---
title: "AI adoption, productivity and employment: evidence from European firms"
aliases:
  - "BIS European Firm-Level Study"
  - "BIS WP 1325"
  - "Aldasoro et al 2026"
tags:
  - eu
  - us
  - central-bank
  - causal
  - firm-level
  - productivity
  - augmentation
  - working-paper
source: "/research/2026-01-bis-eu-firms-ai-adoption.pdf"
extract: "/research/extracts/2026-01-bis-eu-firms-ai-adoption.md"
authors: "Iñaki Aldasoro; Leonardo Gambacorta; Rozalia Pal; Debora Revoltella; Christoph Weiss; Marcin Wolski"
publisher: "Bank for International Settlements"
date: "2026-01"
---

# AI adoption, productivity and employment: evidence from European firms

> [!info] Source
> PDF (human): [2026-01-bis-eu-firms-ai-adoption.pdf](/research/2026-01-bis-eu-firms-ai-adoption.pdf) · Raw extract (machine): [2026-01-bis-eu-firms-ai-adoption.md](/research/extracts/2026-01-bis-eu-firms-ai-adoption.md)
> BIS Working Paper No 1325 — January 2026 — Empirical firm-level study (Europe + US), 36 pages
> **JEL:** D22, J24, L25, O33, O47 · **ISSN:** 1020-0959 / 1682-7678

## TL;DR

Causal study of AI adoption in **>12,000 EU non-financial firms + 800 US firms** (EIBIS-ORBIS data, 2019–2024) using a novel instrumental-variable strategy that assigns adoption rates of comparable US firms to EU firms. **AI adoption increases labor productivity by 4%**, with gains driven by **capital deepening rather than employment displacement** — i.e., AI augments worker output, not replaces labour, in the short run. Benefits concentrated in **medium and large firms**; AI-adopting firms are more innovative and pay higher wages.

## Key findings

### Causal effect on productivity

- AI adoption causally **increases labor productivity by 4%** on average (after IV identification).
- Effect is economically significant; aligns with mid-range macroeconomic projections (Acemoglu 2024, Bergeaud 2024) — **not** the optimistic "productivity boom" scenario (Baily/Brynjolfsson/Korinek 2023).

### Mechanism: capital deepening, not labour displacement

- Productivity gains stem from **capital deepening**, not job destruction.
- **AI augments worker output without reducing employment** in the short run.
- Consistent with micro-level evidence on AI-aided efficiency in cognitive tasks (Noy & Zhang 2023; Gambacorta et al. 2024).

### Distributional patterns

- **Adoption is highly stratified by firm size:**
  - Large firms (250+ employees): 45% adoption.
  - Small firms (10–49 employees): 24%.
- **Country variation:**
  - Sweden 52%
  - US 34%
  - Romania 22%
- **AI-adopting firms** are: more innovative in general, invest more intensively, face tighter skilled-labor constraints, pay **higher wages**.
- **Productivity gains concentrated in medium and large firms** — risks widening inequality given Europe's SME-dominated industrial structure.

### Complementary investments matter

- Productivity gains larger when firms also invest in **software and data** or **employee training**.
- Hardware subsidies alone insufficient; integration, workflow redesign, continuous learning required.
- Re-skilling priorities: **"fusion skills"** — prompt engineering, data stewardship, human-in-the-loop decision-making.

## Methodology in brief

- **EIBIS** (European Investment Bank Investment Survey): annual, ~12,000 EU firms (27 countries) + 800 US firms since 2019.
- Stratified random sampling by country, sector, size class.
- AI adopters defined as firms using "big data analytics and AI in parts of the business or entire business organised around AI" (EIBIS question on big data analytics + AI inc. ML, RPA, NLP, neural networks).
- **Novel IV strategy:** match each EU firm with comparable US firms (same sector, size, innovation, investment, managerial practices, external finance); assign US adoption rate as exogenous proxy for EU adoption.
- Inspired by Rajan & Zingales (1998) financial-dependence approach to growth.
- Propensity-score balancing tests confirm matched US/EU distributions are near-identical.
- Country, sector, year fixed effects + observable financial controls.

## Implications for AdaptAI

This is the **most rigorous causal estimate** of AI's productivity and employment effects in the batch — directly applicable to UK firm-level prose and the framing of "augmentation vs displacement".

