Mathematical Expressions of Work

How automation and mathematical modeling are essential to transforming organizational roles, workflows, and competitiveness

The quest to express work in mathematical terms is a transformative effort at the intersection of organizational science and practical management. For most of business history, jobs were defined by their titles, not their contents. A “Financial Analyst” and a “Senior Associate” might be doing substantially different work – or nearly identical work – with no formal framework to tell them apart. That ambiguity made workforce planning intuitive at best and political at worst.

Translating work into a mathematical framework changes this. It enables objective measurement and comparison of jobs and tasks, eliminating ambiguity about what constitutes productive work. It supports the automation and optimization of workflows by providing a precise language for what machines can and cannot do. And it gives managers a design tool – not just a descriptive one – for reimagining roles as AI capabilities expand.

This article builds the formal framework in three layers: first, a set-based model of the job as a collection of tasks; second, a partition model that separates automatable from human-required tasks; and third, a continuous degree-of-automation model that captures the reality of human-in-the-loop work. Each layer adds precision and practical utility. The final section translates the math into organizational implications that any manager can act on.

Part I: The Job as a Structured System of Tasks

We begin with the most fundamental question: what is a job, mathematically? The answer is not a role, a title, or a headcount entry. A job is a structured collection of tasks, each with attributes, dependencies, and constraints. The following six definitions build this model progressively from minimal to complete.

1. Basic Set-Based Definition

The simplest and most foundational model: a job is a finite set of tasks. This captures what the job entails, without yet specifying how tasks relate to one another or what attributes they possess.

Definition

Let a job be defined as a finite set of tasks:

Where

J

Job

The structured system of tasks that constitutes a defined unit of organizational work. A job is not a person, a title, or a role — it is a collection of tasks that can be decomposed, analyzed, and redesigned.

t_i

Individual Task

A discrete, bounded unit of work within the job. Tasks are the atomic unit of the framework. Every job is reducible to its constituent tasks, and it is at the task level — not the job level — where automation decisions are made.

n

Number of Tasks

The cardinality of the task set. Two jobs with the same title but different n, or different task compositions, are mathematically distinct jobs — regardless of what the org chart says.

Implication

Job titles lose explanatory power the moment you define the job mathematically. Two analysts with the same title, performing different tasks, are in different jobs. This has direct consequences for compensation equity, automation planning, and performance measurement.

Organizational consequence

Workforce design shifts from headcount planning (how many people?) to task allocation planning (which tasks need to be performed, by whom or what, and how often?). This is the foundational shift that AI makes necessary.

2. Job as an Aggregation Function

A set tells us what a job contains, but not how those tasks combine into a coherent unit of work. The aggregation function F captures the logic of combination – sequencing, coordination, execution flow, and interdependence among tasks.

Definition

If we want the job to be an aggregation of tasks, not just a list:

Where

F(·)

Aggregation Function

The function that combines individual tasks into the job. F could represent sequencing logic (task A must precede task B), coordination logic (tasks A and B must be executed in parallel), or execution constraints (task C can only begin when tasks A and B are complete). F is what makes a job more than the sum of its parts.

t₁…t_n

Ordered Task Inputs

The tasks passed to F. The order and structure of these inputs matters — F is not necessarily commutative. Changing the sequence of tasks can affect the job's output quality, duration, and cost.

For a simple additive workload (tasks are independent and sequential), F reduces to a sum:

Implication

Process redesign is the act of changing F, not just changing which tasks are in the set T. Many automation initiatives fail because they replace individual tasks (changing T) without redesigning the aggregation logic (changing F). A workflow designed around human sequential processing may yield no value when individual tasks are automated, yet F remains unchanged.

Organizational consequence

AI-driven workflow redesign requires changing both T (the task composition) and F (the sequencing and coordination logic). The highest ROI transformations address both simultaneously — removing tasks that exist only because of human coordination overhead and collapsing sequential steps that AI can execute in parallel.

3. Tasks with Attributes

A task is not just an activity — it is a measurable unit with quantifiable properties. Adding attributes to the task model transforms it from a conceptual tool into an analytical one. Three attributes are sufficient to describe most tasks for automation planning and workforce design purposes.

