocate those resources in the best way to reduce waste.

Inside of Microsoft, our Real Estate & Security (RE&S) team has worked to digitally transform our Microsoft and LinkedIn campuses using machine learning (ML). To facilitate those efforts, the LinkedIn Global Workplace Strategy team and the RE&S team recently partnered with the Core Services Engineering and Operations (CSEO) Data and Insights team to redefine how we measure space utilization across our facilities worldwide. Using ML technology, the team created a new set of models and metrics to quantify and communicate space utilization per room, per floor, and in special areas like restrooms and cafes.

Intelligent planning empowers facilities managers, IT teams, and business leaders

Agile and informed decision-making requires data and a depth of understanding to drive that data toward insight. As we aimed to explore how our spaces were being used, we also envisioned the ways that business and facilities decision-makers would make informed decisions based on that data.

Here’s what stakeholders can do with space usage data, ML, and dynamic visualization:

  • Use spaces in a way that better facilitates desired business outcomes; such as productivity, employee satisfaction and retention, and proper sizing
  • Make data-driven decisions across the life cycle of real estate planning
  • Purchase or lease commercial properties with real usage data backing those decisions
  • Accommodate team needs and allocate spaces based on usage patterns
  • Allocate resources and utilities in special areas like restrooms and cafes
  • Reduce food waste and energy consumption in unused areas of any building
  • Maintain building systems proportionately to their usage and wear
  • Enhance facilities to reflect employee needs, preferences, and patterns

Space usage is only made more complex as our collective reimagining of “work” and “the workplace” evolves. New paradigms in workplace culture have caused a dynamic trend in office usage.

People who worked in cubicles and static workstations for years became mobile, moving between individual and shared workspaces, between buildings and campuses, and between the office and their home office. Knowing for certain how spaces are serving us, and being able to forecast the demand for those spaces, equips leaders to make swift and informed decisions when unforeseen circumstances arise. We’ve seen in 2020 how surprise changes can immediately (and perhaps irreversibly) impact the way we do business. In that, we’ve uncovered new reasons for facilities managers and data scientists to join forces. As the world of work continues to evolve, data and AI will continue to be paramount.

Space decisions require real-time data

Workplace strategists and facilities managers are tasked with long-term decision-making and need to think forward. This level of predictive demand requires a toolkit that can accurately and acutely forecast the demands of buildings, rooms, and resources in the workplace. We posed the question: How can we determine how many people use Microsoft buildings at any given time, which business groups they represent, and how that space is being used?

We need this data to:

  • Make budget-sure, informed building lease and purchase decisions
  • Allocate space and resources with a focus on waste reduction and energy conservation
  • Right-size spaces and load balance buildings based on business group utilization patterns
  • Proportionately maintain whole-building systems, like HVAC
  • Implement a dynamic work environment that promotes employee productivity
  • Create space programs to support a dynamic workforce and greater collaboration

For us, this is part of a larger effort within the Microsoft ecosystem to understand employee needs, increase satisfaction, and manage spending. Our Campus Refresh initiative is set to make more intentional use of our space and redefine what our spaces offer to employees.

What we can achieve with machine learning

By employing our solution, we hoped to achieve or facilitate a more data-centric approach to all space-related decision-making. Our goal is to aggregate and manipulate usage data to facilitate conversations between departments and promote efficiency, productivity, and satisfaction across the entire organization.

  1. Data-driven decisions
    Workspace and facilities planning requires a great deal of insight into the use of those facilities: how employees move between and connect with space, the engagement or pace of work within the environment, and how it fulfills its purpose. Anecdotal and qualitative insights are important, but we need data to solidify and correct for biases and lapses in observation.
  2. Meaningful conversations
    RE&S teams make large, multi-year commitments both in real estate and in structural and building maintenance. They need real-time data to demonstrate how these spaces impact the business. This data can help facilitate dialogue between facilities managers, data analysts, and business leaders. This interdisciplinary focus will be key as workplace decision-making evolves. As the future of work demands more agility and cross-functional thinking at every level, we need solutions to keep everyone informed.
  3. Efficiency
    Changes in headcount growth projections can lead to a lack of needed workstations, restrooms, conference rooms, cafeterias, and kitchens. The opposite is also true, as empty and wasted spaces are common when projections miss the mark. More reliable forecasting would also reduce any deficit or overflow of resources within these spaces, such as office supplies, restroom stock, and kitchen items. Equipped with dynamic, real-time usage data, facilities managers can make changes to immediately reduce waste while forecasting potential scarcities or influx.
  4. Productivity
    Our campuses are built and designed to serve employees, business guests, and visitors. To facilitate employee productivity requires massive intel on what employees need, how they work, and where the current spaces fall short. Likewise, guests should be given correctly sized and properly optimized spaces in which to interact. Surveys can be costly, timely, and put the onus on employees to self-monitor their own patterns and behaviors. Data on these behavior patterns empower facilities managers to accommodate employees’ preferences, tendencies, and even their subconscious rhythms at work.

