Robotic Process Automation

Robotic process automation (RPA) is a category of enterprise software in which software robots, usually just called bots, operate other applications through their graphical user interfaces, reproducing the clicks, keystrokes, and copy-paste work that a human clerk would otherwise perform ^[1][2]. Despite the name, RPA has nothing to do with physical robots. The "robot" is a script running on a desktop or server that logs into applications, reads fields from one system, and types them into another, following rules written by a developer ^[2].

RPA is also distinct from artificial intelligence in the classic sense. A traditional RPA bot executes a predefined workflow on structured data and makes no predictions; academic work on the field describes it as an "outside-in" approach to automation, in which existing information systems stay unchanged and the bot simply replaces the human sitting in front of them ^[1]. That property made RPA cheap to deploy against legacy systems that lack APIs, and it turned the technology into one of the fastest-growing enterprise software segments of the late 2010s. Gartner measured the worldwide RPA software market at $3.2 billion in 2023, up from $2.6 billion in 2022 ^[13]. Since 2024 the category has been reshaped by agentic AI: every major RPA vendor now sells LLM-based agents alongside its rule-based bots, while Anthropic, OpenAI, and Microsoft have shipped "computer use" agents that overlap directly with what RPA does ^{[16][17][19][20]}.

Origins

The mechanical ancestors of RPA are much older than the term. TechTarget traces the lineage back to macro recorders and to the customized data-scraping tools of the 1980s and 1990s, which pulled data out of applications' screens so it could be reused in other programs ^[2]. This screen scraping heritage, a close cousin of web scraping, is still visible in modern RPA products: reading data out of one interface and keying it into another remains the core use case.

The companies that defined the category were founded years before anyone called it RPA. Blue Prism, long headquartered in Warrington, England, was founded in 2001 by David Moss and Alastair Bathgate and shipped its first commercial product in 2003 ^[4]. Automation Anywhere began in San Jose, California in 2003 under the name Tethys Solutions and took its current name in 2010 ^[5]. UiPath started in Bucharest, Romania in 2005 as DeskOver, founded by Daniel Dines and Marius Tirca; it renamed itself UiPath in 2015 ^[6].

The label "robotic process automation" is generally dated to 2012, though its coinage is contested. SS&C Blue Prism states that the company coined the term that year ^[3], and its former marketing chief Pat Geary is often credited as the individual author ^[25], while TechTarget attributes the term to Phil Fersht, founder of the analyst firm HFS Research, also in 2012 ^[2].

However the name arose, it stuck, and the money followed. Blue Prism listed on London's AIM market in March 2016 ^[4]. Automation Anywhere raised a $550 million Series A in 2018 and a $290 million Series B led by Salesforce Ventures in November 2019, the latter at a $6.8 billion post-money valuation ^[5][7]. UiPath priced its initial public offering at $56 per share and began trading on the New York Stock Exchange under the ticker PATH on April 21, 2021, raising about $1.3 billion in the base offering ^[8].

How it works

An RPA deployment has three standard parts: a design tool in which developers record or assemble the steps of a process, the bots that execute those steps, and an orchestrator that schedules bots, distributes work, and logs what they did ^[2]. Bots identify what to act on through selectors, which are structural descriptions of interface elements (a button's position in an application's object tree, for example), falling back on coordinates or image matching when nothing better is available ^[2]. Working through the interface this way is slower than integrating systems through an application programming interface, but it requires no change to the underlying systems, which is the point ^[1][2].

The industry splits bots into two operating modes. Attended bots run on an employee's own machine and are triggered by that person, typically to speed up a task in the middle of other work, such as pulling up a customer's records across several systems during a call. Unattended bots run on schedules or queues on servers or virtual machines with no human present, processing batches of transactions end to end ^[2].

Wil van der Aalst and colleagues, in an influential 2018 overview in Business and Information Systems Engineering, positioned RPA as the economic middle ground of automation. Processes with huge volumes justify traditional software engineering: rebuilding systems and integrating them properly. Rare, judgment-heavy work stays with people. In between sits what they call the "long tail of work": repetitive tasks whose volumes are too small to justify reengineering the underlying systems but large enough to be tedious and error-prone when done by hand. RPA targets exactly that band, because the existing systems remain unchanged ^[1]. Typical examples are moving new-hire data between HR, payroll, and learning systems, reconciling invoices, or copying claim details from email attachments into a claims platform ^[1][2].

Vendors and market

Scroll sideways for more →

Microsoft's entry changed the economics of the market. Power Automate added RPA through its UI flows feature, announced at Ignite 2019 and generally available on April 2, 2020 ^[9][26]. In May 2020 Microsoft bought Softomotive, maker of the WinAutomation desktop automation tool, to strengthen the desktop side of the product ^[10], and at Ignite in March 2021 it made Power Automate Desktop available to Windows 10 users at no additional cost ^[11]. Bundling a basic RPA tool into Windows put a floor under pricing for standalone vendors.

