How to Analyze Your Job With AI: A Task-Level Template (Updated 2026)

What is a task-level AI job analysis?

A task-level AI job analysis breaks a role into recurring activities, then evaluates each activity for AI assistance, human review, accountability, and career value. It is more useful than assigning one permanent risk percentage to a job title because people with the same title often perform different work.

The International Labour Organization's 2025 exposure index uses task-level data because AI affects parts of occupations unevenly. The OECD's 2026 exposure measure reaches a related conclusion: current systems are closer to routine information processing and codifiable work than to contextual judgment, interpersonal understanding, complex decisions, and responsibility.

This guide turns that research logic into a worksheet you can complete without sharing confidential employer data.

The aim is not to maximize automation. It is to find where assistance improves the work without hiding review effort, errors, or responsibility.

The seven-step job analysis

1. Record one representative week

List what you actually did, not what your job description says. Use calendar entries, project notes, tickets, and deliverables to reconstruct five working days. Group small actions into meaningful tasks such as "prepare weekly client report" or "review supplier exception."

For each task, estimate:

hours or percentage of the week;
frequency;
main input and output;
who checks or uses the result;
what happens if it is wrong.

Do not paste customer records, private documents, personal data, credentials, or proprietary material into a public AI tool.

2. Separate production from decisions

A task often contains both repeatable production and higher-stakes judgment. "Prepare a forecast" may include gathering data, cleaning a spreadsheet, selecting assumptions, investigating anomalies, explaining uncertainty, and approving a recommendation.

Split the task until each line has one clear outcome. This prevents an AI-generated draft from being confused with ownership of the final decision.

3. Score the structure of each task

Use a simple 0 to 3 scale for five dimensions:

| Dimension | 0 | 1 | 2 | 3 | |---|---|---|---|---| | Repetition | New each time | Some repeated steps | Mostly repeatable | Highly standardized | | Digital inputs | Mainly physical or tacit | Mixed inputs | Mostly digital | Fully digital and structured | | Verifiability | Expert judgment needed | Costly to check | Clear review criteria | Automatically testable | | Context and exceptions | Constantly changing | Many exceptions | Some exceptions | Stable rules | | Consequence of error | Minimal | Rework required | Customer or financial effect | Legal, safety, or major business effect |

High repetition and digital-input scores suggest technical feasibility for assistance. High consequence, unstable context, or difficult verification increase the need for qualified human review. Do not combine the five numbers into a false scientific probability.

4. Test one low-risk slice

Choose a reversible task with non-sensitive inputs and an output that can be checked. Examples include outlining a meeting agenda, classifying public feedback, generating spreadsheet formulas from dummy data, or drafting questions for a document review.

Define the test before using AI:

expected output;
source of truth;
acceptance criteria;
prohibited data;
reviewer;
stop condition.

NIST's AI Risk Management Framework recommends documented testing, evaluation, verification, and validation throughout an AI system's lifecycle. A workplace experiment should follow the same basic discipline: test in context, record failures, and keep a human owner.

5. Measure the whole workflow

Saving ten minutes on a draft is not a useful result if review takes twenty extra minutes. Compare the old and new workflow using measures that fit the task:

total cycle time;
review time;
error and rework count;
missed exceptions;
customer or colleague acceptance;
cost of recovery when the output fails.

Run several comparable examples. Record where AI helped, where it added work, and which cases should remain outside the workflow.

6. Classify the next action

Place each task in one of four practical categories:

| Category | Meaning | Next action | |---|---|---| | Automate carefully | Repeatable, testable, low consequence | Build a controlled workflow and monitor it | | Assist with review | AI can draft or classify, but context matters | Keep an expert review and escalation rule | | Redesign the role | Execution shrinks but the outcome remains valuable | Move time toward diagnosis, decisions, and coordination | | Keep human-led | Trust, accountability, physical work, or high-stakes exceptions dominate | Use AI only for bounded support, if appropriate |

The category may change as tools, regulation, and organizational controls change. Review important tasks every quarter or after a major workflow change.

7. Turn the result into evidence

Choose one skill that improves the redesigned workflow. Build a work sample that shows:

the original problem and baseline;
the task you changed;
the controls and review process;
the result;
known limitations;
what you would test next.

This is stronger career evidence than saying you are "good at AI." It demonstrates judgment, verification, domain knowledge, and responsibility.

A worked example: weekly marketing performance report

This example is hypothetical. It illustrates the method and is not a Jobisque customer result.

An analyst spends four hours collecting channel data, correcting labels, drafting commentary, checking anomalies, and presenting recommendations. Data collection and first-draft commentary are repeatable and digital. Explaining a sudden conversion drop is less standardized because tracking changes, promotions, seasonality, or data quality may be involved.

The analyst tests AI only on a sanitized copy of the table. Acceptance criteria require every numerical statement to match the source data and every causal explanation to be labeled as a hypothesis. The first test saves drafting time but invents a cause for one anomaly. The workflow is therefore classified as "assist with review," not fully automated.

The role redesign is specific: automate formatting, use AI to propose questions, and spend more time validating data, investigating causes, and explaining trade-offs to decision-makers. The evidence is a before-and-after workflow with error logs and review time, not an unsupported productivity claim.

Copyable task-audit worksheet

Use one row per task:

| Field | Your notes | |---|---| | Task and desired outcome | | | Weekly time or frequency | | | Input and output | | | Repetition score (0-3) | | | Digital-input score (0-3) | | | Verifiability score (0-3) | | | Context and exception score (0-3) | | | Consequence of error score (0-3) | | | Current reviewer or decision owner | | | Safe test and acceptance criteria | | | Result: automate, assist, redesign, or human-led | | | Skill and work sample to build | |

For a quick directional baseline, use the task-level AI risk calculator. The free Jobisque audit can create a fuller task map for your role.

Mistakes that make a job analysis unreliable

Scoring only the title: it hides differences in seniority, industry, tools, and responsibility.
Treating exposure as replacement: technical capability does not determine adoption, staffing, regulation, or timing.
Testing with confidential data: a convenience experiment can create privacy, contractual, or security risk.
Measuring draft speed only: review, correction, coordination, and failure recovery belong in the calculation.
Ignoring the decision owner: someone remains accountable for using or rejecting the output.
Claiming certainty from a small test: a few examples reveal failure modes; they do not prove universal performance.

What this analysis cannot predict

This method cannot tell you whether or when an employer will eliminate a position. It does not observe local labor demand, company strategy, regulation, budgets, model changes, or your performance. It is a planning tool for identifying controlled experiments and defensible skills.

If you want a prioritized 90-day plan after the audit, the Jobisque Career Playbook turns task findings into actions. It is a one-time $19 offer, not a promise of employment or income.

How to Analyze Your Job With AI: A Task-Level Template