Skip to main content
Guides & How-To

What is RPA? Software robots, solutions and limits

RPA is software that carries out repetitive tasks according to fixed rules, for example by operating buttons and fields and moving data between systems. Here is what an RPA robot can do, what kinds of RPA solution exist, and where the limit lies.

Author: Agent360 editorial team
Published: 3 September 2026
Last updated: 19 October 2026
Reading time: 9 min read

In short: RPA (Robotic Process Automation) is software that imitates the clicks and keystrokes an employee makes when data has to be moved between systems or the same fields filled in again and again. An RPA robot follows a fixed recipe and doesn't think for itself. It suits repetitive tasks with clear rules, and not tasks where something has to be understood or judged.

What is RPA?

RPA stands for Robotic Process Automation, in Danish "automatisering af processer med softwarerobotter". Imagine an employee who logs in to three systems every morning, pulls the same figures, keys them into a spreadsheet and sends a report on. The task is dull, takes time and occasionally goes wrong because a digit is keyed in incorrectly. RPA does the same thing, but as a recipe the computer follows step by step.

RPA can operate existing systems through the buttons and fields of the user interface. An RPA solution can also combine this with direct data exchange between the systems. That can make it possible to automate a task without changing the programs. That is one of the reasons the technology has become so widespread: it can often be laid on top of an older system that otherwise can't exchange data with anything else.

RPA is not new technology. It is a well-known type of tool that many businesses have used for years for back-office tasks, meaning the administration behind customer contact. Its strength is not intelligence. Its strength is consistent execution of the programmed steps. Errors in rules, input or systems can, however, give wrong results or stop the run.

What is an RPA robot?

The word robot confuses some people. There is no physical machine. An RPA robot, also called a software robot, is a program that runs on a computer or a server and carries out tasks in other programs. You can think of it as a very disciplined digital temp, who can only do what it has been set up to do, and has to be monitored so that errors are noticed.

A robot typically consists of three parts:

  • The recipe: the steps the robot is to carry out and the rules it is to follow, for example "if the amount is above the limit, send the case for approval".
  • The access: a user account in the systems the robot works in, with the permissions the task requires, and no more.
  • The monitoring: a log of what the robot has done, and a message to a person when something goes wrong.

Monitoring has to be there from the start. A robot that stops silently can be worse than no robot, because nobody notices that the work isn't being done.

RPA solutions: the types that exist

RPA solutions can be divided up by how the robot works together with the employees.

  • Robots that help an employee. The robot runs on the employee's own computer and is started when it's needed, for example to pull information from several systems while the customer is on the phone. The person steers the process, and the robot takes the mechanical steps.
  • Robots that work on their own. The robot runs on a physical or virtual computer on a schedule or when an event occurs, for example every night or when a new file lands in a folder. Here monitoring and error messages are especially important, because nobody is watching.

AI as a supplement to both ways of operating. The robot still follows a fixed recipe, but individual steps use AI (artificial intelligence), for example to read information out of an invoice in a new format or to sort incoming emails. This widens what can be automated, but also makes the solution more complex to test and maintain.

Which type suits depends on the task, the volume and how stable the systems are. What matters is rarely the tool itself, but whether the process is described well enough for a robot to follow it.

What can RPA be used for?

RPA is best suited to tasks that are repetitive, rule-based and predictable. Some of the most common:

  • Moving data between systems. Pulling figures from one system and putting them into another, for example from a webshop to an accounting system, or between two systems with no shared integration.
  • Keying in and recording. Creating customers, bookings or orders from information that arrives in a fixed form.
  • Reconciliation and checking. Comparing two lists, for example invoices against payments, and flagging what doesn't match.
  • Reports. Collecting the same data every week or month and putting it into a fixed template.
  • Routine replies and confirmations. Sending a standard message when a particular event occurs in a system.

The common denominator is that the task can be written down as a clear recipe. A good rule of thumb: could you write a guide that would let a brand-new employee carry out the task without using their own judgement? If you can, the task is probably suitable for RPA. If the employee has to keep assessing and interpreting, you have gone beyond what RPA is good at.

The business value, and what it costs

The value of RPA lies mainly in two things: time freed up from tedious repetition, and fewer keying errors in data that is used further on. How great the value is depends on how often the task is repeated, and how costly the errors are today. So do the sums for your own task rather than relying on general figures.

On the other side of the calculation are the licence or subscription, the time to describe and build the process and, not least, maintenance. When the robot operates the user interface, it can grind to a halt when a button moves, a field changes name, or a supplier updates its system. RPA solutions therefore need ongoing care. You don't build them once and forget them.

The limits: when rules aren't enough

The robot can't read meaning out of an email written in free text, assess an unclear case or handle something it has no rule for. If an invoice arrives in a new format, or a message is worded in a new way, the robot doesn't know what to do. Anything that requires judgement or an understanding of language lies outside classic RPA.

AI can be used to interpret text and handle variation, but the results can be wrong. An AI agent can plan and carry out several steps towards a goal; human approval of important actions has to be built into the process. The difference is described in more detail in RPA vs. agentic AI. In practice it is seldom either/or: let the rule-based parts stay rule-based, and put the understanding where there is reading and judging to be done. See examples of where automation can be applied under use cases on automation.

Frequent manual intervention is a signal to investigate the cause: unclear rules, varying input, unstable systems or shortcomings in the robot. If some steps require interpretation, they can be handled by a person or supported by AI, while the rest continues to be automated with fixed rules.

RPA and personal data

Robots often work with information about customers and employees. Give the robot access to what the task requires, and no more, and make sure it is documented what it does with the information. If a supplier processes personal data on your behalf, you must have a data processing agreement. The Danish Data Protection Agency's guidance describes the basic principles.

How to get started with RPA

Start with one process, not a strategy for the whole business. Choose a task that is repeated often, has clear rules and uses stable systems. Describe it step by step, including the exceptions the employees handle today without thinking about it. Build a small first version, measure the time and errors it saves, and have one person own the robot and its maintenance. Only expand once the first one is running stably. Before you scale up, you should measure errors and the need for manual intervention, and make sure that monitoring and maintenance can keep up. Train the employees who will start the robot, handle errors and take over the task when it stops.

If you'd like help working out whether a specific task in your business suits RPA or requires AI, get in touch.

Sources

Frequently asked questions

What does RPA mean?

RPA stands for Robotic Process Automation. It means automating fixed workflows with software robots that click and type like an employee.

What is the difference between RPA and AI?

Classic RPA follows fixed rules. AI can be used to interpret text and handle variation, but the results can be wrong. The two can be combined in the same process.

Is an RPA robot a physical robot?

No. An RPA robot is a program that runs on a computer or server and works in other programs.

What can RPA be used for?

Typically for moving data between systems, keying in and recording, reconciling lists and producing fixed reports. The common denominator is tasks with clear rules that are repeated often.

How do you get started with RPA?

Choose one frequent task with clear rules and stable systems, describe it step by step, and build a small first version. Measure the effect, and only expand once it is running stably.

Which of your manual steps could disappear?

We map where data is moved by hand between your systems, and what can realistically be automated.