Pragmatic CenterPragmatic CenterTechnology studio
Home/Journal/Case study
Case study · 28 October 2026

Working time analytics without manual reports

A manager needs to understand what the team's working day goes on, but we did not want to make people fill in timesheets. We built a system that puts the picture together by itself from the work computers: which programs and tasks take up time, and how the workload changes from day to day. Here is how it is built and why we built it this way.

The question “where does the time go?” comes up in any company where work is done at a computer. How many hours a week go on the accounting system, how many on email and messaging, how many on the browser and work services. The answer is needed not to count minutes but to see the workload: where processes eat up time, which routine tasks are worth automating, how the load changes from day to day.

The usual way to get such an answer is timesheets. But manual tracking works badly precisely where it is needed most: people fill it in from memory at the end of the day, round figures off, forget the switches between tasks. And the very need to account for every hour takes up time and is irritating.

The task: see the workload without putting extra load on people

The requirements were formulated simply. The picture should assemble itself, without employees' involvement. The data should stay inside the company. Getting an answer should be as easy as asking a colleague a question. And installing on a new computer should not turn into a separate project.

How it works

The system consists of two parts: a small program on the work computers and the company's server, where the data is collected.

The program on the computer runs in the background and records which window is active and when the person is actually working at the computer. An icon in the notification area shows that the system is switched on. Every few minutes the data is sent over a secure connection to the company's server. If the connection drops, nothing is lost: the program sends what it has accumulated once the server is available again.

On the server, each computer's data is stored separately, by day. Two tools work on top of it: a web interface with a timeline by application, and an analytics module that calculates totals — time by program, a breakdown by day, a comparison of computers with one another.

Why it is built this way

Automatic collection instead of timesheets. Manual tracking shows what people remember about their day, not how it actually went. Automatic collection does not take a single minute of employees' time and does not depend on memory. We deliberately did not add any forms or reminders: the aim was to remove reporting, not to move it somewhere else.

Honest time counting. A window can stay open for hours while the person is in a meeting or at lunch. So the system distinguishes between two measures: the time a program was in the foreground, and active time, when it was actually being worked with. If nothing happens at the computer for a minute, that interval is not counted towards active time. A locked screen and being away from the desk drop out of the count by themselves. A window left open overnight does not turn into hours of work either.

Data on the company's own server. Information about the working day is internal information, and there was no reason to hand it over to an external cloud service. So the server is deployed on the company's own infrastructure, and the computers send data only there.

An AI assistant on top of the data. Timelines and tables are good, but a manager usually needs an answer to a specific question, not a report in which it has to be found. So an AI assistant is connected to the analytics: you can ask “how much time went on 1C this week?” or “how did the workload change by day in October?” and get an answer with figures straight away. The calculation is done on the company's server; the assistant requests ready-made totals through a separate connector that only reads the data and changes nothing in it.

Installation with one command. The company has several computers, and installing the program on each one by hand, with keys and settings, is inconvenient and error-prone. So the server itself provides the installer: on a new computer it is enough to run one command, and the program downloads, configures itself for the company's server, enables autostart and starts working. Nothing needs to be entered by hand.

What was built

  • A program for work computers on Windows (there is also a macOS version): it runs in the background, starts with the system, and survives reboots and dropped connections.
  • A server for receiving and storing data: each computer's data is kept separately, and data is accepted only over a secure connection and only from the company's own computers.
  • A web interface with a choice of computer and a timeline coloured by application.
  • An analytics module: time by program or window over a period, a breakdown by day with the start and end of the working day and the busiest program, a comparison of computers, a minute-by-minute chronology of the day.
  • A connector for the AI assistant with a separate password login.
  • An installer that the server provides itself, and daily backups.

Practical value

The main thing the system provides is an objective picture at no cost of collecting it. You can see which programs really take up the working day and how the workload changes by day and by week. This is a good basis for a conversation about processes: if a noticeable share of time goes on manual operations in the accounting system, that is a reason to automate them; if the workload is steadily growing, it becomes visible before deadlines start being missed.

The assistant removes the last barrier: to get a figure, there is no need to work through reports and filters. The question is asked in plain words, and the answer comes back with specific hours.

Transparency and careful handling of data

A subject like this cannot be treated carelessly, so the rules are built into the system itself rather than left to discretion.

  • Employees know about the system. It was not introduced in secret, and the program's icon is visible on every computer.
  • Data stays inside the company. The raw data is stored only on the company's servers. Only ready-made totals, such as hours by program for a week, go to the AI assistant on request.
  • Limited retention period. Data older than 90 days is deleted from the server automatically; in addition, there is a limit on the total volume.
  • Access is protected. The web interface is password-protected, and the assistant's connector has separate authorisation. The analytics only reads the data and cannot change it.
  • Sensitive information is excluded. Password managers are not tracked, and password input fields are not saved.

The system was conceived as a tool for understanding workload, not for assessing individual people. Its purpose is to help make decisions about processes.

Outlook

The architecture allows the system to be developed without reworking its foundation. New computers are connected with one command, and the analytics module can be extended with new views: for example, grouping programs by type of work or comparing the workload of different periods. Since the answers are built on the same data, each new view immediately becomes available to the assistant as well.

For us this project is an example of how to approach a delicate task: first determine what really needs to be known and which rules must be followed, and only then design the technology around them.

Let's talk

Already have a task in mind?

A website, an online store or automation — tell us about it and we will say honestly how we can help.

The Pragmatic Center journal

  • 01How strong websites and stores are built.
  • 02Why speed matters more than it seems.
  • 03What automation really gives a business.