Point of View
Artificial Intelligence · Healthcare · Risk Management
The Automation Question Every Organization Should Be Asking -- and Almost None Of Them Are
More than half of all work hours can theoretically be automated. That's the beginning of the strategic question, not the answer. In industries where the cost of a wrong decision is catastrophic and irreversible, the real value of AI is not efficiency -- it's error reduction.
Jan 2025
Points of View is Vaxa's opinion column. Each article states a position on a question where reasonable people disagree — and makes the argument for that position directly.
VAXA POINT OF VIEW
Automating half of all work hours is available to every organization simultaneously — which means it creates no competitive advantage. The organizations that win will not be the ones with the highest automation rates. They will be the ones that automated the right work. And the most important work to automate is not the cheapest or the most repetitive. It is the work where the accumulation of human error is quietly producing catastrophic outcomes that no one is measuring correctly.
Frame
The analysis has been made by multiple research institutions and is now broadly accepted: when cognitive AI agents and physical AI robotics are combined, the theoretical automation potential of current work exceeds fifty percent of all work hours. The finding is striking. It has generated significant attention. And it has prompted a predictable response from most organizations: a renewed focus on where AI can reduce headcount, compress timelines, and cut operating costs.
That response is understandable. It is also strategically incomplete — and in some cases, strategically counterproductive.
The problem with framing AI's potential primarily around automation is that automation, at this scale and on this timeline, will be available to every organization in every industry simultaneously. When every competitor has access to the same automation potential, no individual organization gains a durable competitive advantage from deploying it. The cost base across the industry compresses. Margins normalize at the new lower cost level. The organizations that moved first capture a temporary efficiency advantage. The organizations that moved last catch up. The competitive landscape looks roughly the same as before — except leaner.
That is not a bad outcome. Lower costs and higher efficiency are genuinely valuable. But they are not a growth strategy. They are a survival condition. The organizations treating AI automation primarily as a cost reduction exercise are optimizing the wrong variable — and missing the more important strategic question entirely.
The more important question is not how much can be automated. It is which work should be automated, and why — and what becomes possible when the human capability freed by automation is redeployed to work that AI cannot do. That question has a different answer in every industry. Getting it right is where the real strategic advantage lives.
Errors
The analysis has been made by multiple research institutions and is now broadly accepted: when cognitive AI agents and physical AI robotics are combined, the theoretical automation potential of current work exceeds fifty percent of all work hours. The finding is striking. It has generated significant attention. And it has prompted a predictable response from most organizations: a renewed focus on where AI can reduce headcount, compress timelines, and cut operating costs.
That response is understandable. It is also strategically incomplete — and in some cases, strategically counterproductive.
The problem with framing AI's potential primarily around automation is that automation, at this scale and on this timeline, will be available to every organization in every industry simultaneously. When every competitor has access to the same automation potential, no individual organization gains a durable competitive advantage from deploying it. The cost base across the industry compresses. Margins normalize at the new lower cost level. The organizations that moved first capture a temporary efficiency advantage. The organizations that moved last catch up. The competitive landscape looks roughly the same as before — except leaner.
That is not a bad outcome. Lower costs and higher efficiency are genuinely valuable. But they are not a growth strategy. They are a survival condition. The organizations treating AI automation primarily as a cost reduction exercise are optimizing the wrong variable — and missing the more important strategic question entirely.
The more important question is not how much can be automated. It is which work should be automated, and why — and what becomes possible when the human capability freed by automation is redeployed to work that AI cannot do. That question has a different answer in every industry. Getting it right is where the real strategic advantage lives.
Automating the cheap work is straightforward. Automating the work where human error is producing irreversible outcomes — and where AI can fundamentally change the quality of human judgment — is a different and far more consequential challenge.
Evidence
Emergency medicine is the most time-compressed, highest-stakes decision environment in any industry. A practitioner in an emergency room is making diagnostic and treatment decisions in minutes -- sometimes seconds -- drawing on their own training, their own clinical experience, and whatever colleagues they can consult in real time. The quality of that decision is directly constrained by the limits of individual human experience and the impossibility of recalling, under time pressure, every relevant case from medical literature and clinical records.
An AI system with access to the full accumulated record of clinical presentations, diagnoses, treatment interventions, and outcomes changes that decision environment fundamentally -- not by replacing the clinician's judgment, but by giving that judgment access to an informational base no individual human could accumulate in a lifetime of practice. The value is not that the AI makes the diagnosis. It is that the AI surfaces the patterns in an enormous body of prior cases that the clinician can then weigh against what they are seeing in front of them, with the full benefit of their clinical intuition and real-time observation. The error reduction potential -- in misdiagnoses, missed diagnoses, treatment selection errors, and drug interaction oversights -- runs into hundreds of billions of dollars annually in avoided complications and preventable harm in the US healthcare system alone.
The same structural logic applies well beyond medicine, in any domain where institutions historically underinvest in preparedness because the cost of the disaster that did not happen never appears in any accounting system -- disaster response, structural engineering, legal reasoning, agricultural risk. In every case, an AI system that mines the full accumulated record of prior outcomes changes the quality of the decision being made under pressure, without replacing the judgment making it.
Implication
Across every high-stakes decision environment, the same pattern holds: AI augments human judgment by giving it access to an informational base no individual could accumulate or synthesize in the time available. The quality of the decision improves because the information it is based on improves -- the human judgment itself remains the irreplaceable element, because what AI cannot access is the knowledge that was never recorded: the contextual read of a room, a patient's affect, a political dynamic that exists nowhere in any dataset.
The organizations currently deploying AI primarily around cost reduction and efficiency are making a defensible choice -- they are capturing real value and will be better positioned than organizations that delay. But the organizations that will look back at this period as a genuine strategic inflection point will be the ones that asked a different question: not where can we automate to reduce cost, but where is human error producing outcomes that are catastrophic, irreversible, or preventable -- and where does AI give us the ability to fundamentally change the quality of the judgment producing those outcomes. That question has answers in medicine, in legal systems, in engineering, in disaster preparedness, and in a dozen other domains this article has not had space to examine. The finding that more than half of all work hours can theoretically be automated is the beginning of that conversation, not the end of it.
The Question to Ask
Where in your organization is human error producing outcomes that are unacceptable — not just inefficient — and where does AI give you the ability to change the quality of the judgment that is producing those outcomes? That question has a different answer than "where can we reduce headcount?" It also has a more important one. The organizations that are asking it are building something that the ones focused only on cost reduction will not be able to replicate quickly when they realize what they missed.