The Automation Graveyard: Post-Mortem of 5 Real RPA Failures

Here's a number that should make every operations leader uncomfortable: up to 50% of initial RPA projects fail outright, according to Ernst & Young.

Those are just the loud failures. The quiet ones are worse — bots that "work" but cost more to maintain than the humans they replaced, pilots that never scale beyond 10 bots, automation programs that consume 18 months and $1.6M only to become full-time maintenance jobs.

We analysed dozens of post-mortems and case studies to identify the five failure modes that keep appearing. If any sound familiar, your program might already be in the graveyard.

Failure #1: Automating the Mess

A European telecom deployed RPA across finance to automate invoice processing and bank reconciliations. The bots mirrored exactly how operators navigated disconnected ERP systems — replicating every inefficiency at machine speed. When exceptions arose, there was no process logic to handle them because nobody had documented the actual workflow.

The lesson: RPA doesn't fix processes — it fossilises them. Before automating, ask: "Could a new hire follow this process in writing without asking for help?" If not, you need process redesign, not automation.

Failure #2: The Department of 50 Fragile Bots

A multinational bank gave each department its own bots, developers, and priorities. Within 18 months: 50 bots, zero coordination. When the ERP pushed a quarterly update, twelve bots broke simultaneously. Some developers had left. Others had built workarounds so idiosyncratic that nobody could debug them.

The lesson: Governance isn't bureaucracy — it's insurance. Every bot needs an owner, documentation, and a maintenance schedule. If your estate can't answer "who owns this bot?" for every deployment, you have a graveyard in waiting. As the agentic AI vs RPA comparison shows, this governance gap only widens with more autonomous systems.

Failure #3: The Set-It-and-Forget-It Myth

A logistics company deployed bots for sales orders and invoice processing. Initial results were excellent. Management declared victory and budgeted zero for maintenance.

Then supplier portals redesigned their interfaces. SAP pushed patches. Login flows changed. Each update broke bots. HfS Research estimates licensing is only 25–30% of RPA's total cost of ownership — the remaining 70–75% goes to maintenance, emergency fixes, and developer premiums. For a 50-bot deployment, that's roughly $680,000 in maintenance over three years.

The lesson: Budget at least 20–30% of implementation cost annually for maintenance. If your business case only works with zero maintenance cost, it's fiction.

Failure #4: The Coordinate-Based Time Bomb

A US accounting firm spent 15 months building an RPA bot for audit workflows. It interacted with tax software through its GUI — clicking coordinates, typing into fields. It worked perfectly until the software pushed an interface update and buttons moved.

This is RPA's core architectural weakness: bots follow coordinates, not intent. For a mid-sized enterprise running 15 interconnected systems that update regularly, that's 60 potential breaking points per year. Forrester estimates maintenance can account for up to 60% of total RPA expenses.

The lesson: If your vendor's answer to "what happens when the ERP updates?" is "you rebuild the bot" — you're buying a time bomb with a quarterly fuse. The shift toward AI agents is partly driven by this brittleness.

Failure #5: Nobody Asked the Humans

At the same telecom, employees withheld process information from bot developers, fearing job losses. Some actively obstructed automation efforts. At the accounting firm, a working bot saw low adoption because participation was voluntary and nobody championed it internally.

39% of organisations cite lack of internal expertise as the #1 barrier to RPA success (Deloitte). Not technology. Not budget. People.

The lesson: Your RPA program needs a change management workstream budgeted like the technology workstream — role redesign, training, and clear messaging about what happens to the humans whose tasks are automated.

Self-Diagnosis Checklist

Before your program joins the graveyard, score yourself honestly:

Process Health — Every target process documented end-to-end. Exception handling defined in writing. Process stable for 6+ months. You've asked: "Should this process exist at all?"

Governance — Every bot has an assigned owner. Central register of all bots with last-tested dates. Approval process for new deployments. Documentation standards enforced.

Financial Reality — Business case includes 20–30% annual maintenance. Emergency remediation budgeted. TCO calculated over 3 years. You know the difference between license cost and total cost.

Change Management — Affected employees informed. Training planned for post-automation workflows. Executive sponsor actively championing. Employee futures defined (upskilled, redeployed — not "TBD").

Architecture — Plan for underlying system updates. Bot resilience tested against UI changes. Fallback procedures in place. You've evaluated whether you need deterministic RPA or adaptive AI agents.

The Bigger Picture

The graveyard isn't going to stop growing. As enterprises layer agentic AI onto already-fragile RPA foundations, Gartner predicts 40% of agentic AI projects will be abandoned by 2027.

The companies that succeed won't have the most bots. They'll have treated automation as a design discipline — understanding processes before automating them, governing bots as production systems, budgeting honestly, and remembering that the humans in the loop aren't obstacles. They're the reason automation exists.

Frequently Asked Questions

Why do so many RPA projects fail?

The most common root cause isn't technology — it's automating processes that were never well-defined. Ernst & Young reports 30–50% of initial RPA projects fail outright. Companies automate broken workflows, skip governance, budget zero for maintenance, and neglect change management. The failures are organizational, not technical.

What is the real total cost of ownership for RPA?

Software licensing represents only 25–30% of RPA's total cost. The remaining 70–75% goes to implementation, maintenance, emergency fixes, and developer premiums. For a typical 50-bot deployment, that translates to roughly $680,000 in maintenance over three years — costs often missing from the original business case.

Should companies switch from RPA to AI agents?

Not necessarily. AI agents solve some RPA limitations — like navigating by intent rather than screen coordinates — but they introduce new risks. Gartner predicts 40% of agentic AI projects will be abandoned by 2027. The answer depends on your use case: deterministic, rule-based tasks still suit RPA, while adaptive, multi-system workflows may benefit from AI agents.

Sources: Ernst & Young, HfS Research, Forrester, Deloitte, Gartner, Accounting Horizons / Zhang et al., CIO Tech Outlook.