AI Disruption Index: how the score is calculated

The Nestorbot AI Disruption Index gives each of the 2,982 occupations in the European ESCO classification a score from 0 to 100. This page explains where the score comes from, what it tells you and what it does not, and how to cite it.

Method version 1 · page last updated 19 November 2026

What the score measures

The score estimates how much of an occupation's work, described through its skills, AI tools and automation could take over. 0 means almost none, 100 means almost all.

It describes the occupation as a whole, as ESCO defines it. It is not a forecast of job losses, not a prediction for a specific person, employer or country, and not a probability.

Data we use

  • ESCO version 1.2.1, the European Commission's classification of occupations and skills: 2,982 occupations and 13,559 skills and knowledge items.
  • 120,086 links between occupations and skills, each marked as essential (64,308) or optional (55,778) for the occupation.
  • The ISCO-08 code of each occupation, as given in ESCO.

Vacancies, wages and how fast employers adopt AI are not part of the score.

Step 1: two ratings for every skill

Every skill received two ratings on a scale from 0 to 1:

  • Automation exposure: how likely AI or robots are to perform the skill at a human level within 3 to 5 years.
  • AI complementarity: how much AI tools help a person who has the skill to do more or better work.

13,211 skills (97%) were rated on 16–17 March 2026 by a language model, Claude Haiku 4.5 from Anthropic. For each skill it saw the name, the skill type and the start of the ESCO description, together with written guidelines and examples: for instance, routine processing of text or data rates high, while physical work in unpredictable places and hands-on care rate low.

The other 348 skills received fixed values from simple rules based on the skill type and keywords in the name (for example, exposure 0.45 for knowledge items).

Each skill was also placed in one of four groups: high risk (exposure 0.60 or more), moderate risk (0.35 to 0.59), low risk (below 0.35) or AI-enhanced (exposure below 0.45 and complementarity 0.55 or more).

Step 2: from skills to an occupation

For each occupation we combine the ratings of all its skills into three numbers from 0 to 100. They are shown on every AI-risk page:

  • Skill vulnerability: the average automation exposure of the occupation's skills, where an essential skill counts 1.5 times and an optional skill 0.7 times.
  • Task automation: the share of the occupation's skills in the high-risk group, plus half of the share in the moderate-risk group.
  • AI enhancement: the sum of the complementarity ratings above 0.5, divided by the number of all the occupation's skills.

Base score = 0.5 × skill vulnerability + 0.3 × task automation − 0.2 × AI enhancement

The result is rounded and kept between 0 and 100. AI enhancement lowers the score: where AI mainly helps people work, the work is less likely to be taken over.

Step 3: calibration

On 17 March 2026 the same language model reviewed the base scores in small batches. It saw only the occupation names and their scores, with guidelines describing typical levels (for example, 55 or more for work that is mainly processing text or data, 15 to 35 for physical work in unpredictable environments) and a target spread with a median of about 50. Where it judged a score to be more than 8 points off, it proposed a new one, and that became the final score.

This step changed 2,411 of the 2,982 scores (81%), mostly upwards: the median change was 11 points and the average score rose from 29 to 45. The final score therefore combines the skill formula with the model's judgement, and it can differ from what the three numbers on the page would give on their own.

We describe this openly because it affects how the score should be read: it is better at comparing occupations with each other than at measuring an absolute level of risk.

Risk levels

ScoreLevelOccupations
0–14Very low243
15–34Low826
35–54Moderate890
55–74High706
75–100Very high317

The median score across all occupations is 44.

What else an AI-risk page shows

  • Most exposed skills: the five skills of the occupation with the highest automation exposure.
  • Most resilient skills: the five with the lowest automation exposure.
  • Skills AI enhances: up to five skills with the highest complementarity above 0.5.
  • Safer related occupations and upgrade paths: occupations from the same ISCO group with a clearly lower score, chosen when the scores were computed.
  • Occupations with a similar score: from the same ISCO group, within 5 points.
  • The descriptive texts were written on 28 March 2026 by the same language model, based on the score and the skill lists.

Limitations

  • The ratings are estimates made by a language model following our guidelines. They have not been checked against a study of real job tasks or labour-market outcomes.
  • The score uses ESCO's description of an occupation. Real jobs with the same title differ by employer, country and seniority.
  • It does not take into account wages, costs, regulation, how quickly employers adopt AI or how much demand there is for the work.
  • Calibration pulled the scores towards a chosen spread, so differences between occupations say more than any single number.
  • The scores are a snapshot from March 2026, and AI tools change quickly.

Updates

All scores were computed on 17 March 2026 and have not been recomputed since. When we recompute them, for example after a new ESCO version or a change in the method, this page will show the date and what changed.

How to cite

Nestorbot (2026). AI Disruption Index, method version 1 (scores computed 17 March 2026). https://www.nestorbot.com/disruption/methodology

When you quote the score of a single occupation, link to its AI-risk page, for example https://www.nestorbot.com/disruption/jeweller.

Occupation and skill data: ESCO classification v1.2.1, © European Union, https://esco.ec.europa.eu

Check the AI risk of your occupation →