AI Glossary / Agents and automation / Beginner
Robotic process automation
Definition
Robotic process automation, or RPA, is software that performs repetitive, rule-based digital tasks by mimicking how a person uses applications: clicking, typing, copying and moving data between systems, usually at high volume and without supervision.
What it means
Despite the name, no robots are involved. RPA is software that drives the same screens your staff use, following recorded or scripted steps. It logs in, moves through screens, reads fields, enters data and moves files, twenty-four hours a day without tiring.
RPA's sweet spot is the structured, stable, boring: tasks with fixed rules, digital inputs and high volume. Copying confirmed figures between systems, processing standard forms, generating routine correspondence.
Its boundary is just as clear. When inputs vary in meaning rather than format, or judgment is required, plain RPA fails. That gap is what AI-enhanced variants, often called intelligent automation, were built to fill.
The business case is arithmetic: hours of re-keying removed, error rates on routine entries cut, staff moved to exception handling. The honest ledger also includes building, testing and maintaining the bots, which break when the systems they drive change.
For a business owner, RPA is usually the first serious automation purchase, and the discipline it teaches, stable inputs, documented rules, monitored runs, is the foundation every later AI project will need.
How it works
An RPA bot is a script of steps against application interfaces: open, read field, apply rule, enter result, confirm, log. Modern platforms use element identifiers rather than screen positions, queue work items, and run many bots in parallel.
Rules are explicit: if the invoice total matches the order, post it; if not, route it to the exception queue. Everything the bot cannot resolve by rule goes to a person, which is why exception design is half the project.
Control rooms manage the fleet: scheduling runs, watching success rates, stopping bots when systems change, and keeping the audit trail that finance and compliance teams require.
Adding AI creates the common hybrid: a document-reading or classification step feeds structured data to the bot, which then does what bots do, the reliable mechanical part.
Three practical examples
- Finance: match supplier invoices to purchase orders overnight and post the clean matches, queuing mismatches for morning review.
- HR: create accounts, access rights and payroll entries for new joiners from a single approved form.
- Banking: copy customer updates across the several systems that must agree, with a reconciliation report as proof.
Think of it this way
Think of a tireless clerical worker who does exactly what the procedure says, perfectly, at enormous speed, and is completely stumped the moment the procedure does not cover the case.
The limit: even that worker would ask someone when a form looks wrong. A bot needs the exception rules written in, or it processes the wrong thing with the same perfect confidence as the right thing.
Worked example
automating supplier statement reconciliation
Suppose a firm receives monthly statements from forty suppliers, and two staff spend a week matching them to the ledger.
An RPA flow logs into the supplier portals, downloads statements, matches line items by amount and reference, and produces a differences list. The week becomes an afternoon of reviewing genuine discrepancies. The figures are invented; the match-by-rule-plus-exception pattern is standard.
The rules were written from six months of resolved reconciliations, so the bots encode what the team actually did, not what the procedure document claimed.
When one supplier changes its statement format, the bot fails loudly on that supplier only, the run log shows exactly where, and the other thirty-nine still process.
Illustrative case study
A fictional insurer rolls out RPA across claims entry and celebrates a seventy percent time saving. Then a core system upgrade moves every screen element, and the whole bot fleet fails on the same Monday.
Recovery takes a fortnight because nobody owned bot maintenance after go-live. The rebuild assigns an owner, schedules post-upgrade retests and adds element-based selectors that survive most layout changes.
The invented lesson: RPA shifts cost from doing the work to maintaining the bots. Budget for the second, or the first saving evaporates.
Common misconceptions and mistakes
- RPA is AI. It follows explicit rules; adding AI components makes it intelligent automation, but plain RPA is deterministic scripting.
- It works on any process. Unstable interfaces, varied documents and judgment calls all defeat plain RPA.
- Bots are a one-off project. Every target-system change is a maintenance event, so bots need owners and retest schedules.
Business use
Strong candidates: high-volume, rule-stable, digital processes such as reconciliations, data migration, form processing and multi-system updates.
Document the real process with its exceptions before building, choose targets with stable underlying systems, and assign each bot a named owner with a maintenance budget from day one.
People also ask
Does RPA need developers?
Modern platforms are low-code, but exception design, testing and maintenance still need structured thinking, whoever does it.
What is the difference from intelligent process automation?
IPA adds AI for varied inputs like documents and emails; RPA handles the structured, rule-based core.
How do bots fail?
When the systems they drive change, or when an input falls outside the rules, which is why exception queues and monitoring are essential.
Check your understanding
A firm's whole bot fleet fails after one core-system upgrade. What two practices would have limited the damage?
Answer: A named bot owner with scheduled post-upgrade retests, and element-based selectors that tolerate layout changes instead of raw screen positions.
Related terms
Learn next
AI Agents and Automation for Small Business builds rule-based workflows with exceptions and approvals: https://aiforbusinessandlife.com/academy/ai-agents-automation-small-business. [C4]
More in Agents and automation
- Problem-solving agent - A problem-solving agent is an agent that works toward a goal by searching: it imagines sequences of actions, evaluates where they lead, and chooses a...
- Proto-agent - A proto-agent is one of two semantic proto-roles from linguistics: the participant in an event who acts, causes, moves or experiences, as opposed to the...
- Rational agent - A rational agent is an agent that, given what it has perceived, chooses the action expected to achieve the best outcome according to its performance measure.
- Router agent - A router agent is an agent whose job is to decide where work goes: it reads an incoming request and directs it to the right specialist agent, tool, model or...
- Simple reflex agent - A simple reflex agent is the most basic kind of agent: it chooses its action from the current percept alone, using condition-action rules, with no memory of...
- State machine agent - A state machine agent is an agent built as an explicit set of states and transitions: it is always in one known state, such as waiting, collecting details...
Sources
[S3] NIST, AI glossary (AIRC). Full URLs are in the source ledger. Verified 11 October 2026.
Previous: Robotic governance Next: Robotics