Observation: Australia’s Zombie Pilot Problem
Zombie pilots are not an innovation failure. They are a decision failure.
There is a pattern you start to notice once you have sat through enough steering committees, portfolio reviews and “let us do it next quarter” conversations.
Pilots do not fail. They linger.
They generate activity, updates and learning decks, similar to any corporate project, long after learning has stopped compounding. A pilot continues to receive funding (to a point), remains staffed, and is regularly reported on, yet it never quite crosses the threshold into real operational commitment. The point where experimentation stops and implementation starts.
This is frequently framed as an innovation problem. But it is not.
What Australia is experiencing, across government, enterprise, and venture portfolio, is a decision failure. Specifically, a failure to force scale-or-stop decisions once a pilot has done the job it was meant to do.
That is why expressions such as “pilotitis” and “pilot purgatory” are no longer simply hallway language. They are explicitly named in Australian contexts because the pattern is persistent and recognisable.
When a pattern gets a name, it is already widespread.
Australian enterprise and technology commentary has increasingly used the term “pilot purgatory” to describe promising proofs of concept that never graduate to implementation. This does not refer to experiments that collapse or receive a timely stop decision when hypotheses are disproved through rigour. This refers to experiments that continue, as no one ever defines what “good enough to scale” actually means. Experiments that have no entry criteria, no exit criteria and no moment when continuation has to be actively re-earned.
Australia’s Digital Transformation Agency used the term “pilotitis” in the context of AI and digital government, referring to initiatives that demonstrate feasibility but stall before wider implementation and adoption.
When a central digital authority adopts this language, it signals that the issue is not about capability or creativity; it is about how decisions are structured.
The audit trail tells the same story.
Australia’s audit bodies have been documenting this failure mode for years. The Australian National Audit Office has been clear about the purpose of pilots: they exist to inform adoption decisions. Monitoring and evaluation are not box-ticking exercises; they are meant to support judgments about impact, efficiency, and whether something should scale, change, or stop.
Yet the recurring finding is that evaluation is often not designed to force a decision.
Performance reporting focuses on activity rather than effect. This means the reporting is focused on reporting outputs and the tracking of milestones. However, outcomes are not framed in a way that compels a material commitment or a clean stop.
The same theme appears again in audits of performance statements across major government entities, where reporting describes what was done rather than what was achieved in terms that would justify scaling. Or even the inverse: a quick, efficient stop decision. Both are valuable.
At the state level, the NSW Audit Office has identified similar issues, particularly the lack of effective planning to scale successful pilots. When scale is not planned for, pilots do not fail. They simply remain pilots.
How zombie pilots are created
It is this environment where zombie pilots thrive:
Enough activity to look legitimate
Enough ambiguity to avoid hard calls
Enough sunk cost to make stopping uncomfortable
As a result, effort stays high, but learning plateaus and continuation get quietly re-approved. This is not because the case is strong, but because stopping carries consequences that someone has to own up to.
But you can also usually tell a zombie pilot by how it is talked about.
Often, the update is jam-packed, full of activity, and increases over time, and the team is always engaged and ready for the next iteration of the experiment. But when someone asks the simple question -
“So are we actually committing to this?” the room goes quiet. Not because people do not care, but because no one is quite sure who is meant to answer. That pause is the tell. The pilot is not blocked by technology or talent. It is blocked by ownership.
This is not a flaw in the people involved. It is a predictable outcome of systems in which decision rights are diffuse and risk is asymmetric.
AI did not create this problem. It exposed it.
AI has become the most effective stress test of these decision systems, as pilots are being launched faster than governance frameworks can absorb them. They touch regulated data, workforce impact, reputational risk, and external examination simultaneously, and scaling them is no longer a technical decision; it is an organisational one, raising the stakes.
AI pilots often succeed on activity metrics:
models trained
vendors engaged
proofs delivered
workshops run
But they avoid outcome-based measures that would force the real question:
Are we equipped to operationalise this, or should it stop?This is exactly what the Digital Transformation Agency was pointing to. AI does not tolerate fuzzy ownership or vague decision rights. It exposes them.
Why is this flaring now
This pattern is strengthening as Australia’s innovation and funding environment has rebounded in aggregate terms. Experimentation is still encouraged. Pilots are still funded. Venture and innovation activity continues. But governance has not kept up.
As capital tightens and tolerance narrows, continuation bias increases. It is often easier, politically, operationally, and personally, to extend a pilot than to terminate it. Especially when stopping means reallocating people, admitting uncertainty, or absorbing short-term discomfort.
In the end, pilots persist, not because they are winning, but because the system makes stopping expensive.
The issue underneath
Strip away the language and the decks, and the same three unanswered questions appear again and again:
What must be true for this pilot to move to production?
Not “what did we test,” but what evidence would justify real commitment.Who owns the operational and reputational risk if it does scale?
Technical risk is easy to discuss. Organisational risk is not.Who is explicitly authorised to stop it when learning plateaus?
Not slow it. Not reshape it. Stop it.
Until those questions have owners, pilots will continue to build up without compounding value.
This is what leads to an unwelcome truth:
The blocker is rarely the pilot itself. It is the cost of being the person who turns it off.
A final observation
Australia may not necessarily have an innovation problem, but there are indications of a decision problem. And until continuation has to be actively re-earned, zombie pilots will remain a rational outcome of an irrational system.
This is not an interesting observation. It is a solvable problem, but only if decision-making and outcomes, not ongoing experimentation, become the focus.
References / Further Reading
Australian National Audit Office, Evaluation of Australian Government Pilot Programs
https://www.anao.gov.au/work/performance-audit/evaluation-of-australian-government-pilot-programsAustralian National Audit Office, Performance Statements of Major Australian Government Entities: Outcomes of the 2024–25 Audit Program.
https://www.anao.gov.au/work/performance-statements-audit/performance-statements-of-major-australian-government-entities-outcomes-of-2024-25-audit-programNSW Audit Office - Their futures matter (program evaluation and scale challenges)
https://www.audit.nsw.gov.au/our-work/reports/their-futures-matter
Digital Transformation Agency, Global leaders come together in Australia to shape the future of digital government and AI
Australia’s startup funding outlook / ecosystem commentary, SmartCompany
https://www.smartcompany.com.au/startupsmart/australia-startups-2025-funding-report/
Disclaimer
This article is based on the author’s professional experience and interpretation of publicly available information. It is provided for general informational purposes only and does not constitute advice. Any views expressed are the author’s own and do not refer to any specific organisation, program, or individual.