blog に戻る

2022年11月03日 George Gerchow and Kenneth Barry

If and how to return to the office: Data-driven decision making

Data-driven decision making

At the height of the pandemic, remote work was vital and relatively unquestioned as necessary for productivity to continue. But as variants, vaccines, and future of work conversations showed uncertainty in predicting what’s to come, we saw the conversation shift away from setting a definite date for all employees to return to offices – instead we all started talking about hybrid work.

At Sumo Logic, we offered employees hybrid work options while we tried to determine what that would look like. In a hybrid environment, how much space is needed to best service and support employees? How can you do this predictably and cost-effectively?

Like most organizations, we started with surveys asking our teams how often they planned to come into the office. As we later saw, these answers did not accurately reflect the reality of how often they came in person.

Of course, we did the only thing that made sense as Sumo Logic. We decided to drink our own champagne, and use our platform to look at the data. Here’s how we approached the data project, including some code/queries you’re welcome to copy-paste and try in your environment. Our results were surprising to us, and we’d love to hear if you find the same when you look at your data.

What questions to ask when evaluating the future of work

Before we can find the answers, we need to make sure that we’re asking the right questions. It’s not enough to simply find out how many people seats are used in an office on a given day. We wanted to track trends over time and see if there were patterns around which days were the most popular or if certain teams preferred working in the office compared to others. Also, so many people relocated during the pandemic (and we were fortunate enough to hire distributed teams), we needed to see how that impacted the geographical distribution of our employees.

Eventually, these were the core questions we sought to answer:

  • How much real estate do we need to service employees returning to an office?

  • How many people are coming into offices and which ones?

  • What is the frequency of their visits?

  • If you want to see people in the office, what’s the best day for rubbing elbows?

  • Which teams are using the office?

  • How far did people move? If we were to put together a new location, what’s a good place with a cluster of people?

It’s also important to note that our current data might not be able to answer future questions. For example, do we currently have employees who are unvaccinated and would like to return to the office if we changed our vaccination policy? Because we strive to address the future of work without the emotion and surveys that this topic usually elicits, this is yet another question that we can only address as that data becomes available.

The art of possible: How to set up your data for office usage and needs

To get these answers, we needed to start with data sources. This will differ for every company, so this approach might not be ideal for your scenario. We initially thought that we would use laptop-wifi data, tracking when people jumped online in the office to track who was onsite. However, we found that if someone simply came for a meeting or two, they might not need to get their laptop online.

Instead, our building uses access control systems powered by HRIS and access control badging. With this, we could easily track entry into the building and accurately account for presence onsite. We augmented this data by linking it with a feed from HR to determine who these people were. This feed gave us info about who they were, which team they were on, and their home address.

By building an integration connected to the API of our payroll system, we could map the distance from their home to our facility, effectively tracking their travel distance to come into the office. Initially, this was distance as the crow flies, but we’ve found ways to enhance this to show traffic data for greater accuracy around travel time and eventually commute options.

Once the data was coming into Sumo Logic, we needed to do some engineering to show a year of data. Typically, our platform is built to track and notify real-time logging and metrics with our default displays showing no more than 30 days of data. To show usage trends in the office, though, we needed to be able to look back over a year of data to see seasonality or event-based spikes.

We did this, you can do it, too. TL;DR, we’ll save our data into a Lookup Table (query #1), then recall the lookup table in the dashboard’s query. Dashboards can only show data with a time range of fewer than 32 days of data. Lookup data is persistent and always appears when searched for, especially if there aren’t fields called _timeslice or _messagetime. With that bit of trivia, this is what you do.

We’re going to assume you already have a query ready and want to show the data. We need to obfuscate the record times in the data. We added this to the end of our query

// obfuscate the time field
| _timeslice as fake_time
| fields - _timeslice

We set ours to run daily and set ‘Yesterday’ as the time range. Create a Lookup Table to house the aggregate data from your search. Save and schedule the search. In the ‘Alert Type’, we choose Save to Lookup, we selected the lookup table we created earlier and set the Save Operator Method to ‘Merge’

Each day the query will run, get the results from yesterday, and add them to the lookup table. That's the first query.

In the dashboard panel, your query should start something like this:

cat path://your/lookup/table
// un-obfuscate the time field
| take_time as _messagetime
| fields - fake_time
// Do whatever aggregating/transforming here
| transpose ...

Set the search time range to -15m. Because we are pulling data from a Lookup Table, any time range works.

Now that our data was ingested and tracked over the necessary period, the next major step was the make sure that this data could be consumed and questioned easily. The executive team was a key stakeholder in this project, and the dashboards needed to be simple enough for non-technical users to understand and ask their own questions. As such, we configured our dashboards to make it easy to search with natural language using dashboard filters.

Complete visibility for DevSecOps

Reduce downtime and move from reactive to proactive monitoring.

Sumo Logic cloud-native SaaS analytics

Build, run, and secure modern applications and cloud infrastructures.

Start free trial

George Gerchow and Kenneth Barry

CSO and SVP of IT | DevOps Engineer

View more posts by George Gerchow and Kenneth Barry.

More posts by George Gerchow and Kenneth Barry.