Five Things Your AI Job Risk Score Cannot See

The short answer: the score measures how your reported week distributes across tasks the research rates as exposed. It does not know whether your employer has bought anything, what your regulator requires, what your quality bar is, what your labour market is doing, or what your employer actually values about you. Everything downstream of the measurement is judgement, yours and mine.

It does not know whether your employer has bought anything

Task exposure is a property of the work. Adoption is a property of an organisation, and the gap between them is measured in procurement cycles, security reviews, pilot programmes, and years. A task can be fully exposed and still done by hand in your building for the rest of the decade because nobody has signed a purchase order. The score cannot see the purchase order.

It does not know your regulatory environment

In some fields a task can be entirely automatable and still legally require a licensed human to perform or sign it. Audit opinions, prescriptions, engineering approvals, court filings. The accounting projections show how much that requirement is worth: it is the difference between an occupation growing and an adjacent one shrinking. The score does not know which side of a licensing line your tasks sit on.

It does not know your quality bar

The ratings ask whether output holds up. They cannot ask whether it holds up for your clients, at the standard you are held to, with the consequences you carry when it is wrong. A first draft that is good enough for an internal memo may be nowhere near good enough for a filing. The score rates the task; it cannot rate your firm.

It does not know your labour market

An occupation can shed tasks and add headcount at the same time if demand grows faster than the tasks shrink. BLS projects software developer employment up 15% from 2024 to 2034, in an occupation carrying some of the highest measured task exposure in the research. A high score in a growing occupation and the same score in a shrinking one are different situations, and the score alone cannot tell them apart. That lookup is free at BLS, and I recommend doing it next to your score.

It does not know what your employer values about you

What your employer actually values is frequently not on the task list at all. The relationship with the account that renews every year. The institutional memory. The person the team goes to when something breaks. A great deal of the reason a specific person is employed is invisible to a catalogue of tasks, and the score is built on the catalogue.

Nor does it know how badly this genre has aged before

Frey and Osborne put 47% of US employment at high risk of computerisation in 2013, over perhaps a decade or two. Most of that window has passed. Their unit was the occupation, which is the mistake this tool exists to avoid, but the caution generalises. Confident estimates about automation have a track record, and it is not a good one. That is a reason to publish the limits of a measurement alongside the measurement, which is what this page is.

Why a smaller claim

This is a smaller claim than most tools in this category make. I think it is the only one the data supports. A tool that reported a probability of job loss would be easier to headline and easier to sell, and it would be a number I made up. So the score reports the composition of your week, the methodology shows how, and the five gaps above are where your own judgement has to take over.

If you find yourself wanting the tool to know one of these five things, that is a useful signal. It tells you which gap matters most in your situation, and that is the one to go and check yourself.

Frequently asked questions

Does the AI Job Risk Score predict whether I will lose my job?

No. It measures how your reported working hours distribute across tasks the research rates as exposed. Adoption, regulation, quality standards, labour demand, and what your employer values are all outside the measurement.

Should I check my score against anything else?

Yes. Read it next to your occupation's BLS employment projection. A high score in a growing occupation and the same score in a shrinking one are different situations, and the projection is free to look up.

Why publish the limitations of the score?

Because confident automation estimates have aged badly before. Frey and Osborne's 2013 figure of 47% at high risk has mostly not materialised over its stated window. Publishing what the score cannot see keeps it from being read as a forecast.