Robotic Process Automation: Where It Fits and Where It Does Not
Robotic process automation (RPA) bots operate software the way a person does. That makes them useful for systems that cannot be connected any other way, and fragile almost everywhere else.

Robotic process automation (RPA) sounds like physical robots, but it is software that operates other software the way a person does: clicking buttons, copying values, filling forms and downloading files. It can be very useful. It is also frequently misused, because it gets applied to problems that have better and more durable solutions.
What RPA actually does
An RPA bot follows a recorded or scripted sequence of steps on a computer screen. It logs into a system, opens a screen, reads or types values, moves to another application and continues. Tools in this space include UiPath, Automation Anywhere and the desktop flows in Microsoft Power Automate, among others.
The key point is that RPA works through the user interface. It does not need the target system to offer an integration or an API (application programming interface, a published way for systems to exchange data). That is its strength, because it can automate systems that cannot be connected any other way. It is also its weakness, because user interfaces change, and when they do, the bot breaks.
A bot also inherits the speed and limits of the screens it uses. If a portal takes ten seconds to load each page, the bot waits ten seconds too. It can work through the night and does not get bored, but it is not a substitute for a direct connection between systems when one is available.
Where RPA fits and where it does not
Good fits
- Legacy systems with no integration options. An old system that cannot export data or accept imports, but that people use through its screens every day.
- External portals. Government, bank, insurer or supplier websites where staff download statements, upload documents or check statuses, and no API is offered to you.
- High-volume, rule-based copying. Hundreds of similar transactions moved from one screen to another, with clear rules and few exceptions.
- A temporary bridge. Automating a process during the months before a planned system replacement, when building a proper integration would be wasted effort.
Poor fits
- A direct connection is available. If two systems can exchange data directly, an integration will be more dependable, and cheaper to maintain, than a bot clicking through both.
- Unstable processes. If the steps change often or exceptions are common, the bot will fail often and someone will spend their days fixing it.
- Work that needs judgement. Decisions that depend on reading context, negotiating or interpreting unclear documents need a person, possibly supported by other tools.
- Wasteful processes. Automating six unnecessary steps produces a fast bot doing unnecessary work. Fix the process first.
- No clear owner. Bots need monitoring and maintenance. Without an owner, they become a silent source of errors.
Comparing the options
| Approach | How it works | Strength | Main risk |
|---|---|---|---|
| Native system feature | Built-in automation or import in the system itself | Supported and upgraded by the vendor | May be limited in flexibility |
| API integration | Systems exchange data directly through defined interfaces | Dependable at high volume | Needs technical build, and an API must exist |
| Workflow tool | A cloud tool connects apps through ready-made connectors | Quick to build for common apps | Complex logic becomes hard to manage |
| RPA bot | Software operates screens like a user | Works with systems that cannot be connected | Breaks when screens change; needs maintenance |
In practice, these approaches are often combined. A workflow tool might receive a trigger, call an API for most of the work, and hand one step to a bot because a single legacy screen has no other way in. That is a sensible design, as long as the bot does only the part nothing else can.
A scenario: insurance claims at a clinic group
Imagine a clinic group in Amman that submits claims to several insurers. Some insurers accept electronic files. Others require staff to log into a web portal and enter each claim by hand, copying details from the clinic’s billing system. Two staff members spend most of their week doing this, and typing errors lead to rejected claims.
For insurers that accept file uploads, the right answer is an export from the billing system in the required format, not a bot. For the portals that only allow manual entry, RPA is a reasonable choice: a bot reads approved claims from the billing system, enters them into the portal and saves the submission reference back against the claim. Staff handle the exceptions the bot flags, such as missing approval numbers.
The clinic group also accepts that when an insurer redesigns its portal, the bot will need updating. It assigns an owner and a support arrangement from day one, and keeps the manual procedure documented so staff can step in if the bot is down.
Questions to ask before building a bot
- Is there a native feature, import option or API that could do this instead?
- Is the process stable, documented and mostly free of exceptions?
- How often do the screens involved change, and who will know when they do?
- Whose credentials will the bot use, and do the terms of the external portal allow automated access?
- Who monitors the bot daily, and who fixes it when it fails?
- What happens to the work if the bot stops for a week?
If these questions have clear answers, RPA can save real effort. If they do not, the bot is likely to become a maintenance burden that someone inherits without warning.
Choosing the right automation, not the most visible one
RPA earns its place when systems cannot be connected any other way and the process is stable. Used everywhere else, it adds fragility. Our Intelligent Automation work starts by understanding the process and the systems around it, then chooses the simplest durable approach, in line with the KEMLAT Method view that automation should follow understanding, not lead it. Sometimes that is a bot. More often it is a better process and a proper integration.



