MONSa Skills Auditor for the AI era

How we score

Where the numbers come from.

Every job-role figure on this site is built the same way, from evidence about the role’s own tasks. We show ranges rather than exact percentages, because these are estimates.

  1. 01

    Start from the job’s tasks.

    Each of the 2,001 job roles comes with its key tasks — about 19 per role — from Singapore’s national Skills Framework.

  2. 02

    Match each task to research.

    Every key task is matched by meaning to the closest of about 18,000 task descriptions in O*NET, the US occupational database that AI-exposure research is built on. 98% of tasks find a close match.

  3. 03

    Ask how far AI reaches it.

    For each matched task we combine two sources: expert and model ratings of whether a language model can halve the time it takes (Eloundou et al.), and whether people actually use AI for it (the Anthropic Economic Index). Where AI mostly does the task, it counts as AI can do this today; where people use it to assist and check, AI does it, you check.

  4. 04

    Separate out hands-on work.

    Physical and in-person work that today’s AI can’t do is shown as Hands-on. It comes from how much of each required competency happens away from a computer, and how physical the role’s tasks are. The rest of the role is Compounds: judgment, relationships and decisions.

  5. 05

    Give the type a confidence.

    We re-score every role 50 times, each time leaving out a random fifth of its tasks and competencies. If its type holds in at least 80% of runs we say it leans that way; otherwise it sits between two types.

How we tested it

We set the pass marks before running any test, and kept the test data out of the scores. The ranking tests use rank correlation: 1 means the same order, 0 means no relation.

TestResultPass mark
Role ranking vs the ILO’s 2025 index of exposure to generative AI 0.70 at least 0.5 Pass
Sector ranking vs the same ILO index (35 sectors) 0.88 at least 0.6 Pass
Type changes when a fifth of a role’s data is left out 7.4% of roles at most 10% Pass
Singapore practitioners agree on the type (about 100 roles, 2–3 raters each) In progress at least 70% Pending
Practitioners’ split of the work vs ours In progress within 10 points Pending

Scores version evidence-v1-bb196320f7, built 4 October 2026. Our earlier scores, built from AI ratings of competencies alone, ranked jobs against where AI is actually used. That is why we rebuilt them on task evidence in October 2026. Until the practitioner tests pass, figures are shown as ranges.

What it can’t tell you

  • It reads the role as defined in the Skills Framework, not your actual week. The diagnostic covers that.
  • No role comes out as mostly automatable: on current evidence, AI can do this today stays below about 30% for almost every role. Exposure to AI is not the same as losing a job.
  • Seniority makes less difference than you might expect: senior and junior roles score much the same on AI can do this today.
  • Matching Singapore tasks to US task descriptions is approximate, and some of the research we use is itself partly rated by AI models.
  • Usage data comes from one AI assistant and leans towards desk work.
  • Per-competency splits, and 18 roles with too few matched tasks, are AI-rated estimates rather than evidence.
  • About half of roles sit between two types. Small changes in the data can tip them either way.
  • AI improves every few months, so we re-score as it changes. Each version has its own number, and a cohort is measured on one fixed version from start to finish.

Sources and credits

  • Anthropic Economic Index. Anthropic. Task-level AI usage, automation vs augmentation, and robot-exposure data. Licensed under CC BY 4.0. Licence.
  • Eloundou, Manning, Mishkin & Rock (2023), “GPTs are GPTs”. Task-level exposure ratings. Data under the MIT licence.
  • O*NET task statements. This site includes information from the O*NET database by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA), used under the CC BY 4.0 licence. O*NET® is a trademark of USDOL/ETA. We have modified some of this information; USDOL/ETA has not approved, endorsed or tested these modifications. Licence.
  • Skills Framework. Skills and Workforce Development Agency (SWDA), formerly SkillsFuture Singapore. Job roles, key tasks and required competencies, from the Skills Framework dataset (Q3 2026).

Not used to build the scores. Used to test them, and the ILO index is also shown beside each role as an outside reference:

  • Gmyrek, P., Berg, J., Kamiński, K., Konopczyński, F., Ładna, A., Nafradi, B., Rosłaniec, K., Troszyński, M. (2025), Generative AI and Jobs: A Refined Global Index of Occupational Exposure, ILO Working Paper 140, Geneva: International Labour Office. © ILO, licensed under CC BY 4.0. Also shown beside each job role as an outside reference, adapted by matching the role’s key tasks to ILO task statements. This is an adaptation of a copyrighted work of the International Labour Organization (ILO). This adaptation has not been prepared, reviewed or endorsed by the ILO and should not be considered an official ILO adaptation. The ILO disclaims all responsibility for its content and accuracy. Responsibility rests solely with the author(s) of the adaptation.
  • Felten, Raj & Seamans (2021), Occupational, industry, and geographic exposure to artificial intelligence, Strategic Management Journal 42(12).