Definition

Let each task be a vector of three core attributes:

Where

δ_i

Duration

The time required to execute task i. Duration drives labor cost, throughput capacity, and the time-value calculation for automation investment. AI reduces δ_i dramatically for cognitive and documentation tasks — analysis that takes a human 4 hours may take an AI 4 minutes.

c_i

Cost

The monetary cost of executing task i, including labor, tooling, and overhead. When a task is automated, c_i shifts from variable (human labor rate × δ_i) to fixed (infrastructure and model costs). This shift in cost structure is one of the central economic effects of AI adoption.

σ_i

Skill / Judgment Level

The cognitive or judgment requirement of task i. This is the key automation-eligibility signal. Low-σ tasks (data entry, formatting, retrieval, verification) are highly automatable. High-σ tasks (client judgment, strategic framing, ethical reasoning, novel problem-solving) require human execution. σ_i is the variable that determines task fate in an AI economy.

The job as a whole can then be described as an aggregate of its task attributes:

Implication

This model makes it explicit that total cognitive demand (σ) accumulates only in human tasks — and it is this accumulated judgment load that defines what is irreplaceable about a given job. As automation absorbs low-σ tasks, the remaining human work becomes cognitively denser and more valuable per unit of time.

Organizational consequence

Managers can now audit their teams’ task portfolios by σ level. Tasks where σ_i is low and δ_i is high are the highest-priority automation candidates. Tasks where σ_i is high are where human development investment should concentrate. This is the analytical foundation for the workforce redesign.

4. Job as a Weighted Task Function

Not all tasks contribute equally to a job’s output or value. A client call that drives $1M in revenue and an email that updates a calendar both appear as tasks in the set model, but they are not equivalent. The weighted task model assigns importance weights to each task, enabling more precise analysis of where value is created and where automation has the most leverage.

Definition

If tasks contribute unequally to the job’s output, define:

Where

w_i

Task Weight

The relative importance or priority of task i in determining the job’s output quality, business value, or strategic contribution. Weights sum to 1 when normalized. High-w tasks are where performance variation matters most; low-w tasks are where automation savings are most painless.

t_i

Individual Task

The task input to the weighted aggregation, as defined in the set-based model. The task itself is unchanged — the weight is an additional property that describes the task’s contribution to the job’s total output.

Implication

Weight and automation-eligibility are independent dimensions. A task can be high-weight and automatable (e.g., financial report generation), low-weight and automatable (e.g., scheduling), high-weight and human-required (e.g., a critical negotiation), or low-weight and human-required (e.g., a brief peer check). The highest-value automation targets are tasks that are both high-weight and automatable — they deliver cost savings without sacrificing the work that creates the most value.

Strategic consequence

AI investment prioritization should be driven by the product of w_i and automation-eligibility, not by headcount impact. A firm that automates ten low-weight tasks may see less economic impact than one that automates one high-weight task at the center of its delivery model.

5. Job with Dependencies and Cycles

Most jobs do not consist of independent tasks. In practice, tasks have dependencies (task B cannot begin until task A is complete), parallelism (tasks C and D can run simultaneously), and cycles (task E repeats after task F feeds back into it). A directed cyclic graph formally captures this structure.

Definition

Define the job as a directed cyclic graph (DCG), where nodes are tasks and edges are dependencies:

Where

T

Task Set (Nodes)

The set of all tasks in the job, now represented as nodes in the graph. Each node has the attributes (δ, c, σ) defined in the attribute model.

E

Dependency Edges

The set of directed edges E ⊆ T × T, where (t_i, t_j) ∈ E means task t_i must precede task t_j. The structure of E defines the critical path, parallelism opportunities, and cycle points of the job.

DCG

Directed Cyclic Graph

A graph that permits cycles — meaning some sequences of tasks repeat. This is the realistic model for most operational jobs, where steps like review-and-revise or the collect-analyze-report cycle are repeated continuously rather than terminating at a single point.

A practical example: BIP Capital’s Quarterly Performance Reporting cycle is a DCG. Every quarter, the team: (A) evaluates portfolio company financial performance, then (B) runs valuation models, then (C) calculates investor returns, then (D) publishes performance reports. The DCG is A → B → C → D → A, cycling every quarter indefinitely.

