They go into this in the paper; it sounds like they have a quite granular level of detail here:
> The data also include information on hours worked, our primary input measure. This is measured in a sophisticated way, as the analytics software takes into account whether an employee actually engages in a relevant task (which counts as work time) or merely procrastinates at their desk (not counted), by monitoring which software tools the employee uses. Our key outcome measure is Productivity, output divided by hours worked. Thus, in contrast to studies of productivity during WFH based on surveys, our outcome variables are based on relatively objective analytics and monitoring data.
> Moreover, our data include (for a subset of employees) how time was allocated to various activities. That includes meetings, collaboration, and time focused on performing work without distractions. It also includes information on networking activities (contacts) with colleagues inside and outside the firm. Finally, we have data on employee characteristics such as age, experience, tenure at the company, gender, whether or not there is a child in the home, and an estimate of commute time during WFO
This sounds overall quite invasive, so at least from the description, they claim to be able to tell the difference between "an hour logged into slack doing household tasks" and "an hour working with 100% focus on a task".