Too many organizations are making AI workforce decisions in the easiest unit to measure: people. That is already the wrong starting point.

AI operates at a different level. It changes tasks, workflows, decisions, handoffs, and the amount of human effort required inside roles. Most organizations have never fully mapped those things.

When AI success gets defined as labor savings, cost per head, or positions eliminated, headcount reduction quickly becomes the objective. That creates the wrong sequence.

Recent evidence highlighted by Harvard Business Review shows how often that sequence fails. Among 600 HR leaders who made AI-driven layoffs, only 8.4% said the restructuring delivered as promised and that they would make the same decision again. One in three said they lost critical skills and expertise. Gartner found that 80% of companies reduced headcount when implementing autonomous business capabilities, with no correlation between lower employee numbers and return on investment. That should change how leaders approach AI transformation.

The measurement system should come first

Organizations optimize around what they measure. If the primary AI metric is labor removed, then labor removal will become the goal. The measurement system needs to change before the workforce model does.

Leaders should define what value actually looks like before deciding what AI means for staffing. That requires measuring far more than cost reduction.

A stronger AI value framework includes:

• Capacity created

• Decision quality and speed

• Revenue and margin impact

• Customer outcomes

• Rework and escalation

• Innovation throughput

• Time redirected toward higher-value work

• The economic value created from newly available capacity

That last measure matters most. AI ROI should capture more than avoided labor cost. It should capture what the organization can now do better, faster, or at greater scale because AI created additional capacity.

If AI gives a team back 20% of its time, the first question should not be, “How many people can we remove?”

The better question is:

• What can this organization now do that it could not do before?

• Can that capacity support more customers?

• Can it accelerate product development?

• Can it improve pricing, selling, retention, service, or decision-making?

• Can it create more room for experimentation and innovation?

• Can employees move away from repetitive execution and toward work where judgment, creativity, relationships, and business context create substantially more value?

Those questions produce a very different AI strategy.

Capacity is a strategic asset

Productivity gains create capacity. Capacity creates options. Leaders can reinvest that capacity into growth, customer experience, innovation, faster execution, or entirely new capabilities before treating it as excess labor.

The Harvard Business Review authors make a similar argument when they call for a business strategy rather than an AI strategy.

JPMorgan Chase offers a useful example. The company reduced some operations and support roles while expanding client-facing and revenue-generating teams. AI-created efficiencies did not automatically translate into broad workforce reduction. The company redirected resources toward areas that could create more value.

That is the more important economic question. Productivity is only one part of the return. The larger opportunity comes from what the organization does with the capacity that productivity creates. Companies that treat every efficiency gain as a labor reduction risk eliminating the very people who could convert that capacity into growth, innovation, and better decisions.

Decompose the work before changing the workforce

AI changes tasks, not job titles. That distinction matters. Most roles contain a combination of visible work and invisible work.

The visible work is easy to describe: produce the report, answer the customer question, reconcile the data, create the forecast, review the transaction, prepare the analysis.

The invisible work is harder to see: interpret the exception, challenge the recommendation, coordinate across teams, understand historical context, escalate the right issue, recognize when the standard process does not apply. That work rarely appears in a job description. It still creates enormous value.

When organizations eliminate roles before understanding their component tasks, they often automate the visible output and unintentionally remove the judgment that made the process work.

Citigroup's approach provides a better model. The company started with 50 processes targeted for greater automation and allowed that analysis to inform staffing decisions. The work analysis came first. The staffing decision followed.

That sequence matters because a role is rarely a single unit of work. It is a bundle of tasks, decisions, relationships, and responsibilities. Leaders need to understand that bundle before deciding what AI can replace, what it can augment, and what still requires human ownership.

Protect the invisible work

Organizations often discover the missing human capability only after it disappears.

• Oversight.

• Exception handling.

• Context translation.

• Judgment.

• Accountability.

• Institutional knowledge.

These are the capabilities that become visible when the people who performed them are gone. AI may produce an output, but someone still needs to determine whether that output makes sense in context. Someone still needs to own the consequences. Someone still needs to know when the model is technically correct but operationally wrong. Those capabilities need explicit owners.

That is especially important as organizations introduce more autonomous systems. The more work AI performs, the more deliberate leaders need to become about where human judgment remains essential.

Design for reversibility

AI capabilities are changing quickly. So are the ways organizations are learning to use them. That uncertainty should affect workforce design.

Forrester expects half of AI-attributed layoffs to eventually be reversed. Reversal is expensive. Rehiring takes time. Institutional knowledge does not automatically return with the replacement. Relationships, context, and organizational understanding can take years to rebuild. That makes reversibility an economic design decision.

Organizations should use redeployment and attrition where possible, phase workforce changes, and understand the cost of rebuilding a capability before removing it.

The question should not only be: What do we save if we eliminate this work?

Leaders also need to ask: What will it cost if we discover eighteen months from now that we still need it?

Those are very different calculations.

The sequence matters

The strongest AI transformations will follow a different order:

1. Measure value.

2. Create capacity.

3. Deploy that capacity toward growth and better decisions.

4. Redesign the work.

5. Then determine what workforce structure the new operating model actually requires.

That sequence moves the workforce decision downstream, where it belongs.

Some roles will still change. Some work will disappear. Some organizations will ultimately need fewer people in certain areas. But those decisions should emerge from a redesigned operating model and a clear understanding of economic value, rather than from a preselected headcount target.

Smaller is not the same as more valuable

The core issue is becoming clearer. Human judgment, institutional knowledge, and accountability have economic value. They should be measured alongside productivity gains.

The companies that start with a headcount target risk becoming smaller without becoming more valuable, productive, capable, or intelligent. The companies that get this right will use AI to redesign how work gets done and deliberately determine where human and machine capabilities create the most value.

That is a much harder management problem than cutting positions. It is also where the real AI return will come from.

I explored the broader risk in “The Headcount Trap: How Workforce Reduction Undermines AI Value.” The central argument is closely related: organizations can undercut the value AI is meant to create when they remove the judgment, capability, and institutional knowledge required to create and govern that value in the first place.

Related reading: The Headcount Trap: How Workforce Reduction Undermines AI Value
https://ambailey91406.substack.com/p/the-headcount-trap

Source: Harvard Business Review, “AI Transformation Requires Redesigning Work, Not Cutting Roles,” Faisal Hoque, Tom Davenport and Paul Scade, August 28, 2026.

Contact

Start a conversation about AI strategy, governance, or operating model design.

I work with leadership teams that need clearer direction, stronger governance, and a practical path from AI experimentation to execution. For consulting inquiries, advisory support, or speaking requests, reach out directly.

Email

Amanda@BaileyAppliedIntelligence.com

© 2025. All rights reserved.

Linkedin

www.linkedin.com/in/amandakbailey

PHONE

+1 (424) 249-9862