6. General Abstract Definition

The five models above can be unified into a single general definition that is flexible enough to accommodate any job structure: simple or complex, linear or cyclic, uniform or weighted. This is the most complete representation and the one on which the automation analysis in Part II is built.

Definition

Let the most general job model be defined as:

Where

T

Task Set

The complete set of tasks {t₁, t₂, …, t_n}, each with attributes (δ, c, σ) and dependency relationships captured in E.

r

Automation Degree Vector

A vector of automation degrees [r₁, r₂, …, r_n], one per task, where r_i ∈ [0,1]. This will be defined precisely in Part II. At this stage, r is introduced as a placeholder for the automation dimension.

M(·)

Machine Execution Function

The function that maps automatable tasks to machine-executed outputs. M(·) captures everything the AI or automation system does: speed, cost structure, output format, and reliability profile.

H(·)

Human Execution Function

The function that maps human-required tasks to human-executed outputs. H(·) captures judgment, creativity, accountability, and the relational dimensions of work that cannot be replicated by current AI systems.

Ω

Constraint Set

All binding constraints on how the job can be executed: regulatory requirements, ethical boundaries, quality standards, timing windows, resource limits, and governance rules. Ω defines the feasibility region within which F, M, and H must operate.

Implication

Every job in every organization can be described by some instance of this model. The value of having the general form is that it provides a shared language across operators, technologists, executives, regulators, and investors — enabling precise conversations about what a job entails, where automation applies, and which constraints must be preserved.

Organizational consequence

The general model serves as the foundation for AI ROI modeling, workforce redesign, and governance frameworks. Any organization that can describe its key jobs in this form has the analytical foundation to make automation decisions rigorously rather than intuitively.

Part II: Partitioning the Job — Automatable vs. Human-Required Tasks

The general job model establishes the structure. Part II applies the most consequential analytical cut: partitioning the task set into work that machines can execute and work that requires human judgment. This partition is the operational heart of every AI workforce strategy.

1. Defining the Task Taxonomy: Partitioning the Task Set

The first step in any automation analysis is formally partitioning the job’s task set T into two disjoint subsets: automatable tasks (T_A) and human-executed tasks (T_H). This partition is not permanent — it shifts as AI capabilities evolve — but it is the necessary starting point for every workforce and automation decision.

Definition

Let a job consist of n total tasks. Partition the task set into two disjoint subsets:

Where

T

Total Task Set

The complete set of all n tasks in the job. The partition covers T entirely — every task is either in T_A or T_H, and no task belongs to both.

T_A

Automatable Tasks

The subset of tasks that current AI, software, or automation systems can execute reliably, at acceptable quality and cost. Membership in T_A is not permanent — tasks migrate from T_H to T_A as AI capabilities advance. Today’s T_H is tomorrow’s T_A.

T_H

Human-Required Tasks

The subset of tasks that require human execution: judgment under ambiguity, accountability, ethical reasoning, relational work, novel problem-solving, and exception handling. These are not ‘leftover’ tasks — they are the highest-value tasks in the job.

Implication

The partition is the starting point, not the endpoint. The strategically important question is not the current state of the partition but its trajectory: which tasks in T_H are moving toward T_A, and how fast? Organizations that track this trajectory continuously can anticipate workforce changes rather than react to them.

Organizational consequence

Task taxonomy mapping should be a standing organizational capability, not a one-time project. As AI models improve, tasks that were in T_H six months ago may now belong in T_A. Companies that re-partition annually will continuously find new automation opportunities their competitors have not yet seen.

2. Indicator-Function Formulation

The partition model is powerful but binary. The indicator-function formulation encodes the partition compactly and integrates it into the job equation, enabling clean mathematical manipulation and analysis of automation coverage across the full task set.

Definition

Define a binary automation indicator for each task:

Using this indicator, the job can be written as a clean sum of machine-executable and human-dependent work:

Where

a_i

Automation Indicator

The binary variable that assigns each task to either the machine-executable (a_i = 1) or human-dependent (a_i = 0) category. The indicator makes it simple to calculate automation coverage: Σa_i / n gives the fraction of tasks in T_A.

