---
id: "lib-2026-07-00-evidence-on-the-adoption-of-artificial"
type: "library"
title: "Evidence on the Adoption of Artificial Intelligence: The Role of Skills Shortage (DP 18793)"
lang: "EN"
org: "Unknown"
source_type: "Paper"
date: "2026-07-01"
week: "2026-W27"
url: "https://nosov.ai/library/lib-2026-07-00-evidence-on-the-adoption-of-artificial/"
original_url: "https://docs.iza.org/dp18793.pdf"
tier: "B"
tier_name: "Useful, limited"
tier_reason: "IZA discussion paper by Carioli, Czarnitzki and Fernandez with a careful instrumental-variable design on German innovation-survey data; sound method, but cross-sectional, pre-generative-AI data with only 86 adopting firms."
summary_basis: "full-text"
original_minutes: "50"
tags: ["ai adoption", "skills shortage", "firms"]
---

# Evidence on the Adoption of Artificial Intelligence: The Role of Skills Shortage (DP 18793)

Tier B (Useful, limited): IZA discussion paper by Carioli, Czarnitzki and Fernandez with a careful instrumental-variable design on German innovation-survey data; sound method, but cross-sectional, pre-generative-AI data with only 86 adopting firms.

Summary based on the full text.

## In short

IZA Discussion Paper 18793 (July 2026) by Paolo Carioli, Dirk Czarnitzki and Gastón P. Fernández asks whether skills shortages push firms to adopt AI. Using the Mannheim Innovation Panel (German CIS), it finds that firms with more unfilled vacancies in 2017 were more likely to adopt AI and used it more broadly, especially when short of academically qualified and STEM staff.

## Key numbers

- 2,973: German firms in the cross-sectional sample, of which 86 (about 3%) introduced AI after 2017
- 4.2% vs 2.1%: AI adoption among firms that could not fill vacancies as planned versus those that could
- 0.11 pp: OLS rise in AI adoption probability per 10% more unfilled vacancies (IV 2SLS coefficient 0.023 vs 0.011)
- 37%: share of sample firms reporting they could not fill some job openings as planned in 2017

## What it reports

- 4.2% of firms with unfilled vacancies used AI versus 2.1% of firms without shortages.
- OLS: a 10% increase in unfilled vacancies raises adoption probability by about 0.11 percentage points; the IV 2SLS coefficient roughly doubles (0.023), significant at 1%.
- Shortage also predicts breadth of AI methods and application areas.
- Effects concentrate in shortages of academic and STEM skills, which the authors link to Autor's idea that AI may "restore the middle-skill set".
- Associated methods: machine learning, image recognition and language understanding, not knowledge-based systems.
- Among adopters, 55% used AI for process automation, 54% in products or services, 44% for data analysis.

## How they got it

Lagged 2017 vacancy data from the 2018 survey wave, AI adoption from the 2019 wave, controls for size, age, R&D intensity, graduate share, technology path and sector. Instruments: log employees in bankrupt firms in the same district and log average vacancies in the same district and sector (Kleibergen-Paap F of 221.78). Supplementary checks use the 2021 to 2025 waves.

## Limits

- Main AI measure predates generative AI.
- Only 86 adopters in the main sample.
- Cross-sectional German data; authors caution about generalising.
- Instrument validity rests on local labour supply not affecting AI use directly.

## Why it matters for companies

It supports the view that AI adoption in firms is often a response to hiring difficulty, not only a cost-cutting choice. For workforce planners in tight labour markets, AI belongs in the shortage plan alongside recruiting and training, particularly for scarce graduate and STEM roles.

Original: https://docs.iza.org/dp18793.pdf