Connecting your software: click robot (RPA) or interface (API)?
When is a software robot that mimics clicks enough, and when do you need an interface? The difference between RPA and API automation – with an example, maintenance effort and a decision guide.
Moving data from one program to another by hand, every day: it costs time and invites mistakes. If you want to automate that, there are two fundamentally different ways:
- RPA (robotic process automation): a software robot operates the programs the way a human would – through the user interface. It clicks, types, copies.
- API automation: the systems talk to each other directly through their interfaces (APIs). No user interface, no clicks.
Both have their place. But they're not equally good.
The comparison
| RPA | API automation | |
|---|---|---|
| How it works | mimics clicks and inputs | systems exchange data directly |
| Requirement | the program has a user interface | the program has an interface (API) |
| Speed | as fast as the interface loads | seconds, even with many records |
| Stability | sensitive to every change in the interface | stable – APIs change rarely and with notice |
| Operation | needs a machine or virtual machine | runs in the cloud or on a server |
| Typical tools | Power Automate Desktop and other RPA platforms | n8n, Make, Power Automate cloud, custom integrations |
When RPA is the right choice
RPA is a bridge. It pays off when there's no other way:
- Legacy systems without an interface: industry software, older ERP systems, programs running via remote desktop.
- Web portals without an API: such as government or supplier portals where data is regularly entered by hand.
- Interim solutions: the system is being replaced within a year anyway, and until then the manual work should disappear.
When API automation is the right choice
Whenever an interface exists – and today that's the case for most cloud tools: CRM, email, calendar, accounting, online shop, project management.
API automation is faster, more stable and cheaper to run. It can run around the clock, react immediately to new events (for example via webhooks) and handle large volumes of data. Which tool fits is covered in n8n or Make?
The typical trap
Many businesses start with RPA because it feels like simply "rebuilding the human". That works well in the demo. In daily use, the limits show:
- An update changes the interface – the robot can't find the button anymore.
- A pop-up appears that nobody expected.
- The machine running the robot installs updates or gets locked.
That doesn't mean RPA is bad. But using RPA where an interface would be available builds in unnecessary maintenance.
A typical example
A business receives orders through a supplier's portal every day and copies them by hand into its own inventory system. The supplier portal has no interface, the inventory system does.
Here the combination makes sense: the inventory system is connected through its interface. Only the step in the portal that can't be solved any other way runs via RPA. That keeps the fragile part small, and when the portal changes you only have to adjust that one step. (This is a typical scenario, not a specific client project.)
What it costs in the long run
A robot that operates user interfaces is quick to build but needs ongoing care: any change to the program can break it. An interface-based solution often costs about the same to set up but runs with much less maintenance afterwards. So don't only count the build, count running it for a year or two.
What about AI?
AI models can now "see" and operate screens themselves. Anthropic and OpenAI offer such computer-use capabilities. That makes automation through the user interface more flexible, because the AI can cope with small changes.
For recurring processes with fixed rules, the interface is still the better way: faster, cheaper, more traceable. AI plays to its strengths where content needs to be read and classified – such as extracting data from an invoice or sorting enquiries.
How to decide
- Is there an interface? Then API automation.
- There isn't, but the system is being replaced soon? Then RPA as an interim solution.
- There isn't, and the system is staying? Then RPA – but well documented, with error alerts and a clear owner.
- Most of the time: a combination. Everything through interfaces, only the one step without an interface via RPA. What that looks like in the Microsoft world is shown in Power Automate cloud vs. desktop.
Which path fits you depends on the process and the systems involved. That's exactly what I look at first in the automation pilot – before anything gets built.
Concrete workflows where this pays off: Automating client onboarding and Creating client reports automatically.
