Introducing AI in your company — but how? What an AI office does when nobody is there

When people talk about AI in the office, it is almost always about the moment somebody asks a question and gets an answer. That is the visible part of an AI-run back office. It is the smallest.

The larger part happens when nobody is looking. It has no screen, no input field and no demo. It has a timetable.

This Sunday edition walks through one day in the AI-run back office — with the times that are in the system of our own operation.

After midnight

The server updates the lists it uses to lock out attackers and synchronises the certificates that secure its connections. Both once a day, both without anyone lifting a finger.

Between three and five

Now runs what would disturb during the day. The databases are backed up and transferred, together with the files, to a separate storage. Seven days, four weeks and six months are kept.

Every agent fetches the current state of its rules — so that tomorrow’s session does not work by yesterday’s rules.

And the website is rebuilt. What is planned for today appears. After that a machine opens every page and every internal link, more than 700 link targets, and checks whether everything answers.

All day long, in rhythm

This is the part that surprises most when you count it for the first time.

• Every two minutes the server checks whether its services respond.
• Every five minutes the mailboxes hosted with other providers are mirrored — so that what arrives there is in the own inbox.
• Every five minutes an agent checks whether a mail has arrived for a mandate that needs a reply. It prepares a draft. It cannot send.
• Every ten minutes a check runs whether the website is up to date.
• Every fifteen minutes calendars and contacts are synchronised.

In between the pre-sorting runs without a break: incoming mails are read, classified and filed; scanned documents get their folder.

There are 38 schedules of our own on our main server. The four most frequent alone — mailboxes, services, calendars, website — add up, by our calculation, to more than 6,000 scheduled runs a day. No human triggers any of them.

At a quarter past seven

The press review is ready: what appeared overnight on the company’s topics, as one mail. In the late afternoon the second edition follows.

In the morning

The scheduled posts appear. Later in the morning a machine looks at the pages where they have to be and speaks up if one is missing. In the evening a second one counts whether everything is planned for the coming days.

And the human?

In this whole timetable a human appears in exactly three places.

They approve. No mail goes out and no post appears without somebody having said yes.

They are called. The phone rings when a backup fails, a service does not respond, a link leads nowhere or a post is missing. On a good day it never rings.

They change the timetable. Each of the 38 schedules is there because a task was added. The next one will again be entered by a human.

The most important line

More than 6,000 runs a day only run quietly if every failure reaches somebody. That is why the most important line in the whole system is not a schedule but the rule that a failure reaches the phone.

And that is why every new schedule comes with the same question: who finds out when it fails?

What this means for your company

When you think about AI in the office, do not picture the chat. Picture the list: what has to happen in your company every night, every hour, every five minutes — and who notices when it does not happen once?

Most small companies do not have this list. They have a person who thinks of it. That works until they are on holiday.

An operated back office is at its core exactly this list, written down, monitored and assigned to somebody who answers for it. The AI in it sorts, reads and drafts. The timetable makes sure it does so every day.

The list grows with every task that is added — it is never finished, and that is its purpose.

What happens in your office tonight without anybody having to think of it?