RPA (Robotic Process Automation) is a technology that uses software (bots) to perform repetitive tasks in computer systems, mimicking the actions a human user would take: clicking buttons, filling in fields, copying data between applications, reading emails, and generating reports. RPA does not change existing systems. It interacts with them the same way a person would, through the user interface.
The global RPA market reached 13 billion euros in 2025, according to Gartner, with annual growth of 20%. However, the same Gartner report finds that 30 to 50% of RPA projects fail or do not reach the expected ROI. The difference between success and failure almost always comes down to selecting the right processes to automate.
What RPA is and how it works
An RPA bot works like a virtual user that executes predefined sequences of actions in computer systems:
- Capture: the bot is "trained" to perform a task, typically by recording the user's actions or through visual step configuration.
- Execution: the bot executes the sequence of actions automatically, following the defined rules. It can open applications, navigate menus, fill in forms, copy data, and generate outputs.
- Simple decisions: the bot can make rule-based decisions (if the value is greater than X, follow path A; otherwise, follow path B).
- Exceptions: when it encounters an unforeseen situation, the bot can stop and notify a human, log the error, or attempt an alternative action.
| Feature | RPA | API integration |
|---|---|---|
| How it accesses systems | User interface (UI) | Programmatically via API |
| Speed | Moderate (simulates clicks) | High (direct data transfer) |
| Fragility | High (breaks if the UI changes) | Low (APIs are stable) |
| Requirement on the target system | None (uses the existing interface) | API available |
| Maintenance cost | Medium to high | Low |
| Best for | Legacy systems without an API | Modern systems with an API |
Types of RPA: attended, unattended, and hybrid
Attended RPA: bots that work alongside the user. The user starts the bot, which performs part of the task while the user does the rest. Useful for processes that require human judgment at some steps.
Unattended RPA: bots that perform tasks completely on their own, without human intervention. They run on servers, typically outside working hours or in response to events. Ideal for batch data processing.
Hybrid: a combination of the two models. The bot automates as much as possible and requests human intervention only when it encounters an exception it cannot resolve.
When RPA is the right choice
RPA is the best option when these conditions are met:
- Systems without an API: legacy applications that do not offer a programmatic interface. RPA is the only way to automate without modifying the system.
- Rule-based processes: tasks with clear, predictable steps that do not require complex judgment. Example: copying data from an email into 3 fields of a form.
- High volume and low variability: processes executed hundreds of times a day in the same format. Example: recording 500 bank transfers in the ERP.
- Temporary solution: when a migration to a system with an API is planned, but the manual process needs to be automated in the meantime.
- Prohibitive integration cost: when building an API integration is more expensive than maintaining an RPA bot, considering the volume and expected lifespan.
When to prefer alternatives to RPA
RPA is not the best choice when (see the detailed comparison):
- APIs are available: API integration is faster, more stable, and more cost-effective in the long run. If the system has an API, using RPA is an inferior solution (see system integration).
- The process requires understanding context: documents with varying layouts, emails with natural language, or decisions that depend on professional expertise. AI is better suited for these cases (see AI for documents).
- The interface changes frequently: if the target system is updated regularly (SaaS applications with frequent releases), the bot will keep breaking. Maintenance costs can outweigh the benefits.
- Volume is low: automating a process that runs 5 times a day likely does not justify the investment in RPA. The setup and maintenance cost can be higher than the cost of manual work.
- Poorly defined process: if the process has many exceptions and variations, the bot will spend more time handling exceptions than processing normal cases.
Real costs and ROI
RPA costs include components that are frequently underestimated:
- Licensing: platforms such as UiPath, Automation Anywhere, and Blue Prism charge per bot and per type (attended/unattended). Annual costs range from 5,000 to 40,000 euros per bot.
- Development: configuring a bot for an average process takes between 2 and 6 weeks. Certified RPA developers have hourly costs comparable to software developers.
- Maintenance: 15 to 30% of the initial cost per year in maintenance. Every update to the target system may require the bot to be reconfigured.
- Infrastructure: unattended bots need machines (physical or virtual) to run on. Cloud compute costs are not negligible.
- Management: someone needs to monitor the bots, manage exceptions, and coordinate with teams. An RPA "center of excellence" has staffing costs.
RPA's ROI is positive when the automated process consumes more than 2 to 3 FTE (full-time equivalents) and is stable (it does not change frequently). For smaller processes, alternatives such as API integration or automation with workflow tools are more cost-effective (see how to calculate ROI).
The future of RPA and its convergence with AI
RPA is evolving in two directions:
- Intelligent Automation: combining RPA with AI to handle processes that require document understanding, natural language, and decision-making. The bot performs the actions, the AI understands the context (see AI-based automation vs. traditional automation).
- Autonomous agents: an evolution from rigid bots to flexible agents that understand high-level goals and decide how to achieve them, adapting to variations in the process (see the future of automation).
The trend is for pure RPA (interface automation) to decline as more systems offer APIs and AI enables smarter approaches. However, for legacy systems with no alternative, RPA will remain a valid and necessary tool.
Engibots evaluates each process individually and recommends the most suitable technology: RPA when it is the best option, API integration when possible, AI when necessary. The goal is to solve the problem in the most efficient and sustainable way.