M(t_i)

Machine Execution Output

The output produced when the machine executes the automatable task t_i. M captures the speed, cost structure, and reliability profile of AI or software execution — typically much faster and cheaper than H at scale, but constrained in judgment capacity.

H(t_i)

Human Execution Output

The output produced when human execution is applied to a human-required task t_i. H captures the judgment, contextual reasoning, and accountability that human workers provide. H outputs carry ethical and legal weight that M outputs do not.

Implication

The indicator formulation makes automation coverage measurable. For any job, Σa_i / n gives the proportion of tasks that are automatable. For any function or team, the weighted average of Σa_i·w_i gives the proportion of value-weighted work that can be automated. These are actionable metrics that leadership can track, target, and report.

Workforce consequence

Roles with high Σa_i (many automatable tasks) face the most near-term structural change. Roles with low Σa_i (mostly human-required tasks) are resilient. Knowing which roles fall where — and how the distribution is shifting — is the foundation of defensible workforce planning in an AI economy.

3. Job as a Tuple of Automated and Human Components

Rather than treating the job as a single aggregated function, we can explicitly define it as a two-component system: one component executed by machines, the other by humans. This tuple representation makes the job's architecture visible and supports separate analysis of each component.

Definition

Define the job explicitly as an ordered pair of its automated and human components:

Where

J_A

Automated Job Component

The portion of the job executed by machines, software, or AI agents. J_A = M(T_A) is the application of the machine execution function to all automatable tasks. This component scales without additional labor cost and is the primary source of margin expansion in AI-augmented firms.

J_H

Human Job Component

The portion of the job executed by humans. J_H = H(T_H) is the application of the human execution function to all human-required tasks. This component is irreplaceable in the near term and represents the concentrated judgment, accountability, and relational work that AI cannot replicate.

M(T_A)

Machine Execution Function Applied to T_A

The composite output of applying M to all automatable tasks in the job. In practice, this is what an AI workflow, RPA system, or autonomous agent delivers when assigned T_A.

H(T_H)

Human Execution Function Applied to T_H

The composite output of applying H to all human-required tasks. In practice, this is what the remaining human workforce delivers after automation has absorbed T_A.

Implication

The tuple representation makes visible something that single-metric automation analyses obscure: J_A and J_H are not symmetric. J_A scales efficiently but produces outputs bounded by current AI capability. J_H does not scale cheaply but produces outputs that carry judgment, accountability, and relational value that J_A cannot replicate. The organization’s competitive advantage lives in J_H — automation exists to create the capacity to expand it.

Organizational consequence

As T_A grows and J_A expands, the strategic question shifts: what do we do with the J_H capacity that is freed? The Labor Redeployment Model in (Model 4) is the economic answer. The Mathematics of Value defines where that freed capacity should be directed.

4. Attribute-Based Expansion: Time, Cost, and Judgment

Adding the task-attribute model from Part I to the partition framework yields a fully specified account of how automation changes the economic and cognitive structure of the job. This is where the math connects directly to measurable financial and workforce outcomes.

Definition

With each task carrying attributes (δ_i, c_i, σ_i), the automated and human components expand to:

Where

δ_i

Task Duration

Time required to execute task i. When task t_i moves from T_H to T_A, its duration contribution shifts from J_H’s time budget to J_A’s time budget — typically at a fraction of the original time cost. The aggregate human time budget Σ(δ_i | a_i=0) shrinks as automation expands.

c_i

Task Cost

The monetary cost of executing task i. When a task is automated, c_i shifts from a variable labor cost (human time × labor rate) to a fixed or marginal infrastructure cost (AI model inference or software execution). This cost structure shift is the primary driver of operating leverage in AI-augmented businesses.

σ_i

Judgment Requirement

The cognitive or skill level required for task i. The automated component carries zero judgment requirement (σ = 0 for all tasks in T_A). Judgment accumulates only in J_H: max(σ_i | a_i=0) is the peak skill demand of the residual human job.

max σ_i

Peak Judgment Requirement

The highest cognitive demand task remaining in the human component. As low-σ tasks migrate to T_A, max σ_i rises — meaning the human job becomes progressively more cognitively demanding, not less. This is the mechanism behind job polarization.

