AI product innovation and occupational exposure: automation and augmentation in commercial AI deployments
Published in Industry and Innovation, 2026
This paper derives a measure of the degree to which each occupation is exposed to disruption by Artificial Intelligence. Using public press release data and a fine-tuned BERT-style classifier, I identify new AI product launches or adoptions, then use a large language model to extract the specific capabilities of each AI product. These capabilities are compared to a list of job-relevant skills compiled by ESCO in terms of cosine similarity, and an occupation-level exposure score is calculated by averaging over the job-skill / AI-capability similarity of all skills related to each occupation.
The results put knowledge-heavy and highly-skilled occupations at the top in terms of their exposure to AI, while occupations requiring physical labor are towards the bottom. Occupations with high exposure tend to be highly paid and with low unemployment, suggesting that AI product innovation may be targeted towards tasks for which the required labor is relatively scarce and expensive. The exposure score is decomposed into two components — AI Product Automation and AI Product Augmentation — based on whether the announcing press release describes the product as replacing or as assisting human labor.
Published in Industry and Innovation (Taylor & Francis, 2026).
Recommended citation: Demirev, G. (2026). "AI product innovation and occupational exposure: automation and augmentation in commercial AI deployments." Industry and Innovation, 1–30. https://doi.org/10.1080/13662716.2026.2623903