What Is a Task Map?

A task map breaks work into the activities that make it up. That helps us ask where AI might matter without pretending an entire occupation is one indivisible thing.

The short answer

A task map is a structured description of the tasks connected to an occupation, workflow, activity, or output. It gives us a more useful unit of analysis than a job title on its own.

A job may include producing an output, checking work, handling exceptions, communicating with people, using software, and taking responsibility for a decision. A task map makes those parts visible.

It does not by itself predict automation, adoption, productivity, job loss, wages, or causation. It is the beginning of the investigation, not the conclusion.

Why tasks matter

O*NET defines occupation-specific tasks as activities and duties performed in a job, and distinguishes them from broader work activities. Its database releases are versioned, so the source and date matter when comparing task descriptions.

Task-based research on automation uses the same basic discipline: analyse the activities inside an occupation instead of assuming every worker and workplace is affected in the same way. That is why a claim about an occupation should be traceable to the tasks behind it, not just to a percentage attached to a job title.

A task map can show where a question belongs. For example, an AI system might be relevant to drafting a routine document but not to deciding how an unusual case should be handled. That example illustrates the distinction; it is not a measured result about any particular occupation.

A task list is not a workflow map

A simple task inventory might list preparing records, answering questions, checking results, and coordinating with other people. That tells us what kinds of activities are associated with the work.

A richer workflow map would also show sequence, dependencies, actors, tools, inputs, outputs, exceptions, and review points. It might show that one person prepares an item, another checks it, an exception sends it back for correction, and a final decision requires human approval.

A task map can include those details, but the label alone does not guarantee that it does. The exact fields depend on how the map was designed.

What it can and cannot tell you

A task map can help identify:

• which activities are being discussed;
• whether the question concerns an occupation, workflow, activity, or output;
• where a specific AI capability might be relevant;
• what evidence needs to be gathered next.

It does not necessarily tell you:

• how often a task occurs or how much time it takes;
• the order of tasks or their dependencies;
• which tools, inputs, outputs, or people are involved;
• whether an AI system performs the task reliably;
• whether an organisation has adopted it;
• whether productivity, staffing, wages, or employment changed;
• whether AI caused any observed change.

Occupational tasks also vary across employers, workers, locations, and technologies. A task map can simplify that variation. That is useful, provided nobody mistakes the map for the territory. The territory is where the missing data usually lives.

A five-question reading test

When someone says an occupation is “exposed” to AI, ask:

1. Which tasks? Is the claim about named activities or only a broad job title?
2. Which setting? For whom, where, and under what workflow conditions?
3. Which details? Does the map include time share, sequence, dependencies, or only a list?
4. Which date and method? What source version produced it?
5. Which outcome? Is the conclusion about technical potential, actual use, adoption, productivity, employment, or causation?

If those questions are unanswered, treat the task map as a starting point rather than a forecast.

For a related explanation of why categories and measurements matter, see What Is AI Exposure? and What Is Jevons’ Paradox?.

Bottom line

A task map breaks work into analysable pieces. That is valuable because occupations are bundles of tasks, not uniform blocks. But a task map does not tell us what technology, employers, or labour markets will do next. For that, we need separate evidence about capability, reliability, adoption, time, cost, quality, and consequences.

Sources

• ONET Resource Center, “The ONET Content Model”.
• O*NET Database Releases Archive.
• Arntz, Gregory, and Zierahn, “The Risk of Automation for Jobs in OECD Countries”.
• Acemoglu and Restrepo, “Automation and New Tasks”.

Research cutoff: 2026-08-03. Last reviewed: 2026-08-03. Update trigger: Material changes to the cited task frameworks or authoritative evidence about task decomposition and AI outcomes.