RPA for business: where it still fits, and where AI replaced it
Short answer: RPA for business still fits one situation well: getting into a system that offers no API, no database access and no file exchange. Everywhere else it has been overtaken, because RPA drives a user interface and breaks the week that interface changes. The work it could never do, anything requiring interpretation, is exactly what moved to AI-led automation.
What RPA actually does
Robotic process automation scripts the clicks a person would make. It opens the application, types into the field, reads the screen, moves to the next system. It has no understanding of what any of it means.
That is its strength and its ceiling in one sentence. The strength is that it works against anything a human can operate, including a thirty-year-old terminal application nobody has the source for. The ceiling is that it is coupled to a presentation layer that was never designed to be a contract.
Why RPA programmes plateau
Three reasons, and they compound.
Brittleness. A screen change, a slower load, an unexpected dialogue, and the bot fails. Maintenance grows with the number of bots until it consumes the team that built them.
No judgement. Anything requiring interpretation, reading the attached invoice, deciding whether two records are the same customer, judging whether a request is unusual, stayed with a person. So the queue never disappeared, it just arrived faster.
Process debt. Automating a bad process makes it a fast bad process. Programmes that skipped process redesign got exactly what they automated.
The result is familiar: an impressive pilot, a plateau well short of the target, and a maintenance burden nobody budgeted.
Where RPA for business still wins
Systems with no other route in. Legacy applications, terminal emulators, vendor software with no API. This is the genuine case and it is not going away.
Short-lived bridges. A migration is eighteen months out and you need the process working now. A bot is the right amount of investment for a temporary problem.
Regulated screen-level parity. Occasionally a control requires that the automated path is provably the same as the human path.
Use it for those, and keep the judgement outside it.
What replaced the rest
Integration at the API or data layer, with a model at the steps that need interpretation.
That combination changes which processes qualify rather than making the old ones faster. A claims process that was 40% automatable because most of it required reading unstructured documents may now be 85% automatable. That is a different business case, not a faster version of the old one.
It also changes the failure mode. Rules fail loudly and predictably. Models fail plausibly, producing a confident wrong answer that reads exactly like a right one. That is why our business process automation services put confidence scoring and an explicit escalation path on every judgement step, and keep everything else deterministic.
The hybrid worth running
Not a replacement, a division of labour.
Deterministic workflow code owns the process and the state. A model handles the steps that need interpretation, with a confidence score and an escalation path. RPA handles the single legacy hop where nothing else can reach, and nothing else.
That last point matters. Teams that use RPA as the backbone inherit its brittleness across the whole process. Teams that use it as one connector contain the risk to one step.
What to measure
Straight-through rate, the share of items completing without a person. Then exception volume by reason code, which turns your queue into a prioritised list of what to fix next. Then bot maintenance hours, which is the number that tells you whether an RPA estate is still paying for itself.
If maintenance hours are growing faster than volume, the estate has plateaued and the answer is redesign rather than more bots.
The takeaway
RPA for business is a connector, not a strategy. Keep it for the systems that offer no other way in, move the process logic into code you own, and put a model only where judgement is genuinely required. Measure maintenance hours alongside savings, because that ratio is what decides whether the programme is working.
EpochC builds business process automation services that integrate at the API and data layer, with intelligent document processing services where documents drive the loop. See the KYC OCR automation case study or start a project.