System of Action

Most programs in companies collect information and display it. A customer management program lists who bought what and when. A human reads this list, decides, and types in the next action themselves. A System of Action reverses this relationship: the software carries out the work step itself instead of merely preparing it. It writes the email, creates the order, or cancels the booking – and the human, at best, only checks the result. The term comes from the software industry and has become popular in recent years through AI programs that can perform tasks independently.

From reference work to employee

To understand the meaning, it helps to look at the older classification. Software in companies was long sorted into two groups. A System of Record is the reliable data repository, i.e. the source of truth: personnel files, invoices, inventory levels. A System of Engagement is the interface through which people work with it, such as a chat program or a dashboard. Neither type acts on its own.

This is exactly where the new term comes in. When software takes over tasks, the value of the programs shifts. A company then no longer pays for access per employee, but for completed processes. Some providers already charge per resolved support ticket instead of per user per month. For investors this is an important distinction, because it changes the entire business model of a software company.

However, the term is also a marketing word. Many providers slap it onto products that, at their core, still only make suggestions. A useful test is the question of whether a human ultimately has to click. If they do, it isn’t a real System of Action.

Tools, permissions, and reversal

Technically, such a system needs three things. First, a model or a set of rules that decides what needs to be done. Second, interfaces to other programs, so-called APIs – these are defined access points through which software can control other software. Third, permissions: the system must actually be authorized to change a booking or transfer money.

In AI-based systems, an agent takes on this role. This refers to a program that breaks a goal down into individual steps and works through them one after another. It checks the result after each step and adjusts the next one accordingly. If a step fails, it tries a different approach.

The trickiest part is not the acting but the undoing. An incorrectly displayed number is annoying, but an incorrectly triggered transfer costs money. That’s why such systems work with limits: amounts above a certain threshold require human approval. In addition, every step is logged so that it can later be traced who initiated what.

Where such systems are already at work

In customer service, the shift is furthest along. In the past, a chatbot would suggest help articles. Today it can create a return, track the shipment, and refund the money. The difference for the customer is significant: the problem is solved by the end of the conversation, not merely described.

The approach can also be found in programming. Tools don’t just suggest code, they modify files, run tests, and submit the change for review. Something similar happens in accounting, where software matches invoices and initiates payments.

The term regularly appears in news about companies like Salesforce, ServiceNow, or Microsoft. There it usually signals: in the future, we will sell outcomes, not screen interfaces. Whether this works out depends on how reliably the systems act. A program that does the wrong thing in five percent of cases is unpleasant in customer service and unacceptable in accounting.