How our automation-risk scores are calculated

What does the score mean?

A low score means the occupation appears difficult to replace as a whole human role. A high score means many core parts of the role look structured, repeatable, software-driven, physically predictable, or otherwise suitable for automation.

For example, an occupation with a score of 10% is likely to be much less exposed to replacement than an occupation with a score of 90%. The number should be treated as a useful signal, not an exact forecast.

How do we calculate the score?

We start with occupations where the replacement-risk judgement is relatively clear. These are labelled conservatively:

• Low replacement risk: occupations that clearly still require a human role.
• High replacement risk: occupations that appear clearly replaceable end-to-end.
• Excluded: occupations that are too mixed, ambiguous, broad, or only partially automatable.

We then train a machine-learning model using occupation data such as abilities, skills, knowledge, work activities, and work context. The model learns which patterns are associated with higher or lower replacement risk, then applies those patterns across occupations with enough source data.

User votes are also used as a sense-check. They help us spot questionable results and review assumptions, but they do not simply decide the final score.

What data do we use?

We use data from O*NET, which describes the abilities, skills, knowledge, activities, and work contexts associated with occupations.

Some examples of signals used or reviewed include:

The base scoring model currently focuses on O*NET level and work-context ratings. These help describe the capability required to perform the work, and the setting in which the work happens.

Importance ratings are also useful, but mainly for public explanation. For example, they can help show which human strengths are important in a job, even when those signals are not the only inputs into the automation score.

What makes an occupation harder to replace?

Occupations are generally harder to replace end-to-end when they depend on human qualities that are difficult to reduce to a repeatable process.

These factors do not guarantee safety. AI, software, and robotics can still change parts of the work. The question we are trying to answer is whether the occupation itself is likely to remain a distinct human role.

What makes an occupation higher risk?

Higher-risk occupations often contain core work that is easier to specify, repeat, measure, or hand over to software or machines.

• Tasks are predictable and repeated in similar ways.
• Inputs and outputs are structured or digital.
• The work follows rules, scripts, forms, or standard procedures.
• There is limited need for human judgement, trust, creativity, or care.
• The physical environment is controlled enough for machines or robotics to operate reliably.

My occupation has a high score. Should I be worried?

A high score is worth taking seriously, but it should not be treated as a reason to panic. Automation risk depends on your specific role, employer, industry, location, and the direction your career is moving in.

The most useful response is to look for practical next steps:

• Compare safer related careers that use some of your existing skills.
• Move towards more human-centred responsibilities, such as judgement, communication, leadership, care, or client trust.
• Learn to work with AI and automation rather than competing with it directly.
• Broaden your skillset by combining technical confidence, communication, self-direction, and practical industry knowledge.
• Build specialist or practical expertise that is harder to replace end-to-end.
• Watch how your industry is changing so you can adapt before change is forced on you.

The goal of this site is not just to show a risk score. It is to help you move from worry to a clearer career decision.

The occupation I am interested in is not listed. Can you calculate it?

Usually not. We can only score occupations where there is enough reliable source data. If an occupation is missing, it may be too new, too niche, too broad, or not separately covered in the O*NET data we use.

Why have some scores changed over time?

The website originally showed predictions from the 2013 Oxford Martin study, The Future of Employment. The current scores are calculated in-house and use a more recent methodology.

Scores can change for several reasons:

• O*NET updates its occupation data over time.
• Occupations themselves change as tools, workflows, and expectations change.
• Our modelling has moved towards role-level replacement risk rather than simple task automation.
• Some older assumptions have been reviewed, adjusted, or excluded where the evidence is mixed.
• User votes and public feedback help us identify results that deserve closer review.

This means older and newer scores may differ. We think that is better than pretending the world of work, AI, and robotics is standing still.