Most conversations about automation risk stay at the job-title level — “accountants are at risk,” “nurses are safe” — without accounting for the fact that two people with the same job title can have wildly different levels of exposure depending on what they actually do each day. A more useful unit of analysis is the task, not the role. The 5-step framework below is designed to be run by anyone on their own job, in about an hour, using no tools beyond a notebook or a spreadsheet.
| # | Step | What to produce |
|---|---|---|
| 1 | List every task | A raw inventory of your actual work |
| 2 | Rate repetitiveness | High / Medium / Low for each task |
| 3 | Rate judgment required | High / Medium / Low for each task |
| 4 | Rate data availability | High / Medium / Low for each task |
| 5 | Score and prioritize | A ranked list of tasks by automation exposure |
1. List Every Task You Actually Do
Start by capturing the real work, not the job description. Spend fifteen minutes listing every distinct task performed in a typical week or month. Include the small things: formatting reports, attending standups, approving invoices, answering the same three questions from colleagues repeatedly, and the harder-to-name work like “figuring out what the client actually wants” or “deciding whether this situation needs to escalate.”
Aim for 15 to 30 items. If the list is shorter, it’s likely that some tasks are being grouped at too high a level. “Communications” isn’t a task — “drafting the weekly project status email” is. The more granular the list, the more informative the audit.
This step also produces a useful side effect: most people realize their job contains more variation than they assumed, which itself is often a signal of lower average automation risk than the title suggests.
2. Rate Each Task for Repetitiveness
For each task, answer one question: does this task follow the same steps in the same order most of the time, or does it require adapting the approach based on context?
Highly repetitive tasks follow fixed procedures and produce consistent outputs. Data entry, invoice processing, standard report generation, and scheduling fit here. Low-repetitiveness tasks require reading a situation and deciding which approach applies. Handling a complaint that doesn’t fit any policy category, advising a colleague on an interpersonal issue, or recommending a strategy given conflicting information all sit at the low end.
Mark each task High, Medium, or Low. Don’t overthink it — the relative ranking matters more than perfect precision.
3. Rate Each Task for Judgment Required
Judgment, in this context, means the degree to which the task requires weighing trade-offs, tolerating ambiguity, or making a call where reasonable people could disagree. Tasks that are purely information-retrieval or execution require little judgment. Tasks that involve interpreting context, managing relationships, or making decisions without a clear precedent require a great deal of it.
High-judgment tasks are the ones where, if a wrong call were made, it would require a human to explain why. Low-judgment tasks are the ones where the right answer could be specified in advance as a rule. This is the dimension most relevant to automation exposure: current AI systems handle pattern-matching and rule-following well; they handle genuine trade-off reasoning poorly, particularly in novel situations.
4. Rate Each Task for Data Availability
Automation requires data. A task is easier to automate when the inputs are already structured and digital — a database record, a form submission, a transaction log. A task is harder to automate when the relevant inputs are unstructured, tacit, or social: what the client implied in conversation, what the team dynamic suggests about an approach, what institutional history a decision needs to account for.
For each task, rate how available and structured the inputs are: High (fully digital, structured, already in a system), Medium (partially structured, some interpretation required), or Low (primarily unstructured, tacit, or relational). Tasks rated High on this dimension are generally more automatable, all else being equal.
5. Score and Prioritize by Exposure
With three ratings for each task, a simple composite picture emerges. Tasks rated High-High-High (highly repetitive, low judgment, structured data) have the greatest automation exposure. Tasks rated Low-Low-Low have the least.
The point isn’t to calculate a precise score. It’s to identify which tasks are at the high-exposure end of the distribution and make a deliberate choice about them. Three options exist for any high-exposure task: accept that it may be automated and plan accordingly, work to incorporate more judgment or context into how the task is performed (making it harder to automate), or shift time and attention toward the lower-exposure tasks in the same role before the high-exposure ones disappear.
This is also a useful document to have in hand when thinking through broader strategies for AI-proofing an existing role — knowing which specific tasks are at risk, and being able to explain why, is a more productive starting point than general anxiety about the job title.
Doing this audit once is useful. Doing it again in 12 months, with fresh eyes and awareness of what new tools are available, is where it becomes genuinely valuable as a career practice. The distribution of task exposure isn’t fixed — it shifts as tools improve and as roles naturally evolve in response to what gets automated around them.