Every leader wants an answer to a simple question: which jobs will AI affect? The honest answer is that no model can provide certainty. Technology capability, adoption, regulation, economics, data access, process design, and employee trust all shape what happens in practice.

That uncertainty is not a reason to avoid analysis. It is a reason to analyse at the right level and communicate results correctly.

AI exposure is best treated as a property of an activity, not a prediction that a job will disappear. It asks: given the nature of this activity and the capabilities of available technology, how plausible is meaningful automation or augmentation?

A disciplined assessment considers several drivers:

• Structure: Are inputs and outputs standardised or highly variable?
• Repeatability: Does the activity recur in a consistent form and volume?
• Rules and exceptions: Can the decision logic be made explicit, or are exceptions central?
• Information type: Is the work based on text, images, structured data, conversations, or physical context?
• Judgment and accountability: Does the activity require professional interpretation, ethical trade-offs, or formal authority?
• Data and systems: Are the necessary data accessible, lawful to use, and of sufficient quality?

These factors should generate two related but distinct assessments: automation potential and augmentation potential. A high score in one does not automatically imply a high score in the other.

For example, structured invoice matching may be strongly automatable. Executive briefing preparation may be difficult to automate end-to-end because context and accountability matter, yet highly augmentable through research synthesis and first-draft generation. Sensitive performance feedback may be technically assistable, but its relational and ethical dimensions make any score only one input to a careful design decision.

Scores should be interpretable. A 0–100 scale can work well if anchored with examples and if the score is accompanied by a reason. More importantly, the score should never stand alone. Leaders also need to see:

• How much time the activity represents.
• Whether it is core or peripheral to the role’s value.
• The confidence of the assessment.
• The conditions required for impact, such as data access or tool adoption.
• The likely direction of change: automate, augment, redesign, or monitor.

This avoids a common mistake: ranking job titles by a single exposure number and treating the result as a redundancy list. The meaningful analysis happens in the distribution. Is exposure concentrated in a small amount of peripheral administrative work? Is it spread across many cognitive activities? Does AI change the work’s quality, speed, autonomy, or risk profile? These are much more useful questions for workforce planning.

Good exposure assessment is therefore a decision-support tool, not an oracle. It helps organisations prioritise where to investigate, pilot, redesign, and develop skills. It should be regularly updated as technology and work practices change.

**Next article:** A score becomes useful only when it can be translated into a transparent estimate of possible value without turning optimism into false precision.