The machine learning approach to quantifying the workplace

When data is the solution to a problem, the solution has two prongs: data acquisition and data visualization. Our approach retrieves data from multiple sources while using existing infrastructure: badges, Wi-Fi telemetry, employee devices, and weather data. By employing AI and ML solutions using Microsoft Azure, we can:

  • Accurately and anonymously measure, track, and monitor how many people are using a given space at any time, all the time
  • Calculate the requirements of any given space using a supply and demand equation
  • Employ data sensors, organizational data, and Microsoft Office 365 data to understand how each space is being utilized
  • Integrate with Microsoft Power BI to convert ML insights into usable answers for space planners and strategists
Diagram showing Wi-Fi,  badge,  and organizational data blended with CSEO data to arrive at space insights for Microsoft and LinkedIn.

Figure 1. Badge and Wi-Fi telemetry and organizational data blend with CSEO data to provide space insights for decision-makers.

A new way to quantify space usage: Peak average attendance

Peak average attendance (PAA) refers to the average number of employees or customers utilizing a given space over a period of time. PAA empowers teams to quantify space utilization, assigns a universal metric to help communicate space usage forecasts, and allows teams to make more informed, data-driven decisions about manipulating the measured spaces and optimizing for better outcomes. We can use PAA to augment space planning initiatives, facilitate food service programs, inform campus maintenance agreements and schedules, and quantify employee behavior and engagement within every space. 

Technology specifications

Using a combination of Microsoft Azure Data Factory and Microsoft Azure Machine Learning technology, Microsoft Azure SQL, Microsoft Azure Monitor, and Microsoft Power BI, we’ve arrived at an end-to-end solution, from data acquisition to visualization. Our first step was to collect and aggregate data across a few key data sources, then move that data through a variety of customized ML models, to arrive at a visualization output that decision-makers could use and manipulate.

Data sources

To holistically analyze and visualize how Microsoft and LinkedIn spaces were being used, we took a strategic and layered approach to the data we pulled. It was important that all data was secure—encrypted where necessary—and that we could rely on a stable and consistent flow of data. We also wanted to make sure we were looking at space and its usage from multiple angles, including general building capacity, actual functional use of spaces, and predictive fluctuations in use over time. Here’s what we gathered:

  • Badge telemetry: Badge-in swipe counts allow us to analyze building capacity patterns
  • Wi-Fi telemetry: Wi-Fi telemetry helps to validate building capacity at a given time, and match occupancy to employee device details
  • Facilities details: Floor maps, room listings, square footage, and a variety of other building details help to contextualize usage and capacity
  • Employee devices: To better understand usage, we developed an estimation model to approximate which groups are using a given space at a given time
  • Organization data: We used organization data to generally group and categorize employees into predictive teams, to deepen pattern observation
  • Office 365 data: Calendar and event data add further context to usage patterns
  • Weather predictions: We use Bing Maps to analyze daily forecasts, because fluctuations in weather often predict changes in space usage
  • Room sensor data: Conference room sensors calculate individual room occupancies to provide a more detailed view of how building capacity relates to individual space usage within that building

Machine learning models

With the power of Microsoft Azure Machine Learning, we created four ML models that serve to calculate, compute, and manipulate the data we’ve aggregated into something we can use. Because we knew it wouldn’t be enough to pull the preceding data sources and analyze them at face value, we leaned on ML technology to create an automated and adaptive computational flow.

  1. Data drop mitigation
    With hundreds of offices globally, data leakage and loss were nearly inevitable. Badge and Wi-Fi data is especially dynamic and integral to our understanding of building and space capacity. Because these datasets are co-supplemental, we created a model to cluster the data and flag any instance where Wi-Fi and badge data weren’t congruent.
  2. Device-to-person estimation
    Employee privacy was a key factor in implementing our entire solution successfully. While individual device data was essential in contextualizing building usage over time, we didn’t want to track individual employee movements to any degree. Instead, we created a model that allowed us to encrypt which device belongs to whom, while still assigning each device to a group, function, or team. This results in at least 85 percent accuracy in pairing devices to individuals with full encryption, helping us to determine the functions of business taking place in a given space at a given time.
  3. PAA
    To calculate PAA, we created an ML model to compute hourly and daily usage and identify times of peak space utilization. Using this computational model allows us to access past usage data and predict usage patterns across spaces and at given intervals.
  4. Occupancy forecasting
    At the intersection of all of our data, we can now predict how many people are expected to visit or use a building at one-hour intervals for the upcoming two-week period. This allows us to take past insights and move them forward to make predictive decisions.