Consolidation followed at the top of the market as well. SS&C Technologies, a financial services software firm, completed its acquisition of Blue Prism Group plc in March 2022 for approximately $1.6 billion (£1.25 billion), after a bidding contest with Vista Equity Partners; at the time Blue Prism reported more than 2,000 customers in over 170 countries ^[4][12].

On market size, figures vary widely between research firms and many published numbers are forecasts rather than measurements. Gartner's market share analysis, an after-the-fact measurement, put worldwide RPA software revenue at $2.6 billion in 2022 and $3.2 billion in 2023 ^[13]. RPA remains a small fraction of enterprise software spending; its prominence came from growth rates and from how visible the savings were in back-office headcount.

Limitations

RPA's weaknesses are as well documented as its benefits, and most of them follow from the same design choice that makes it cheap: bots interact with interfaces built for people.

The best-known problem is fragility. A bot that locates a button through a selector or a screen position breaks when the application is updated, a field is renamed, or a layout shifts, and it cannot repair itself. Microsoft's own automation team, introducing computer use in Copilot Studio in 2025, cited the fragility of UI elements as the traditional limitation, and marketed the new tool on the promise that when buttons or screens change it keeps working ^[19]. Van der Aalst and colleagues raised the same issue in 2018 from the other direction, warning that RPA agents risk making incorrect decisions when the context changes in ways their rules never anticipated, and arguing that the technology would need machine learning to become robust ^[1].

Fragility produces maintenance debt. Each deployed bot is a small program tied to the exact state of one or more user interfaces, so a portfolio of hundreds of bots requires continual technical support to keep pace with system changes ^[2]. Organizations that automated aggressively found they had, in effect, built a new legacy estate on top of the old one.

Scaling proved harder than the pilot projects suggested. Many programs stalled after automating the first handful of processes; a 2019 Gartner assessment found that only about 13 percent of enterprises had managed to scale their early RPA initiatives, and Gartner's coining of "hyperautomation" that year was in part a response to RPA's scaling limits ^[14]. The pattern repeated often enough to become an industry cliche: dramatic savings on the first process, then a lengthening backlog of processes that were too variable, too exception-laden, or too poorly documented to automate with rules.

Finally, RPA automates tasks, not processes. Because the bot reproduces what a clerk did, it can entrench an inefficient workflow instead of prompting anyone to redesign it, and critics have long argued that it postpones rather than replaces proper system integration ^[1].

Convergence with AI

From roughly 2019, the vendors' answer to these limits was to attach AI to the rule-based core, sold under labels such as intelligent automation and hyperautomation. Gartner coined hyperautomation in 2019 to describe the combination of RPA with process mining, machine learning, optical character recognition, and low-code development tools, aimed at automating more of the enterprise than RPA alone could reach ^[14].

The acquisitions of that period map the strategy. UiPath bought ProcessGold, a Dutch process mining company, in October 2019, so that customers could discover which processes to automate from system logs, and Re:infer, a London natural language processing startup, in 2022 to read unstructured messages ^[6]. Document understanding, the extraction of fields from invoices and forms with OCR and computer vision, became a standard platform module across vendors, since documents are where rule-based automation most often ran out of road.

Generative AI accelerated the shift. At its Imagine 2024 conference, Automation Anywhere made AI Agent Studio generally available, a low-code tool for building AI agents grounded in enterprise data and governed by guardrails, and repositioned its platform around combining such agents with its existing automation ^[15].

The agentic AI era

Between 2024 and 2026 the boundary between RPA and AI agents largely dissolved, from both directions at once.

From the model side, foundation model developers shipped agents that operate computers directly, which is RPA's home turf. Anthropic announced a computer use capability for Claude 3.5 Sonnet on October 22, 2024, the first frontier model offered in public beta that could look at a screen, move a cursor, click, and type; it scored 14.9 percent on the screenshot-only OSWorld benchmark, against 7.8 percent for the next best system ^[16]. OpenAI followed on January 23, 2025 with Operator, a research preview for ChatGPT Pro users in the United States built on a Computer-Using Agent model, which reached 38.1 percent on OSWorld and 58.1 percent on WebArena ^[17][18]. OpenAI retired Operator on August 31, 2025 after folding browser operation into the broader ChatGPT agent ^[18]. Microsoft brought the same idea into its business automation stack: Copilot Studio gained a computer use capability in April 2025, explicitly marketed as UI automation that keeps working when buttons or screens change rather than breaking like selector-based RPA, and the feature reached public preview that September ^[19][22]. These systems are less predictable than scripted bots, and their benchmark scores show they still fail at a large share of computer tasks, but they handle interface changes and unstructured input in ways rule-based bots cannot ^[16][19].

From the RPA side, the incumbents rebuilt their platforms around agents. UiPath launched what it called the first enterprise-grade platform for agentic automation on April 30, 2025, combining Maestro, an orchestration layer that coordinates AI agents, RPA robots, and people in a single workflow, with an Agent Builder in UiPath Studio and a UI Agent for natural language driven interface navigation ^[20]. CEO Daniel Dines described the aim as a platform "that unifies AI, RPA, and human decision making" ^[20]. The company has since extended the line, adding native integration for coding agents in May 2026 and Maestro Case, an orchestration product for exception-heavy case work, in June 2026; its current self-description is a company building systems where "AI agents reason, robots act, and people lead" ^[21][23]. Automation Anywhere followed a parallel path with multi-agent orchestration and, in May 2026, a set of platform updates for running what it calls AI-driven processes across the enterprise ^[15][24].

