RPA
Also known as: Robotic Process Automation, software robot, bot automation, intelligent automation
Robotic Process Automation (RPA) is software that mimics how a human interacts with computer applications -- clicking buttons, reading screens, copying data between systems, and filling out forms -- to automate repetitive, rule-based tasks.
Unlike traditional integration software that connects via APIs, RPA bots interact with the user interface of existing applications without requiring those systems to be modified. Leading RPA platforms include UiPath, Automation Anywhere, and Blue Prism.
RPA, explained at three levels
At home
RPA is software that can do the boring, repetitive computer tasks that humans do -- copying information from one system to another, filling in forms, downloading reports, and sending emails based on rules. Instead of a person spending hours doing data entry, an RPA bot does it automatically, faster, and without making typos.
At work
Organizations deploy RPA bots to handle high-volume, repetitive processes: processing invoices by reading the vendor's PDF and entering values into the accounting system, onboarding new employees by creating accounts in multiple systems from a single HR form, or extracting data from spreadsheets and entering it into a different database. These tasks take humans hours; bots do them in minutes.
At home
RPA can automate tasks even when the underlying systems don't have APIs or can't be connected in traditional ways. A bot can 'see' what's on a screen, read text from a PDF or website, type into fields, click buttons, and navigate menus -- just like a human would, but much faster. This is why RPA is often described as a 'last mile' automation tool for legacy systems that haven't been modernized.
At work
RPA bots are classified as attended (running alongside a human, triggered by human action, helps speed up a task the human is doing) or unattended (running independently on a schedule or trigger, without human involvement). Financial services use unattended bots to run overnight reconciliation reports. Customer service centres use attended bots to pull customer data from multiple systems and display it on one screen while the agent is on the call.
At home
Intelligent Process Automation (IPA) combines RPA with AI capabilities: OCR for extracting data from unstructured documents (not just structured PDFs), NLP for understanding email content and routing it appropriately, and computer vision for navigating UIs that change layout dynamically. These capabilities move RPA from purely rule-based processes to semi-structured inputs that previously required human judgment.
At work
Enterprise RPA programs face several known challenges: bots break when UI layouts change (a button moving in an application update can stop a bot), creating maintenance overhead; governance requires a Centre of Excellence to manage bot lifecycle, access credentials, and performance monitoring; and RPA often automates inefficient processes rather than redesigning them -- a risk of 'automating the mess.' The most mature RPA programs combine process mining (tools that analyze event logs to identify automation candidates) with lean process redesign before building bots.
Real-world examples
At home
Fatima worked in accounts payable and spent two hours every morning manually entering invoice data from PDFs into their accounting system. Her company implemented an RPA bot that read the PDFs overnight and entered the data automatically. Her morning data entry task was eliminated, and she spent that time on exception handling and vendor relationship management instead.
At work
Jerome's insurance company processed 5,000 claims per week, each requiring data entry across three legacy systems. An RPA bot was deployed to extract claim details from the intake form, create records in all three systems, and send an acknowledgment email to the claimant. Processing time dropped from 15 minutes per claim to 90 seconds, and error rates dropped to near zero.
How it compares
| feature | RPA | API Integration | Zapier |
|---|---|---|---|
| Integration method | UI interaction (clicks, typing) | System-to-system data exchange | Trigger-action via APIs |
| Requires API | No (works on any application UI) | Yes | Yes (app must be in Zapier's library) |
| Legacy system support | Excellent (works without modifying systems) | Only if APIs exist | Limited to modern SaaS apps |
| Maintenance | Higher (breaks when UIs change) | Medium (breaks when APIs change) | Low (Zapier maintains connectors) |
| Best for | Legacy systems, complex UI workflows | Modern systems with APIs | Simple SaaS-to-SaaS automation |
RPA is the right tool when you need to automate processes that involve legacy applications with no API, or complex multi-system workflows that require UI navigation. API integration is faster, more reliable, and lower maintenance when the target systems have modern APIs. Zapier is the simplest option for connecting modern SaaS tools without technical resources.
For analysts
Before recommending RPA, complete a process mining or time-and-motion study to confirm the process is genuinely a good automation candidate: high volume, rule-based, stable UI, and low exception rate. Automating a process with frequent exceptions or changing UI generates high bot maintenance cost that erodes ROI. RPA works best when the process is boring precisely because it's simple and repetitive.
History & origin
RPA emerged in the early 2000s from screen scraping and test automation software that could interact with application UIs. The term 'Robotic Process Automation' was popularized by Blue Prism around 2012. UiPath and Automation Anywhere followed, and the category grew rapidly as enterprises discovered they could automate back-office work without replacing legacy systems. The RPA market grew from nearly nothing to over $3 billion in 2023. Gartner consistently named RPA the fastest-growing enterprise software category through the late 2010s. RPA platforms expanded from attended automation (bots that assist humans in real time) to unattended automation (bots that run overnight without human involvement), and more recently added AI/ML capabilities to handle less structured inputs.