Organizations have been reducing headcount for decades. There is nothing particularly new about restructuring a workforce when demand changes, costs increase, markets shift, or an organization needs to rebalance capacity.
What is different today is the environment in which those decisions are being made.
Businesses are larger. Many operate globally. Enterprise systems have connected functions that once operated far more independently. Decisions cross organizational boundaries, technical systems, customers, suppliers, geographies, and functions.
And now AI has entered the equation.
That makes what appears to be a relatively simple question increasingly difficult:
How many people does the organization actually need?
The easiest answer is to start with a percentage.
If the objective is to reduce headcount by 20%, identify the positions that appear redundant, consolidate responsibilities, automate repetitive activities, redistribute the remaining work, and reduce the organization accordingly.
Mathematically, the exercise is simple.
Organizationally, it may be anything but.
Start With the Work, Not the People #
Before deciding which positions can be eliminated, organizations need to understand the work being performed.
One starting point is relatively straightforward: ask leaders and their teams to identify repetitive work.
What activities consume time but require limited human intervention? Which could be simplified, eliminated, automated, or performed differently?
Then map the processes.
Not only the processes contained within individual departments, but those that cross departmental boundaries.
Where does Finance depend upon Operations?
Where does Supply Chain depend upon Commercial?
Where does Customer Service depend upon Planning?
Where does one person’s work become another person’s input?
Traditional process mapping can begin to expose these relationships.
But in an AI-enabled organization, process mapping alone is no longer enough.
We also need to map decisions.
Where are decisions made?
Who makes them?
Who influences them?
Where are decisions changed?
Which decisions are critical?
Which decisions intersect with others?
Which depend upon information from inside the organization, and which depend upon customers, suppliers, regulators, or other external sources?
Most importantly, what human capability is required to make those decisions well?
That is where the headcount equation becomes much more complicated.
A Position Is More Than Its Tasks #
Imagine an employee whose work is 60% repetitive.
AI or automation may be able to perform much of that work.
It would be tempting to conclude that the position can therefore be eliminated.
But that conclusion assumes the repetitive work represents the value of the position.
What about the remaining 40%?
That employee may recognize when a customer’s behavior deviates from the normal pattern. She may understand why two apparently identical products need to be planned differently. She may know which supplier commitment should be challenged. She may understand why a process was designed the way it was because she remembers what happened the last time the organization changed it.
She may also be the person others call when the process stops working.
Those activities may never appear clearly in her job description.
They may not even appear on a process map.
Yet those activities may represent the very expertise, judgment, institutional knowledge, relationships, and organizational memory upon which other parts of the organization depend.
Eliminating the repetitive work does not necessarily mean that the human capability surrounding that work is no longer required.
That distinction becomes increasingly important as organizations redesign roles around AI.
Capacity and Capability Are Not the Same Thing #
There is another variable that workforce models need to consider: capacity.
An organization can retain highly capable people and still overwhelm the system.
Suppose a restructuring removes 20% of the workforce and distributes the remaining responsibilities among those who stay.
Those employees may possess the necessary expertise.
But how much capacity do they have to use it?
How many decisions are they now expected to make?
How complex are those decisions?
How many functions must they coordinate with?
How many AI-generated recommendations must they evaluate?
How frequently are they interrupted?
How much time remains for investigation, reflection, learning, collaboration, and challenging assumptions?
A person’s expertise does not disappear simply because workload increases.
But the capacity available to exercise that expertise is not unlimited.
This creates an important distinction:
Human capability asks what a person is capable of doing.
Human capacity asks how much of that capability can reliably be deployed within the demands of the system.
Organizations need to understand both.
AI Changes the Equation #
AI can unquestionably perform work that humans perform today.
It can analyze information, identify patterns, automate repetitive activities, generate recommendations, summarize enormous quantities of information, and dramatically increase the speed at which organizations process information.
That does not automatically tell us how many humans the organization requires.
Instead, AI forces us to ask a different question:
What work should humans perform, what work should AI perform, and where must they work together?
Some processes may become almost entirely automated.
For others, AI may conduct the analysis while a human makes the decision.
In another process, AI may make a recommendation that a human must challenge, interpret, or approve.
And for some consequential decisions, human expertise, judgment, leadership, or broader contextual understanding may remain essential.
Each configuration creates a different human-capacity requirement.
There is also a possibility organizations need to consider carefully: AI may reduce work in one part of the decision system while increasing work somewhere else.
If AI enables an organization to generate significantly more analyses, recommendations, alerts, scenarios, and potential decisions, someone still needs to determine which matter.
The analytical bottleneck may disappear while a decision bottleneck emerges.
The constraint has moved.
It has not necessarily been eliminated.
Before You Cut 20% #
This is why determining the “right” number of employees in an AI-enabled organization cannot begin and end with a headcount target.
Process mapping is an important starting point.
But organizations increasingly need to understand several interconnected dimensions:
What work is being performed?
How does that work move across the organization?
Where are decisions being made or changed?
Which decisions are critical or interconnected?
What expertise is required?
Where is judgment required?
Where do leadership and broader contextual understanding matter?
What organizational knowledge exists primarily within people?
How much decision-making capacity will the remaining workforce actually have?
Which activities should AI perform?
Which should humans retain?
And how will those requirements change as roles evolve?
Only after answering those questions can an organization begin to understand what its future workforce should look like.
This changes the fundamental question.
Instead of asking:
“If AI increases productivity, how many people can we remove?”
Perhaps leaders should ask:
“What combination of human capability, human capacity, and artificial intelligence does our future decision system require?”
That analysis may eventually produce the same 20% reduction.
They may produce 10%.
They may produce 30%.
They may reveal that one function can be dramatically reduced while another requires investment.
The point is not that organizations should avoid reducing headcount.
The point is that the percentage should be an outcome of understanding the future operating system—not the assumption from which the design begins.
Because an organization can eliminate a position relatively easily.
Understanding everything that disappears with it is much harder.
References