### Calculator (`/calculator`)
- **Validates the augmentation framing** — the paper finds NO short-run aggregate displacement. The calculator's High-exposure bands should not over-promise displacement; the language ("many tasks may be automatable or heavily AI-assisted") is well-calibrated.
- **`industryModifier` direction is consistent** — large firms in finance/IT/professional services are the highest adopters and the highest productivity gainers. No change required.
- **`jobTitlesWeighted` defaults** for **junior vs senior** roles in the same occupation may need different `weight` values: paper finds adoption skews to large firms, where employment patterns differ (more compositional change). Junior roles in large firms more affected; senior roles either stable or higher-paid.
- **Methodology modal:** worth citing the BIS 4% productivity number alongside the OECD 0.4–1.2pp range — gives users both micro and macro figures.

### Big Picture (`/big-picture`)
- **`bigPictureData.aiAdoptionSpeed.adoptionIndex`:**
  - UK currently estimated; benchmarks: Sweden 52%, US 34%, Romania 22%. UK should sit in upper-mid (probably 35–45 range based on the DSIT data in the GOV.UK paper).
  - Paper's large vs small firm split (45% vs 24%) supports the existing "adoption depends on firm size" narrative.
- `bigPictureData.skillsDemand.growing` — paper explicitly mentions "fusion skills" (prompt engineering, data stewardship, human-in-the-loop) — already in the `skillsDemand.growing` list as Prompt Engineering, AI/ML Engineering. Add **Data Stewardship** and **Human-AI Collaboration** if not already (Human-AI Collaboration is there).
- `bigPictureData.industryExposure.aiInvestment` — supports the existing investment-by-sector ranking; large firms = higher AI investment.
- **Crucially:** paper's headline that **AI adoption does NOT reduce employment in EU firms in 2019–2024** is the cleanest direct counterweight to the Big Picture's "15/17 industries showing job decline" framing. The aggregate decline projection should be presented as a forward-looking projection, not a current reality.

### Industry Insights
- `/industries/finance` — supports current narrative; finance is a major AI-adopter and productivity-gain sector.
- `/industries/tech` — software/data complementary investment is exactly where Tech firms are concentrated.
- `/industries/manufacturing` — paper covers non-financial firms broadly; manufacturing in the EU sample.
- All industry pages should soften "displacement" claims with "augmentation found in EU firm-level data 2019–2024" caveat.

### How to Adapt
- `/adapt/upskilling` — **direct policy recommendation** from the paper: re-skilling should prioritise "fusion skills" (prompt engineering, data stewardship, human-in-the-loop). Useful for the page's recommendation list.
- `/adapt/ai-tools` — paper supports the framing that AI tools augment workers; add citations.
- `/adapt/career-transitions` — augmentation framing means transitions can be intra-occupation (re-skill within current role) more often than cross-occupation. Useful for the guide's strategic advice.
- `/adapt/human-advantage` — paper notes higher wages for workers in AI-adopting firms, consistent with the "use AI to outperform" framing.
- `/adapt/rights-protections` — paper warns of widening inequality given Europe's SME structure; useful for the policy section.

## Related notes

- [[Atlanta Fed CFO Survey 2026]] — parallel productivity question, US lens, CFO-survey methodology.
- [[IMF Skill Gaps SDN]] — skills and adoption story; complements this paper's firm-side evidence.
- [[DSIT UK Labour Market Assessment]] — productivity gain figures cited in UK government evidence.

## Caveats & limitations

- "AI adoption" definition is broad (big data analytics + AI tools used in parts of the business). May over-count basic data analytics as AI.
- 2019–2024 sample window is short; paper itself notes "longer-term effects remain uncertain".
- IV strategy identifies effect only of EU AI adoption that is similar to US peers' patterns — local AI variants may differ.
- Capital deepening finding may understate displacement that occurs through reduced hiring rather than layoffs.
- Productivity is firm-level; says nothing about within-firm composition shifts (which other papers — e.g. Atlanta Fed — find substantial).
- EU sample is heavily weighted toward survey-cooperating firms; selection on observables addressed but not exhaustive.
- Paper acknowledges divergence between firm-level (4% gain) and macro (0.07–0.66% TFP) estimates — adoption frictions explain the gap.