Technologies

To deploy our new approach to space analysis, we used the following technology stack to aggregate our data, and then move it through computation and contextualization, delivery, and display. This flow allows for a hyper-efficient pass-through of data, from initial collection through to complete control and visualization. We relied deeply on the level of reliability and efficiency in this stack to ensure our solution would work now and serve teams going forward:

  • Microsoft Azure Databricks for data aggregation
  • Microsoft Azure Data Factory for data contextualization through ML and computation
  • Microsoft Azure SQL for feature engineering and data delivery
  • Microsoft Power BI for total data visualization and easy access to insights

Space planning architecture

This entire framework brings together a variety of data sources, technologies, and tools. Our integrative approach then draws different datasets into their corresponding ML models, depending on which level of processing or computing is necessary to achieve what we’re after. Finally, we see the outcomes of our computational work represented in a well-filtered, highly visual, dynamic dashboard using Microsoft Power BI.

Diagram depicting data sources from telemetry,  room sensor data,  device and facility data,  and more,  taken into 4 core Machine Learning models,  to arrive at a dashboard of insights.

Figure 2. Data inputs from Wi-Fi and badge telemetry, room sensors, and organizational data are aggregated and put through four key Microsoft Azure Machine Learning models to arrive at highly visual space usage insights.

Our approach to implementation

We needed the ability to tie together reliable technology, ease of access for decision-makers and users, and an integrative approach that would empower cross-functional teams to make important space decisions in real time. We prioritize data analysis methodologies that deliver highly visual data insights in a way that you can manipulate. We were careful to protect individual employee privacy and security, prioritize end-to-end data validity, and require as few adaptations of employees or visitors as possible.

Providing decision-making support for facilities managers

This solution offers innumerable benefits to facilities managers, workplace strategists, space planners, and business leaders. The impact of this solution will be felt to the degree that your team is dependent on your space. For Microsoft, our campuses hold the key to much of what we can deliver to our customers. Our buildings are the incubators for our best ideas and world-class solutions. For anyone looking to implement PAA, results will be proportionate to the effort made to use the data wisely.

The dynamic nature of this solution allows us to access both moment-in-time data and living, adaptive insights. This further equips decision-makers and analysts to manipulate data to answer pressing questions about space. Upon re-entry to standard working hours, following this extended work-from-home period, space usage is likely to have shifted dramatically. This solution accounts for fluctuations and allows for a time-inclusive study of historical space usage.

While Microsoft Azure Data Factory empowers ML computation, Microsoft Power BI delivers a malleable and deeply insightful look at how these data points interplay. The following visual dashboard allows facilities managers, real estate planners, and business leaders to make smart, informed decisions about every space on every campus in real time.

Diagram of Microsoft Power BI analytics on space usage including 180 day utilization trends,  badge-in data,  and attendance % numbers.

Figure 3. A Microsoft Power BI dashboard shows building attendance counts, utilization trends, and other filterable space usage insights in a highly visual, agile display.

New metrics for communications and clarity

By developing a roster of metrics that convey various angles of the space usage analysis problem, we’re also creating a lexicon around space measurement and an interdepartmental dialogue about the future of the workplace. PAA is only the beginning of a larger-space utilization conversation. The applications for this data are endless, as are the ways to obtain it. However, we wanted to implement with key metrics set parallel to our objectives and ideal outcomes. PAA gives all organizations a consistent way to measure and refer to space usage. This is especially important as we return to the workplace. In this way, employees have enough space to feel safe at work, and capacity overflow can be predicted and mitigated before it presents a public health or safety challenge.

Total workplace decision-making

Drawing back to our objective to empower more meaningful discussions about space utility, we needed to implement this solution with consideration for all the ways that decisions about space get made. Decisions like whether or not to extend a lease on a building, purchase more space, or engage in a Campus Refresh program like ours can now be made with data. We’re facilitating interdepartmental conversations about space by including a robust approach to visualization through Microsoft Power BI. When data is translated so everyone can access it, data analysis and insights have the power to be more profound.

Optimization for cafe management

Food waste was a big focus for this project and a top priority in our approach to implementing a solution. With insights gained from our ML solutions and the use of AI, cafe managers can purchase and prepare food, arrange tables and serving stations, and anticipate high and low waves of activity with greater precision. Specifically for LinkedIn, where meals are complimentary for employees, cost and waste-minimization are critical. With this data, we can ensure that LinkedIn’s food expense, their third largest behind payroll and real estate, is proportionate to use.

