# IU Study on AI Training: Learning Time Is the Bottleneck

Author: Marin Niehues, Niehues Consulting · Published: 2026-09-25 · AI Adoption & Operations
Source: https://www.niehuesconsulting.de/en/blog/iu-study-ai-training-learning-time/

The bottleneck in building AI literacy lies in how companies are organised, because they rarely give their employees learning time, clear expectations and suitable training. So finds the study "KI in der Arbeitswelt: Wie gelingt der Change?", which IU International University of Applied Sciences published on 24 September 2026. 81.6% of the 2,000 employees surveyed are confident they can acquire new skills on their own, while 43.9% used no working time for AI training in the twelve months before. By our calculation a company with 300 employees can close this gap with around 13 learning hours per person a year, provided it plans learning time as a budget and ties it to real workflows.

**Key facts at a glance**
- IU published the study "KI in der Arbeitswelt: Wie gelingt der Change?" on 24 September 2026, based on a survey of 2,000 employees aged 16 to 65 from 31 March to 14 April 2026.
- 43.9% used no working time for AI training in the twelve months before the survey, while a further 14.3% say AI training does not concern them.
- Of those without learning time, 24.0% say their employer did not want them to take such training.
- Art. 4 of the AI Act has required measures on AI literacy since 2 February 2025. The version amended by the Digital Omnibus has applied since 27 July 2026.

**A company with 300 employees that budgets learning time per role, appoints multipliers with a fixed share of their time and ties learning to a real workflow invests just under 0.8% of its labour costs. It also gains the evidence Art. 4 of the AI Act requires.**

## IU study: almost six in ten employees had no learning time for AI

Almost six in ten employees in Germany used no working time for AI training within a year, even though a majority consider AI literacy a requirement for starting a career. 45.5% of them use AI tools. 54.9% expect AI skills to become a basic requirement for career starters in many fields, rising to 65.6% among managers, as [ingenieur.de](https://www.ingenieur.de/karriere/arbeitsleben/ohne-ki-kenntnisse-kein-berufseinstieg-mehr-als-die-halfte-der-beschaftigten-glaubt-das/) reported on 24 September 2026.

Only 8.9% reach 16 hours or more. Using band midpoints (100 hours for the open top band) gives an average of around 5.5 hours per employee, while the median is zero hours.

::: diagramm Almost six in ten employees used no working time for AI training in twelve months
- !No time at all | 43.9 | 43.9%
- !Does not concern me | 14.3 | 14.3%
- Up to 2 hours | 14.9 | 14.9%
- 3 to 5 hours | 10.5 | 10.5%
- 6 to 15 hours | 7.7 | 7.7%
- 16 to 40 hours | 5.2 | 5.2%
- More than 40 hours | 3.7 | 3.7%
Source: IU International University, KI in der Arbeitswelt: Wie gelingt der Change?, September 2026; n = 2,000 employees; the bands 41 to 80 and more than 80 hours are combined

## After a lack of need, the reasons against AI learning time lie with the employer

Employees without learning time name a lack of need first (47.4%), followed mainly by reasons their employer controls: 24.0% say the company did not want the training, 21.8% lack information and 21.6% find no suitable offer. Across all respondents, roughly one employee in ten (10.5%) had no learning time for AI because the employer did not want it (12.0% of those without a management role).

::: diagramm After a lack of need, employees without learning time mostly cite reasons their company controls
- Not necessary | 47.4 | 47.4%
- !Not wanted by the company | 24.0 | 24.0%
- No information about it | 21.8 | 21.8%
- No suitable offer | 21.6 | 21.6%
- No time | 20.6 | 20.6%
- Approval too complicated | 3.2 | 3.2%
Source: IU International University, KI in der Arbeitswelt: Wie gelingt der Change?, September 2026; only respondents without learning time for AI in the past twelve months, multiple answers possible

The most frequent reason also says something about the organisation, because 65.0% of respondents see no expectation from their company that they use AI tools. Where management states no expectation, employees understandably regard training as dispensable. This link is our interpretation.

