How does employee monitoring software actually work?

Employee monitoring software sounds more sophisticated than it is. Nearly every product in the category is doing the same four things, and understanding them tells you quickly what the output is worth.

What does the agent measure?

A desktop agent samples a small number of signals on a timer. It does not watch continuously, and it does not understand what you are doing.

Notice what is absent: the agent has no idea whether you are writing production code or reading news. It knows a key was pressed and that the window is called something.

How are active and idle time calculated?

Time is divided into intervals, often a minute or less. An interval containing at least one input event is active; one containing none is idle. After a configurable threshold of consecutive idle intervals — typically five to ten minutes — the period is marked idle in the report.

That is the entire algorithm. It explains the category's central weakness: thinking, reading, meetings and whiteboard work all look identical to being away from the desk. Any product claiming to distinguish them from input activity alone is overselling.

In Employee Desk, idle time is always reported as its own figure rather than deducted silently or folded into the total. An employee can see it and explain it, which is the only way the number stays useful. See the time tracking documentation.

What is the activity percentage?

The ratio of active intervals to total intervals in a period, expressed as a percentage. Nothing more.

It is useful for one thing: spotting a tracker left running overnight, or a session that recorded six hours with almost no input. It is actively harmful when used to compare two people, because the number measures typing, and typing is not the job. A guide on measuring remote productivity covers what to use instead.

How is screenshot timing decided?

Good implementations randomise it inside a range you set, redrawing the gap each time, so the timing cannot be predicted or waited out. Employee Desk defaults to a random gap between 2 and 10 minutes with a hard cap of 15, and captures are off by default for a new team.

A fixed interval — exactly every ten minutes — is worse on both counts: it is predictable, so it does not achieve what it was bought for, and it produces the same volume of sensitive data anyway.

Where does the data go?

The agent writes to a local queue first, then uploads. That ordering is what makes offline operation possible: a dropped connection pauses the upload, not the recording, and the backlog goes up in order when the link returns.

Where it uploads to is the question most buyers skip. On a cloud product it reaches the vendor's infrastructure, which matters for organisations whose screenshots would contain client code or customer records.

What can this software not tell you?

• Whether the work was any good. No signal here relates to quality.
• Whether someone is productive. Hands on keyboard is not output.
• Whether someone is struggling. A person stuck on a hard problem may show high activity and no progress.
• What happened away from the machine. Meetings, calls, reading and thinking are invisible to it.
• Intent. It cannot distinguish research from browsing.

Used to answer "how many hours went into this client project", the data is reliable and genuinely useful. Used to answer "who is my best employee", it will mislead you confidently.

Related