Map tasks before deciding what to automate

What it is

Before you decide what to automate, list the tasks in a role or process and judge each one. A job is a bundle of tasks. AI can do some of them, help with others, and is no use for the rest. “Exposure” is the research term for how much of a job AI could, in principle, do.

What the evidence shows

• Few jobs can be fully automated. A study that scored 18,156 tasks found that most occupations contain some tasks machine learning could do, and few could be fully automated. Getting the benefit usually requires redesigning the job.
• Assist, not replace, is the common case. A global estimate by the International Labour Organization predicts that generative AI will mostly assist people in their jobs. Clerical work is the most exposed.
• Exposure does not predict job losses. Across 23 countries, occupations more exposed to AI did not lose employment overall. Where computer use was high, more exposure went with faster job growth. That study used data from before today’s AI chat tools. Since 2022 the evidence is disputed. In US payroll records, employment of 22–25-year-olds in the most exposed occupations is about 19% below trend. The Yale Budget Lab’s tracker finds no clear disruption across the US economy.
• Exposure is not use. Exposure scores show what AI could do, not what it is used for. A recent working paper makes the point directly.

Public scores exist for whole occupations, such as the AI Occupational Exposure measure. They describe a typical job. Your accountant or planner may do different tasks.

How to do it

1. Pick one role or one process to start.
2. List its tasks with the people who do them. Job descriptions are usually out of date. Ask each person which tasks they want automated, and why. In a study where 1,500 US workers rated 844 tasks, they favoured automating 46.1% of them. Most often they wanted AI as an equal partner. On 47.5% of tasks they wanted more human involvement than AI experts judged necessary. The study is a preprint.
3. For each task, note how often it is done, how long it takes and what goes wrong when it is done badly.
4. Judge separately: could AI do it? Could your company put that in place and maintain it? Does the task need a person’s judgement, relationships or responsibility?
5. Mark each task: automate, assist, or leave alone.
6. Check the “automate” list for tasks that juniors learn on.
7. Before you cut roles, check the traces of how the work is really done: overtime, queue lengths, and how often team leaders or others step in. See the box below.
8. Start with a small number of tasks and measure.

Traces to check before you cut roles

At Commonwealth Bank of Australia, the bank said 45 customer-service roles would go and be replaced by an AI voice bot. The union pointed to rising call volumes, overtime offers and team leaders answering calls. The bank then said its assessment had been an “error” and that the roles were not redundant. Overtime, queue and escalation records show whether the work has really gone. Check them before you remove a role.

What to measure

• The share of working time in each group: automate, assist, leave alone.
• For the tasks you change, time and quality before and after.

Where it fails

• Reading the score as a forecast. A high exposure score does not mean a job will go. Treating it that way alarms staff for no reason.
• Mapping forever. A map is useful when it leads to a first change within weeks.
• Stale maps. What AI can do changes quickly. Redo the judgement when the tools change.

By company size

• Micro and small: an afternoon with a whiteboard covers a whole role.
• Medium and larger: map one process end to end before mapping the whole company. Public exposure scores can help you choose where to start.

Assessments for this practice

These are existing questionnaires, tests and guides you can use. Each card gives the question it answers, who answers it and whether you can use it freely. All assessments.

• Which of our roles could AI touch most? AI Occupational Exposure measure (AIOE) Nobody is surveyed · You match your job titles to United States occupation codes Free to use, ask before changing Testing: Moderate Task and job maps
• In which roles could generative AI take over tasks, and in which could it assist? ILO index of occupational exposure to generative AI Nobody is surveyed · You match your job titles to a standard list of occupations Free to use and adapt Testing: Emerging Task and job maps
• Which of our own tasks could AI do or help with? TPAH task analysis Nobody is surveyed. Process owners check the results · An analysis from a description of the company and its roles Paid service Testing: Emerging Task and job maps

See it applied

• Worked example 375 task families mapped in a 118-person company 31 roles broken into tasks and scored on four separate scales, with the tasks to hand to AI, the tasks where AI helps a person, and the tasks to leave alone.

Company cases

Companies that did this

• Klarna Klarna moved named tasks to AI (translation, image production, routine copy), not marketing as a whole. Company-reported result
• META-Regalbau The steps of the picking process and the routes staff walked were recorded and analysed before the new warehouse layout was fixed. Company-reported result

What happened without it

• Commonwealth Bank of Australia The bank later said its assessment “did not adequately consider all relevant business considerations”. Partly walked back

Sources

1. Brynjolfsson, Mitchell and Rock (2018). What Can Machines Learn and What Does It Mean for Occupations and the Economy?. AEA Papers and Proceedings, 108, 43–47. What this study did and found
2. Felten, Raj and Seamans (2021). Occupational, industry, and geographic exposure to artificial intelligence: A novel dataset and its potential uses. Strategic Management Journal, 42(12), 2195–2217. What this study did and found
3. Gmyrek, Berg and Bescond (2023). Generative AI and jobs: A global analysis of potential effects on job quantity and quality. International Labour Organization. What this study did and found
4. Georgieff and Hyee (2022). Artificial Intelligence and Employment: New Cross-Country Evidence. Frontiers in Artificial Intelligence, 5, 832736. What this study did and found
5. Sedai (2026). Exposure Is Not Adoption: Measuring Generative AI and Labor Demand. SSRN working paper. What this study did and found
6. Brynjolfsson, Chandar and Chen (2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab working paper (revised 12 August 2026). Limits or contradicts this practice. What this study did and found
7. Shao, Zope, Jiang, Pei, Nguyen, Brynjolfsson and Yang (2026). Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce. arXiv 2506.06576. Preprint, not peer-reviewed. arxiv.org
8. The Budget Lab at Yale (updated September 2026). Tracking the Impact of AI on the Labor Market. budgetlab.yale.edu
9. Information Age (21 August 2025). CBA reverses AI-driven job cuts, admits error. ia.acs.org.au

Related practices

• For each kind of task, test whether a person with AI beats either one alone Do not assume that adding a person to AI, or AI to a person, gives the best result. Test it for each type of task before you settle the process. Do next · Effort: medium (weeks, a few people) Evidence: Strong Context-dependent Process redesign · Redesign
• Keep the entry-level tasks that newcomers learn on Before you automate routine work, check whether it is how juniors learn the job. If it is, keep some of it for them, or replace the learning another way. Do later · Effort: medium (weeks, a few people) Evidence: Emerging Contested People and skills · Run
• Redesign the workflow instead of adding a tool to it Giving people an AI tool inside an unchanged process changes little. Decide again who does each step, in what order, and where a person checks. Do next · Effort: high (months, several teams) Evidence: Moderate Process redesign · Redesign