RPA · Robotic Process Automation
Robotic Process Automation
What is RPA?
- Software robots take over repetitive office work
- Mimics “hand movements” on existing systems
- Strong at simple, rule-based tasks
What RPA really changes is not the content of the work but where that work sits. Modifying a core system takes six months from requirements definition to acceptance testing. RPA is layered on top of existing screens, so a short task can be off people’s hands in as little as 2 to 3 weeks. For a department that has been stuck in the IT development queue for years and getting by manually, the first time it changes its own work directly is the real change people feel on the ground.
Bots fall into two broad types. Attended bots are launched by a person who presses a button and watches alongside. They suit work like customer support and service desks, where case-by-case judgment is mixed in. Unattended bots run on their own at set times on a server and are used for overnight closing or large-scale data migration. In practice, teams usually apply three criteria when choosing candidate tasks: 200+ repetitive hours per year, processing rules that are at least 90% identical with no branching, and input in a machine-readable form such as a screen or Excel. If any one of the three is missing, API integration or a system improvement is a cheaper choice than RPA.
Look at the cost structure first. Bot licenses are billed annually per bot, and automating a single task usually takes 2 to 5 days of development and testing. In other words, automating a task that takes 2 hours a month will never pay for itself. Steps that require reading scanned documents or handwriting fall under intelligent automation, where OCR and AI are added on top. However high the recognition rate, the design must include a review step where people catch the few percent that are wrong.
The most common misconception is that a bot, once built, keeps running forever. Bots rely on screen positions and button names, so when the groupware is patched or one extra popup appears on an ERP lookup screen, that night's close stops cold. 'Ghost bots' whose owner has left and whose source code exists only on a personal PC, and bot accounts given full administrator rights for convenience, will always be flagged in an audit. Automation with no designated operations owner will have half its bots broken within six months. If you justify adoption as headcount reduction, business units will never give up their work.
Seolyeong ParkGenerative AI and AX training expert with 17 years of hands-on experience and 200+ lectures, who gets non-specialists using AI right away
John KimJohn Kim, CEO of Nanum Consulting and author of “B2B Marketing Design”It emerged in the 2010s alongside advances in business automation software and has since spread as a practical tool for digital transformation.
Consider a hypothetical case: Company A, a mid-sized auto parts maker with 320 employees and annual revenue of around KRW 180 billion (≈ US$134,670,100). Each month, its four-person finance team downloaded about 1,200 supplier tax invoices from Hometax, organized them in Excel, and keyed them into the ERP one by one. During closing week the whole team stayed until 10 p.m., and typos caused three or four payment errors every quarter.
- A finance team senior manager collected two weeks of work logs and tallied invoice processing time: 160 hours a month went to data entry and reconciliation alone, and half of that was simple typing to copy values from one place to another.
- Manager Lee of the information strategy team split the work into four stages: download, organize, enter, and verify. Senior Manager Kim drew the line: "We’ll handle account classification and rejected items. Just leave the transfer work to the bot."
- One unattended bot was deployed to download Hometax data at 5 a.m. every day and enter it into the ERP. In the first week it stopped three times because of ERP notice pop-ups, so a step to close pop-ups was added.
- Bot account permissions were limited to journal entry input only. Two people on the information strategy team were assigned to fix the bot whenever a screen changes, and an alert email goes out at 7 a.m. whenever an error occurs.
- Three months later, data entry work dropped from 160 hours a month to 22 hours, month-end close moved up from D+5 to D+2, and two finance team members shifted to receivables management and supplier price analysis.