Implication

This model makes explicit that automation does not reduce the cognitive intensity of human work — it concentrates it. As the low-σ tasks in T_H migrate to T_A, the remaining human tasks are disproportionately high-σ. The human job becomes harder cognitively even as it becomes shorter in time terms. Workers who do not develop higher σ_i capabilities will find their role shrinking faster than they can adapt.

Workforce consequence

Training and development investment should target the σ_i distribution of the remaining human task set, not the original job description. Organizations that continuously re-map task σ levels and align development programs accordingly will build the cognitive density needed to remain competitive as T_A expands.

5. Degree-of-Automation: The Continuous Model

The binary partition model (a_i ∈ {0,1}) captures the endpoints of the automation spectrum but misses the most common real-world state: human-in-the-loop. Most AI deployments are not fully autonomous — they assist, augment, or partially automate human tasks, requiring human review, oversight, or exception handling at various points. The continuous model captures this reality.

Definition

Instead of binary automation, define a continuous degree-of-automation r_i for each task:

The job model then becomes a weighted combination of machine and human execution for every task:

Where

r_i

Degree of Automation

A continuous value in [0,1] representing the fraction of task t_i that is machine-executed. r_i = 0 is fully human. r_i = 1 is fully automated. r_i = 0.6 means 60% of the task’s execution is handled by AI, with 40% requiring human involvement (review, judgment, sign-off, or exception handling).

r_i·M(t_i)

Machine Contribution to Task i

The portion of task i’s output generated by machine execution, scaled by the degree of automation. As r_i rises toward 1, this term dominates and the machine produces most of the task output.

(1-r_i)·H(t_i)

Human Contribution to Task i

The portion of task i’s output generated by human execution, scaled by the residual human fraction. As r_i rises toward 1, this term shrinks — but it never disappears entirely for tasks where accountability or judgment is required.

Implication

The continuous model captures copilot arrangements, review layers, and partial automation — the dominant real-world AI deployment pattern in 2025. Most organizations are not choosing between fully automated and fully human; they are managing a portfolio of r_i values across hundreds of tasks, each at a different point on the automation spectrum.

Strategic consequence

The right question is no longer ‘will this job be automated?’ It becomes: ‘which tasks are moving along the r_i spectrum, at what speed, and what is the correct r_i target for each given quality, cost, and accountability constraints?’ This is a continuous optimization problem, not a binary classification exercise.

Part III: Organizational Implications of the Mathematics of Work

The following eight implications translate the mathematical framework above into actionable organizational insight. Each is grounded directly in the models introduced in Parts I and II.

1. Jobs Stop Being Atomic: They Become Decomposable Systems

The set-based model makes this inescapable: a job is not a single atomic act performed by a single human. It is a structured system of tasks, each of which is decomposable, measurable, and independently assignable. Once you define jobs mathematically, the concept of a ‘job’ as a stable unit of organizational work begins to dissolve.

2. Automation Is a Continuous Variable, Not a Binary Decision

By defining r_i ∈ [0,1], the model formalizes what practitioners already know: automation is partial, evolving, and task-specific, not a binary switch. Most AI deployments in 2025 are copilot or human-in-the-loop arrangements — not full replacement — and the continuous model captures this accurately.

3. Human Value Becomes Concentrated, Not Eliminated

The attribute-based expansion model is unambiguous: σ_i accumulates only in human tasks. As T_A grows and low-σ tasks migrate out of J_H, the human job becomes cognitively denser — not smaller in value, but smaller in headcount requirements and larger in per-person cognitive demand.

4. Productivity Gains Come from Task Rebalancing, Not Labor Elimination

The productivity gain model makes an important distinction that is frequently missed in public discussions of AI: the gains come from moving tasks from H to M, not from eliminating jobs. The economic value is created by the difference between H(t_i) and M(t_i) for each automated task.

The Productivity Condition

Implication

Automation that removes judgment destroys value. Automation that removes non-judgmental execution time amplifies value. The sign of H(t_i) − M(t_i) is positive only when the human was adding less value to that task than the AI replaces — true for execution tasks, false for judgment tasks.

