back to blog
RPA

What RPA Actually Solves (and Where It Breaks Down)

2026-08-15 · 2 min read

RPA vendors sell it as "automate anything a human does on a screen." That pitch is true in the narrowest technical sense and misleading in practice. After enough implementations, the pattern of what actually works becomes pretty clear.

Where RPA earns its keep

High-volume, low-variance data movement. Nightly reconciliation between an ERP and a spreadsheet-based tracker. Pulling invoice data out of a portal with no API. Re-keying form submissions into a legacy system. These are jobs with a fixed shape, run on a schedule, where "it does the same thing every time" is a feature, not a limitation.

Legacy systems with no integration surface. If a system genuinely has no API and never will (a 20-year-old desktop app, a government portal), UI automation is often the only lever you have. That's a legitimate, durable use case — not a stopgap.

Compliance-sensitive, auditable steps. Because a bot's actions are scripted, you can point to exactly what it did and why. For processes under audit (financial close, regulatory filings), that determinism is worth more than the flexibility an AI agent would offer.

Where it breaks down

Anything with real-world ambiguity. RPA bots don't handle exceptions gracefully — they handle the exceptions you anticipated and scripted for. The moment a process has "but sometimes the customer writes it differently" branching, bot maintenance cost climbs faster than the manual-labor cost it replaced.

UI-dependent bots on systems that change. Every UI redesign, every A/B test, every relocated button is a potential outage. I've seen automation teams spend more engineer-hours maintaining brittle selectors than the process saved in manual labor — the ROI math quietly goes negative and nobody notices until someone finally audits bot uptime.

Anything that should have been an API integration. RPA is frequently used as a workaround for not having engineering time to build a real integration. That's sometimes the right call under time pressure, but it accrues technical debt: a bot doing what an API call should is fragile infrastructure wearing the costume of a permanent solution.

The honest framing

RPA is a wrench, not a Swiss Army knife. It's excellent for a specific, well-understood category of repetitive, rules-based, UI-bound work — and worth deploying deliberately for that category, not as a default answer to "we need to automate this."

share

# comments

loading comments...