Robotic process automation (RPA) is a type of software that executes rule-based tasks by replaying the actions a person performs on application screens: clicking, copying a value from one system into another. The bot follows a script; it does not interpret what it reads.
In finance, RPA links systems with no interface: bots download statements from bank portals when no MT940 or camt.053 feed is in place, or key invoices into an ERP. An API integration does the same job with fewer moving parts.
A script checks only what its author specified, leaving three blind spots:
- Content: the bot copies an amount whether or not it is right, so a duplicate invoice goes through at machine speed
- Change: a renamed field or a new invoice layout stops the bot or makes it read the wrong zone
- Identity: a bot account that can both create suppliers and release payments carries a segregation of duties conflict into every run
The line with AI has a legal edge. Recital 12 of the EU AI Act says the definition of an AI system should not cover systems "based on the rules defined solely by natural persons to automatically execute operations", because a key characteristic of AI systems is the capability to infer. A purely scripted bot generally falls outside it; adding a model that reads invoices can change the analysis under EU AI Act compliance.
RPA differs from cognitive automation, which adds document understanding to rule execution, and from an AI agent, which pursues a goal and handles unscripted cases.
Phacet starts with content. The agent that automates the accounting inbox routes supplier emails, detects invoices and extracts attachments, its workflows blend LLM and deterministic rules, and ambiguous results go to a person with the source document and a confidence score.