Economic consequence

The highest-ROI automation targets are high-δ, low-σ tasks: coordination, data retrieval, formatting, verification, status reporting, and scheduling. These have large H(t_i) values (expensive and time-consuming for humans) and can be reduced to near-zero M(t_i) costs. Decision ownership must remain in J_H.

5. Jobs Become Unstable Over Time — But Organizations Don’t Have to Be

Because r_i is time-varying — tasks that were in T_H at r_i = 0 migrate toward r_i = 1 as AI capabilities advance — the job model J(t) is itself time-varying. A job description written today describes a different task composition than the same job will have in 18 months.

The Time-Varying Job

Implication

A job is never finished being redesigned. Static job descriptions become obsolete artifacts the moment they are written. The job exists in time, and time changes r_i(t) continuously as AI capabilities expand.

Operating-model consequence

Organizations that can continuously re-partition tasks, re-train humans toward high-σ activities, and re-assign freed capacity to value expansion will compound productivity advantages. Those that redesign roles reactively, after the task migration has already happened, will consistently lag by 12–18 months.

6. Accountability Cannot Be Assigned to the System

The partition model makes a critical governance point explicit: J_A = M(T_A) handles execution, but J_H = H(T_H) retains accountability. Automation transfers execution; it does not transfer responsibility. This is not a philosophical position — it is a mathematical one. The human-required task set T_H contains, by definition, tasks for which human judgment and accountability are mandatory.

7. The Model Predicts Job Polarization

As r_i(t) increases across the economy, the distribution of σ in the remaining human task set shifts. Low-σ tasks disappear from J_H as they migrate to T_A. High-σ tasks remain. The result is a polarization of the labor market: low-judgment roles contract, high-judgment roles expand in influence and compensation, and mid-complexity roles face the sharpest pressure.

8. The Model Is a Design Tool, Not Just a Descriptive Model

Every model introduced in this article has a practical use case beyond description. The mathematics of work is not just a way to understand what is happening — it is a tool for deciding what should happen next.

• Automation readiness assessment: Score each task on r_i potential and σ_i level to identify the highest-value automation targets.
• AI impact simulation: Model the effect of moving a subset of tasks from T_H to T_A on total job duration, cost, and peak judgment requirement before deployment.
• Role redesign without layoffs: Identify the high-σ work that is currently buried in high-δ low-σ jobs, and redesign roles to concentrate human effort where it creates the most value.
• Governance mapping: Formally identify which tasks must remain in T_H for regulatory, ethical, or fiduciary reasons, and build those constraints into Ω.
• Investment underwriting: Use the task-level model to assess a target company’s automation potential, cost structure trajectory, and workforce resilience — translating workforce composition into financial projections.

The model also provides a shared language across operators, technologists, executives, regulators, and investors — enabling precise, evidence-based conversations about what automation is doing, what it cannot do, and what accountability obligations remain irreducibly human.

Summary: The Mathematics of Work at a Glance

The table below summarizes each model layer, its type, its core definition, and the key insight it delivers for managers and strategists.

What This Framework Changes

The Mathematics of Work is not a theoretical decoration. Like the Mathematics of Value, it is a practical tool for managers, strategists, and investors. When applied rigorously, it changes four things:

How you assess automation readiness. Task-level σ_i scoring provides a defensible, quantitative basis for prioritizing automation investments that is far more precise than function- or role-level analysis.

How you redesign roles. The partition model separates the tasks that must remain human from those that can migrate, enabling role redesign that preserves accountability while eliminating unnecessary human execution cost.

How you manage workforce transitions. By tracking the time-varying distribution of r_i(t), organizations can anticipate role changes before they occur – creating the runway needed for reskilling and redeployment rather than reacting to displacement after the fact.

How you make the governance argument. The constraint set Ω gives regulators, boards, and clients a formal basis for specifying which tasks must remain in T_H - turning a qualitative policy preference into a mathematically specified governance boundary.

The next article will bring the Mathematics of Work and the Mathematics of Value together — modeling the intersection where task automation on the supply side meets value component delivery on the demand side, and identifying where AI creates the most powerful simultaneous leverage across both dimensions.

‍