Imagine a company with 1,000 employees. Its AI business case estimates that automation and augmentation could remove work equivalent to 500 full-time positions.
Has the technology made 500 people redundant? No. It has identified a theoretical productivity gain.
Technology makes capacity available. It does not make people redundant.
The organization releases capacity only when work actually disappears, roles are redesigned and old commitments are removed. Until then, a calculation about work can become a decision about people without the underlying economic claim ever being tested.
Time Saved Is Not Capacity Released
When an AI business case converts minutes saved per task into annual full-time equivalents, the arithmetic may be correct while the managerial conclusion is wrong.
Capacity can be fragmented across tasks, roles, functions and moments. Ten hours a week saved in each of twenty roles adds up to five full-time equivalents on paper. It does not automatically become five positions that can be removed or five people who can be reassigned. The time may be absorbed by rising demand, transferred into checking and exception handling, consumed by coordination or embedded inside roles whose remaining work is still necessary.
The organization must redesign work before theoretical savings become usable capacity. Without that step, efficiency exists in the model but not in the operating system.
Begin With Net Capacity
AI does not arrive as costless productive capacity. It brings technology and integration costs, data work, cybersecurity, compliance, model validation, human oversight, process redesign, new skills and management attention. During transition, old and new systems may run in parallel. Employees must implement the new capability while continuing to operate the business.
The relevant business-case equation is therefore:
Theoretical time-saving potential, minus fragmentation, transition effort, checking, exception handling, operating requirements, controls and coordination costs, equals the net capacity economically available.
A case that stops at gross time saved overstates the gain, understates the work required to realize it and encourages workforce decisions before the underlying capacity has become usable.
Make The Allocation Decision Explicit
If net capacity does become available, leadership still has to decide what it is for. The organization can reduce cost, absorb rising demand, increase output, improve quality or service, enter new markets, strengthen resilience and control, build further AI capability or perform work that was previously unaffordable.
Growth is one possible allocation. Cost reduction is another. Neither is the inevitable consequence of the technology.
When a business case moves directly from hours saved to positions removed, a preferred allocation has been presented as necessity.
That allocation may still be correct. Competitive pressure, declining demand or financial constraints may make cost reduction necessary. But leadership should own it as a choice, test its consequences and compare it with the alternatives. Technology provides evidence. It does not relieve executives of judgment.
The People Producing The Evidence May Carry Its Consequences
AI transformation commonly asks employees and managers to identify use cases, document work, expose inefficiencies, redesign processes and implement the technology. This is distributed execution within centrally determined boundaries.
The people with the knowledge required to create the productivity gain may also bear its consequences. If every verified saving is assumed to support the removal of roles, protecting tasks, minimizing reported savings or withholding knowledge can become a rational response rather than evidence of irrational resistance.
A general promise of augmentation will not resolve this conflict if the financial case assumes workforce reduction. Leaders must state whether the intended value comes from cost reduction, demand absorption, reinvestment or a deliberate combination. Ambiguity does not preserve optionality. It creates self-protection.
A productivity estimate is evidence about work. It is not yet a people decision.
Preserve Reversibility Until The Evidence Is Strong Enough
An organization should not remove capacity merely because a model performs the most visible part of a task. Roles also contain tacit knowledge, relationships, judgment about exceptions, informal coordination, challenge capacity and resilience that may not appear in the process map.
Before removing them, the CFO-grade question is: What would it cost to rebuild what we remove?
That cost may include recruitment, training, restored customer and supplier relationships, new controls, vendor dependency and the time required to recover when an apparently peripheral capability proves essential.
Reversibility is not an argument against efficiency. It is a discipline for avoiding irreversible decisions before the evidence is strong enough.
Assess Value In Stages
AI value should not be declared at the moment a tool works or a time-saving estimate is approved. It should pass through a staged assessment.
1. Technical potential. The technology can perform or support specified tasks. Nothing has yet been saved, released or realized.
2. Time saving observed. Human effort has demonstrably declined. The saving may still be fragmented, absorbed by demand or offset by new work.
3. Gross capacity made available. Leadership can see which work has disappeared and which work has migrated into review, exceptions, control and coordination.
4. Net capacity established. Transition, operating, governance and management costs have been deducted. What remains is concentrated enough to be economically usable.
5. Capacity allocated. Leadership decides whether the capacity will be removed, used to absorb demand, retained under a defined hypothesis or redirected to a specific purpose.
6. Operational effect demonstrated. The allocation produces an observable result in cost, output, quality, growth, service, risk or resilience. Additional busyness is not value.
7. Finance-confirmed value. Finance can connect the operational effect to an economic result without double counting or ignoring recreated cost. This is Booked Value.
8. Sustained value and reversibility reviewed. The result persists, the operating system remains viable, and lost knowledge, control or resilience does not reverse the gain. This is Booked and Sustained Value.
This AI-specific sequence refines the front of MANREAU's canonical value ladder. Stages 1 to 6 establish the movement from Expected Value toward Realized Value; stage 7 is Booked Value; stage 8 is Booked and Sustained Value.
The complete chain is: Technical capability → time saved → gross capacity made available → net capacity established → capacity allocated → operational effect demonstrated → Booked Value → Booked and Sustained Value.
What The CEO And CFO Must Require
A CEO should be able to bring an AI business case back to five questions in the room:
1. Work Reality. What work will genuinely disappear, and what new work will AI create?
2. Net Capacity. After transition, oversight and operating costs, how much capacity will actually become available?
3. Allocation Choice. What deliberate choice are we making with that capacity: remove it, absorb demand or reinvest it, and who carries the consequences?
4. Reversibility. Which knowledge, capability or resilience could we lose, and what would it cost to rebuild?
5. Value Confirmation. What evidence will prove the operational effect, and when will Finance confirm Booked and Sustained Value?
What will leadership do with the net capacity this case creates, and is that decision written into the case or merely assumed by it?
This Perspective presents a conceptual management test rather than a claim about the prevalence of particular AI business-case practices. Its propositions about fragmentation, new operating work and capacity realization should be tested against the specific work, costs and evidence of each organization.
