Before AI Changes Your Workforce, Redesign the Work
The easiest AI strategy to explain is also becoming one of the most common: Cut headcount.
AI can automate more work. Therefore, fewer people should be needed.
It sounds efficient. It may also be backwards.
A recent Harvard Business Review article, AI Transformation Requires Redesigning Work, Not Cutting Roles, argues that many companies are making workforce decisions faster than the evidence warrants.
In one survey, 55% of businesses that had laid off workers because of AI later said they had made the wrong staffing decisions. In another, only 8.4% of HR leaders said their AI-driven restructuring delivered as promised and that they would make the same decision again.
The problem is not simply that companies are cutting too aggressively.
It is that they are starting from the wrong unit of analysis.
Why are companies redesigning headcount before they redesign the work?
Most jobs are bundles of different activities.
Some are repetitive. Some involve coordination. Some require judgment. Some rely on relationships, institutional knowledge, exception handling, accountability, or context.
AI may automate only part of that bundle.
Yet companies often make workforce decisions at the level of the role. That creates a dangerous shortcut:
↓
We no longer need the person performing the role.
The HBR authors argue that this is precisely where organizations get into trouble. Eliminating a role based on its most visible tasks can expose less visible work that still needs to be done, eventually forcing the organization to reconstitute the role or hire people back.
That gives us an important distinction:
Role ≠ work
And another:
Task automation ≠ workforce redesign
Companies are making decisions ahead of the evidence
The article includes an especially revealing finding.
In one survey of more than 1,000 global executives, only 2% of organizations had made large headcount reductions based on actual AI implementation, while 60% had reduced staff or hiring with an expressed connection to AI.
That gap matters.
It suggests that many companies are restructuring around what they expect AI to become rather than what AI has already demonstrated inside their own workflows.
That is not necessarily strategy. It may be speculation translated into organizational change.
And workforce decisions are not easily reversible. Once experienced employees leave, they take institutional knowledge, relationships, context, and tacit understanding with them. Rehiring later does not simply restore what was lost. The authors therefore argue that reversibility should become an explicit design criterion in AI-driven workforce decisions.
A useful principle follows:
When AI capability is uncertain, irreversible decisions require stronger evidence.
Start with the workflow
A better sequence begins somewhere else.
Not with: How many people can we remove?
But with: What work is actually being performed?
Then:
- Which tasks can AI perform reliably?
- Which tasks still require human judgment?
- Where are the bottlenecks?
- Which activities create customer or business value?
- What breaks if a role disappears?
- What new capabilities become possible if AI improves the workflow?
This is why the HBR authors recommend decomposing roles into component tasks before making staffing decisions. They point to Citigroup as one example: the bank began by identifying processes that could be improved through technology and redesign, then allowed staffing implications to follow from that analysis.
The analysis came first. The restructuring followed. That order matters.
AI strategy should follow business strategy
Another useful point in the article is that companies should not treat AI as an isolated cost-cutting program.
The authors argue that the better question is not: What can AI do to reduce headcount?
It is: What combination of human and technological capabilities best helps the business achieve its goals?
Sometimes the answer will involve fewer people. Sometimes it may mean redeploying people into higher-value work. Sometimes it may mean expanding revenue-generating or customer-facing capacity.
The article points to JPMorgan Chase as an example of this broader approach: some operations and support roles declined while client-facing and revenue-generating teams expanded, leaving overall headcount relatively stable.
The objective was not headcount preservation. It was capability allocation. That is a much more strategic framing.
Cutting people is not the same as creating value
This connects directly to a larger problem in AI economics.
If every company uses AI primarily to reduce labor costs, the initial gains may be real. But the competitive advantage may not last. Competitors can adopt similar tools. Cost savings can become expected. Prices can adjust. Customers can capture part of the gain. What begins as a margin improvement can become an industry baseline.
So cutting people is not enough.
The more durable question is: How does AI make the organization better at creating value?
That may involve cost reduction. But it may also involve:
- faster workflows
- better decisions
- higher service quality
- greater productive capacity
- new products
- better customer experiences
- improved risk management
- redeployment into higher-value work
Headcount is one variable. It is not the strategy.
The real unit of AI transformation is the workflow
This is where AI workforce transformation becomes a decision-systems problem.
The organization needs to understand:
- What work is actually being performed?
- Which tasks can AI perform reliably?
- Which tasks still require judgment, accountability, context, or relationships?
- Where does removing a role break the workflow?
- What productivity has actually been observed?
- What staffing changes are supported by evidence?
- Which changes should remain reversible until confidence improves?
Those questions move the organization away from AI speculation and toward evidence-based redesign.
A stronger sequence looks like this:
The weaker sequence is:
One begins with the business. The other begins with the assumption that fewer people must be better.
Smaller is not necessarily smarter
The HBR authors end with a useful warning: if headcount reduction becomes the primary measure of AI success, companies risk becoming “smaller but not smarter.”
That may be the most important lesson here.
The goal of AI transformation should not be: fewer people.
It should be: better-designed work.
Reduce headcount where the evidence supports it. Redeploy people where human capabilities still create value. Redesign workflows where AI changes what is possible. And preserve reversibility where the evidence is still weak.
Because the real strategic question is not: How many roles can AI eliminate?
It is: How should the work change now that AI exists?