Australian organisations have spent the past two years buying artificial intelligence at a sprint. Copilots, chatbots, drafting assistants and analytics engines have landed in finance teams, marketing departments and customer service floors, often faster than anyone stopped to ask who now owns the output. According to a new opinion piece in Forbes Australia, that rush is quietly building a fresh kind of liability on the corporate balance sheet, one that does not show up in any accounting standard: deployment debt.
The argument comes from Lucio Ribeiro, a digital and AI strategist who has spent years watching companies mistake the act of switching a tool on for the harder work of embedding it. His central point is deceptively simple. Capability is now cheap and abundant, but the surrounding scaffolding that makes capability safe and useful, namely clear ownership, defined workflow and genuine accountability, is being skipped. The result is a growing gap between what an organisation can technically do and what it can actually stand behind.
When four systems all sound confident
The phrase that gives the piece its title captures the problem neatly. In a modern enterprise, a single question can now be answered by four different AI-assisted systems, each pulling from slightly different data, each returning a slightly different number, and each sounding equally authoritative. Nobody in the room is formally responsible for reconciling them, and nobody has been told which version is the one the business will act on. The technology has produced abundance, but abundance without an owner is just noise wearing a suit.
This is where deployment debt differs from the more familiar idea of technical debt. Technical debt is a shortcut in code that a future engineer eventually has to repay. Deployment debt sits in the organisation itself: in the undefined handoffs, the unassigned decisions and the workflows that were never redesigned to account for a machine now doing part of the job. It accrues silently because, on any given day, everything appears to be working. The chatbot answers, the report generates, the summary lands in the inbox. The cost only becomes visible when something goes wrong and there is no one who can say, with confidence, that they own the answer.
Ribeiro reaches for a stark real-world illustration to make the stakes tangible. He points to the phenomenon of ambulance ramping in South Australia, where paramedics keep patients in vehicles parked outside hospitals because there is nowhere inside to move them. The hours involved have ballooned in recent years, and his purpose in raising it is not to blame any single hospital but to show what happens when a system has capability, in this case beds, staff and vehicles, yet lacks the flow and ownership to convert that capability into a resolved outcome. Patients are monitored, everyone is technically doing their job, and still the queue grows. It is a picture of capability stranded without workflow, and it maps uncomfortably onto what is now happening with AI inside offices.
Two ways to read the warning
There are competing ways to interpret this. The more alarmed reading is that many organisations have effectively pre-loaded a governance crisis. Every AI tool bolted onto an existing process without redesigning that process is a small future liability, and the interest compounds as more tools arrive. On this view, the boards now proudly reporting how many AI pilots they are running are measuring the wrong thing entirely. Volume of deployment tells you nothing about whether the deployments are owned, and an unowned system is a risk dressed up as progress.
The more measured reading is that deployment debt is a normal and manageable feature of any fast technology cycle, not a looming catastrophe. Enterprises took on debt when they moved to the cloud, when they adopted software as a service, and when they first digitised paper processes. Each wave produced a period where capability outran control, and each was eventually brought to heel through governance, clearer roles and better tooling. From this angle the lesson is not to slow down but to pay the debt down deliberately: assign owners, redesign the workflow around the tool rather than beside it, and decide in advance which version of the truth the business will treat as canonical.
Both readings, notably, arrive at the same practical prescription. Someone has to own the output. The disagreement is only about how much danger sits in the gap before that ownership is assigned.
Why this lands hard in Australia
For Australian businesses the timing is pointed. Local adoption has accelerated sharply, yet the country’s regulatory settings for AI remain a work in progress, with the federal government still shaping a framework covering jobs, education and small business. That leaves accountability resting heavily on individual organisations at exactly the moment they are deploying fastest. In sectors that carry legal weight, such as financial services, healthcare and law, an unowned AI answer is not merely inconvenient. It can become a compliance failure, a privacy breach or, in a courtroom, a liability with a name attached.
The concern is sharpened by recent local research suggesting Australians are inclined to trust AI outputs a little too readily, including in high-stakes settings such as banking. If frontline staff and customers both assume the machine’s answer is authoritative, and no human inside the organisation has been made responsible for verifying it, deployment debt stops being an abstract management concept and becomes an operational hazard. The four versions of the truth problem is worse, not better, in a culture primed to believe the first confident answer it is given.
There is also a competitive dimension. Australian firms are small by global standards and cannot outspend the hyperscalers on raw AI capability. What they can do is deploy it more thoughtfully, wiring tools into disciplined workflows with clear lines of accountability. Handled well, that discipline becomes an advantage rather than a brake, because a business that can trust its own AI outputs moves faster and with less rework than one constantly reconciling contradictory numbers.
What comes next
The practical next step Ribeiro presses for is unglamorous but decisive: treat deployment as a design problem rather than an installation task. That means naming an owner for every AI-touched process, rebuilding the workflow around the tool, and establishing a single agreed source of truth before the fourth conflicting answer appears. For Australian boards, the more useful metric is not how many AI systems have been switched on but how many have someone genuinely accountable for what they produce. Until that question has a clear answer, the debt keeps quietly accruing, and the bill, as with all debt, eventually comes due.
Sources: Forbes Australia.
