Exposure Score

An exposure score is a number that indicates how strongly something is touched by a change. “Exposure” here means roughly being subjected to something, i.e. the point of contact with a matter. In the context of computer programs that write texts or generate images on their own, the score usually measures: how many tasks of an occupation could these programs take over fully or partially? A translator would receive a high value here, a roofer a very low one. The meaning of this number is important: it describes being affected, not being doomed. A high value can just as easily mean that someone will work faster and earn more in the future.

Why investors and labor market researchers look at this number

Statements like “AI is changing the world of work” are too vague to base decisions on. Anyone valuing a company or planning education policy needs something comparable. A score turns a gut feeling into a quantity that can be sorted in tables and tracked over years. That is exactly why such values now show up in reports from banks, consulting firms, and government agencies.

On the stock market, the number is read in both directions. A company with a high exposure score is considered either at risk or a particularly big beneficiary – depending on whether it deploys the technology itself or is overtaken by it. A call center operator and a provider of speech software can have the same value and yet face completely opposite prospects. The score alone therefore makes no forecast about the share price.

A typical misconception is to read the value as a layoff rate. A widely cited study found that around 80 percent of US employees could hand off at least a tenth of their tasks to language models. In headlines, this quickly became “80 percent of jobs threatened.” What was actually meant was something quite different, namely a touchpoint with individual activities.

From the list of activities to the single number

The calculation almost always begins with a breakdown. An occupation is not assessed as a whole but split into individual activities. For a tax advisor, these would be, for example: sorting receipts, looking up laws, advising clients, filling out forms. Such lists already exist in large occupational databases; in the US, for instance, in a collection called O*NET covering more than a thousand occupations.

In the second step, someone assesses each individual activity. In the past this was done exclusively by experts; today a language model is often additionally asked to judge, and both results are compared. The guiding question is: could this activity be completed significantly faster using the technology without the quality suffering? In the end, the individual judgments are weighted and summed up, usually according to how much work time is spent on each activity.

The result is a number, often between 0 and 1 or expressed as a percentage. It should be treated with caution. The weightings are assumptions, and whether a company actually adopts the technology at all is not captured in the score at all. Related but not identical is the term automation potential: it refers to the complete takeover of a task, whereas exposure also counts mere support.

Where you run into such values in everyday life

You most often encounter exposure scores in news about the labor market. When a chart ranks occupations by AI exposure, there is almost always such a calculation behind it. Career counseling services and study-choice portals now also work with such rankings. It is almost always worth taking a look at the methodology at the end of the report.

The term is also used in entirely different fields. In IT security, an exposure score measures how many vulnerable points a company network has. In finance, exposure describes how much money is tied up in a particular risk, for instance in a single industry. The basic idea is the same everywhere: making being affected measurable. The calculation method, however, differs completely, which is why it is always worth asking which exposure is meant.