RPA in finance: what it solved, where it broke, and what replaced it
RPA in finance automates screen-level actions — logging in, clicking, copying between systems — and was adopted because it required no changes to the underlying systems. That same strength is its weakness: because a bot depends on the screen rather than the data, any interface change breaks it, and large RPA estates accumulate maintenance costs that can exceed the labour they replaced. It remains genuinely useful for stable, high-volume, integration-free tasks, and has largely been superseded by APIs where those exist.
Why finance adopted it so fast
RPA arrived with an unusually attractive proposition for finance: automation that required no involvement from IT, no change to the ERP, and no integration project. A bot logged into the same systems a person did and performed the same clicks.
For finance functions sitting on systems they could not modify, and having watched integration projects overrun for years, this was compelling. Adoption was fastest exactly where systems were oldest.
The work it took over was real: downloading statements, re-keying between a billing system and the ledger, running the same report set every month, moving data into a consolidation template. High-volume, entirely mechanical, and genuinely tedious.
The maintenance problem
A bot that automates a screen has a dependency on that screen. A vendor moves a button, adds a confirmation dialog, or changes a login flow, and the bot does not adapt — it fails, or worse, it half-succeeds.
For one bot this is a minor irritation. For a portfolio of eighty, accumulated over three years by different people, it becomes a maintenance function nobody budgeted for. Several large finance functions found their RPA estate required permanent staffing to keep running, which is a strange outcome for a labour-saving technology.
The second problem was subtler. Because RPA required no process change, it was frequently used to automate bad processes rather than fix them — encoding a workaround permanently instead of removing it.
The lesson generalises beyond RPA: automation that requires no process change also creates no pressure to improve the process, and a fast bad process is still a bad process.
Where RPA is still the right answer
The technology is not obsolete. It is correctly scoped now, which it was not in 2018.
| Characteristic | Good fit | Poor fit |
|---|---|---|
| System interface | Stable, rarely updated | Frequently redesigned SaaS |
| API available | No API exists | API exists — use it instead |
| Volume | High and repetitive | Occasional |
| Exceptions | Rare and well-defined | Common or ambiguous |
| Process maturity | Settled and documented | Under active change |
| Lifespan | Needed for years | Interim workaround |
What came next
Two things changed the picture. First, APIs became common. Where a system exposes its data directly, moving that data is a solved problem and no bot needs to pretend to be a user.
Second, the hard part of finance work turned out not to be the clicking. Getting the numbers out was always the visible cost; deciding what they meant was the expensive one. RPA never touched that, because it had no notion of what any figure represented.
The current approach separates the two. Data movement through integration where possible. Calculation in deterministic code, so it can be checked. Interpretation as a distinct layer on top, working from figures it did not produce. RPA occupies a smaller, more defensible position within that — the fallback for systems that still refuse to talk.
Common questions
What is RPA in finance?
Robotic process automation uses software bots that operate applications the way a person does — logging in, navigating screens, copying data between systems. In finance it is typically applied to statement downloads, re-keying between systems, and running recurring reports.
Is RPA still used in finance?
Yes, but more narrowly than during its peak. It remains the practical option for high-volume work against stable systems that expose no API. Where an API exists, direct integration is more robust and cheaper to maintain.
What is the difference between RPA and AI in finance?
RPA performs predefined actions and has no understanding of the data it moves; it does the same thing every time regardless of what the numbers say. AI is used for interpretation — reading results and explaining them. They solve different halves of the problem and are frequently combined.
Why did RPA projects fail in finance?
Chiefly maintenance. Bots break when interfaces change, and large estates accumulated a support burden that offset the savings. A secondary cause was automating flawed processes unchanged, because RPA deliberately required no process redesign.