The emerging consensus position, visible in all of these product lines, is that deterministic bots and probabilistic agents coexist rather than one replacing the other: scripted automation for stable, auditable, high-volume steps, and LLM-based agent orchestration for the judgment calls, exceptions, and unstructured inputs that rule-based RPA was never able to absorb ^[19][20][21]. Whether the RPA vendors or the model developers end up owning that combined market is one of the open competitive questions in enterprise software.

See also

• Agentic AI
• AI agents
• Anthropic computer use
• OpenAI Operator
• Agent orchestration
• Web scraping

References

1. ^van der Aalst, W.M.P., Bichler, M., Heinzl, A. "Robotic Process Automation." Business and Information Systems Engineering, vol. 60, pp. 269-272, 2018. link.springer.com/...s12599-018-0542-4
2. ^TechTarget. "What is robotic process automation (RPA)?" techtarget.com/...RPA
3. ^SS&C Blue Prism. "About Us" (states the company coined the term robotic process automation in 2012). blueprism.com/about
4. ^Wikipedia. "Blue Prism." en.wikipedia.org/...Blue_Prism
5. ^Wikipedia. "Automation Anywhere." en.wikipedia.org/...Automation_Anywhere
6. ^Wikipedia. "UiPath." en.wikipedia.org/...UiPath
7. ^Automation Anywhere. "Automation Anywhere Announces $290 Million in Series B Financing." November 2019. automationanywhere.com/...on-in-series-b-financing
8. ^UiPath. "UiPath Announces Pricing of Initial Public Offering." April 2021. uipath.com/...uipath-announces-ipo-pricing
9. ^Microsoft Power Automate blog. "Enable end-to-end enterprise automation with Power Automate UI flows." November 6, 2019. microsoft.com/...tion-with-power-automate-ui-flows
10. ^Microsoft Power Automate blog. "Microsoft acquires Softomotive to expand low-code robotic process automation capabilities in Microsoft Power Automate." May 19, 2020. microsoft.com/...ities-in-microsoft-power-automate
11. ^Microsoft. "Microsoft Ignite March 2021 Book of News" (Power Automate Desktop at no additional cost for Windows 10 users). news.microsoft.com/ignite-march-2021-book-of-news
12. ^PR Newswire. "SS&C Completes Acquisition of Blue Prism Group plc." March 16, 2022. prnewswire.com/...f-blue-prism-group-plc-301503933
13. ^Gartner. "Market Share Analysis: Robotic Process Automation, Worldwide, 2023." July 24, 2024. gartner.com/...5614191
14. ^TechTarget. "What is hyperautomation?" techtarget.com/...hyperautomation
15. ^Automation Anywhere. "Here's Every Product Innovation Announced at Imagine 2024 (and Even a Few That Weren't!)." 2024. automationanywhere.com/...magine-2024-and-even-few
16. ^Anthropic. "Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku." October 22, 2024. anthropic.com/...3-5-models-and-computer-use
17. ^OpenAI. "Introducing Operator." January 23, 2025. openai.com/...introducing-operator
18. ^Wikipedia. "OpenAI Operator." en.wikipedia.org/...OpenAI_Operator
19. ^Microsoft Copilot blog. "Announcing new computer use in Microsoft Copilot Studio for UI automation." April 15, 2025. microsoft.com/...soft-copilot-studio-ui-automation
20. ^UiPath. "UiPath Launches the First Enterprise-Grade Platform for Agentic Automation." April 30, 2025. uipath.com/...rade-platform-for-agentic-automation
21. ^UiPath. "UiPath Introduces Maestro Case to Orchestrate Dynamic, Exception-Heavy Business Processes Across the Enterprise." June 16, 2026. uipath.com/...-case-for-dynamic-business-processes
22. ^Microsoft Copilot blog. "Computer use is now in public preview in Microsoft Copilot Studio." September 15, 2025. microsoft.com/...eview-in-microsoft-copilot-studio
23. ^UiPath. "UiPath Becomes First Business Orchestration & Automation Platform with Native Integration for Coding Agents." May 12, 2026. uipath.com/...uipath-for-coding-agents-launch
24. ^Automation Anywhere. "Automation Anywhere Unveils 2026 Platform Enhancements to Run AI-Driven Processes Across the Enterprise." May 19, 2026. automationanywhere.com/...-run-ai-driven-processes
25. ^TechTarget contributor profile. "Pat Geary" (states Geary invented the term robotic process automation in 2012 at Blue Prism). techtarget.com/...Pat-Geary
26. ^Microsoft Power Automate blog. "Microsoft Power Automate adds robotic process automation in April." March 2, 2020. microsoft.com/...botic-process-automation-in-april