Employee productivity and satisfaction

Environment has a tremendous impact on how employees perceive and experience the organization, the work, and the workday. The task of analyzing space usage is rooted in the desire to improve workplace conditions and experiences for employees. For that reason, we centered on acquiring data that could be tied to work styles, behaviors, and movement patterns. We wanted to better understand and evaluate occupant patterns in order to build better infrastructure and reallocate spaces on behalf of the entire Microsoft and LinkedIn staff.

Best practices to successfully understand space usage

We ran into three key challenges during the implementation of this solution. From those challenges, our team developed three core best practices that empower optimal outcomes from this effort.

  1. Aim for low-cost or no-cost implementation and data acquisition
    Microsoft and LinkedIn campuses operate with badge-in swipes but no badge-out. Badge sensor data wouldn’t be enough for us to aggregate accurate capacity and usage data between buildings. We used an employee device prediction model to estimate which Wi-Fi device belongs to which employee. From there, we could identify the occupancy in a building or space at a given time. It was important to us to keep this solution easy and be able to implement it quickly.

    This is especially apt in the current landscape, where space is a major focus for nearly every company as remote employees return to the workplace. We also wanted to keep the solution cost-effective to implement with infrastructure most companies already have in place. Companies with badge-out sensors don’t necessarily need the Wi-Fi device workaround, and the predictive model was a great workaround for campuses that operate like ours.
  2. Employ Microsoft Azure Databricks to parallel-process large reams of data
    The exciting thing about data is that there’s always more to analyze. We eagerly increased the number of space usage metrics we were monitoring, expanding from 10 to 30. Unfortunately, this severely limited our computation capacity due to the complex AI network and large reams of data required to process. We used Microsoft Azure Databricks parallel processing technology to remediate this concern. Data Bricks reduced the runtime for our ML model from 8 hours per day to 40 minutes.
  3. Anticipate potential for data leakage and loss
    We track telemetry data from over 600 global Microsoft facilities. Data loss and leakage are highly possible, but AI technology helps us solve for that possibility. We developed a customized data drop mitigation framework that detects dropped data in real time from any facility worldwide. The system remediates those losses before they happen, to keep metrics reliable and consistent for multi-location measurement.

 

The future of work is fluid: Our spaces should be, too

The future of work is decentralized and, for many of us, remote. Today’s workforces are flexible and their relationship to space is multifaceted. This flux is only further impacted by the COVID-19 crisis, which could have more to say about the way we work together in the office going forward. With all of this change, we need solutions that enable us to right-size our spaces to meet agile and evolving employee needs. ML technology equips us with real data that our facilities managers can use to make decisions in real time. ML solutions bring us data that’s tied to real employee experiences and business outcomes. This fluidity is more important than ever; these analytics help us understand which campus services are operating at less-than-full capacity and identify ways to bolster a safe return to work for our employees.

Now and in the future, as we work to understand how employees move through their day, we’ll uncover patterns of performance that will help us optimize the spaces we use and the ways those spaces serve Microsoft and LinkedIn employees on a global scale.

Conclusion

The CSEO Data and Insights team, in partnership with LinkedIn’s Global Workplace Strategy team and Microsoft’s RE&S team, tapped Microsoft Azure Machine Learning technology and Microsoft Power BI to solve a complex problem: understanding and measuring how space is used in the workplace. Using existing infrastructure, the team discovered quantifiable space usage metrics, streamed data across multiple sources, and developed an ML algorithm to pull it all together. Using building, enterprise, and organizational sources, in tandem with ML, we translated reams of data into insights leaders can use to make decisions that reduce costs, improve conditions for all, and move work forward. This solution is an industry-leading example of how ML can empower teams to make data-driven decisions that lead to greater productivity and engagement in any space, smarter space sizing, and an overall reduction in waste.


You might also be interested in

Turning to DevOps engineering practices to democratize access to data at Microsoft
October 19, 2020

Turning to DevOps engineering practices to democratize access to data at Microsoft

Read blog
Microsoft automates revenue processing with Power Automate
September 22, 2020

Microsoft automates revenue processing with Power Automate

Read case study
How Microsoft applied Azure Cognitive Services to automate partner claim validation
August 06, 2020

How Microsoft applied Azure Cognitive Services to automate partner claim validation

Read Article
Transforming 70 million calls at Microsoft with Microsoft Azure
July 07, 2020

Transforming 70 million calls at Microsoft with Microsoft Azure

Read blog