## Companies support the use of AI tools less often than any other training topic

Companies mainly support subject-matter skills, while they support the use of AI tools least often. 51.6% say their company supports training in subject-matter topics very strongly or fairly strongly, compared with 26.2% for the use of AI tools. Critical thinking is supported according to 33.6%, although Anne Gfrerer, professor of business informatics at IU, calls critical thinking, recognising whether an AI answer is plausible, the key future skill in the study.

::: diagramm Companies support the use of AI tools less often than any other training topic surveyed
- Subject-matter skills | 51.6 | 51.6%
- Communication | 46.0 | 46.0%
- Use of digital applications | 42.0 | 42.0%
- Problem solving | 41.1 | 41.1%
- Critical thinking | 33.6 | 33.6%
- Empathy | 31.6 | 31.6%
- !Use of AI tools | 26.2 | 26.2%
Source: IU International University, KI in der Arbeitswelt: Wie gelingt der Change?, September 2026; n = 2,000 employees; share of answers "very strongly" and "fairly strongly" supported

Among managers 32.4% used no learning time for AI, compared with 47.6% of employees without a management role. 16 hours or more are reached by 18.5% of managers and 5.9% of other employees. At the same time 35.1% of managers feel left alone on AI by their own manager. More on this layer in [Strengthening middle management](/en/blog/strengthening-middle-management/).

**The learning gap in numbers**
- 81.6% | of respondents are confident they can acquire new skills on their own
- 43.9% | used no working time for AI training in twelve months
- 27.6% | say their company offers training in the use of AI tools
- 3.1 times | as often as other employees, managers reach 16 learning hours or more

## The AI learning gap is an organisational problem, because learning time goes unplanned

The learning gap arises in the organisation, because many companies introduce AI on top of existing work while learning time rarely appears in any plan. Of AI users, 27.9% say their workload has risen since, while 21.6% report lower productivity. Philipp Diebold, professor of business informatics at IU, attributes this to AI tying up resources when it comes "on top" of unchanged ways of working.

Organisation design reveals three missing decisions that are standard for other ongoing tasks. Learning time rarely has a budget in the capacity plan, so it loses out to day-to-day business. Passing on knowledge rarely has a role with its own share of time, although 72.3% of respondents say they learn at work mainly by exchanging with colleagues. Workflows usually lack a description of which step will be done with AI in future and who checks it. Tiering skills by role is covered in [Building AI capabilities in the organisation](/en/blog/building-ai-skills-in-the-organisation/). Why acceptance grows from experienced benefit is explained in [Winning employee acceptance of AI](/en/blog/employee-acceptance-of-ai/).

## Since July 2026, Art. 4 of the AI Act makes the measure the subject of the literacy duty

Since 2 February 2025, Art. 4 of the AI Regulation has required providers and professional users of AI systems to take measures on the AI literacy of their staff. The Digital Omnibus Regulation (EU) 2026/1744, in force since 27 July 2026, has amended the provision. The earlier version required companies to make every effort to ensure a sufficient level of AI literacy. The current version requires measures that support the development of AI literacy, without guaranteeing a specific level for any individual.

A budget per role with content and hours used documents that the measures fit how the systems are deployed. Under Section 98(1) of the German Works Constitution Act the works council co-determines in-company training measures, so the budget should be agreed with it early. This assessment does not replace legal advice. Details are covered in [AI literacy under Article 4 of the AI Act](/en/blog/ai-literacy-article-4-ai-act/), the route with employee representatives in [Works council and AI introduction](/en/blog/works-council-and-ai-introduction/).

## A learning time budget for 300 employees costs around 171,000 euros a year

By our calculation the budget for 300 employees comes to around 3,800 hours and 171,000 euros in labour costs a year, that is 12.7 hours and around 570 euros per employee. It distinguishes four groups, because a basic module alone covers neither practice in the workflow nor the passing on of knowledge. The share of regular users comes from the IU study, the other values are assumptions any company can replace.

**Learning time budget for 300 employees per year**
Group | People | Hours per person | Total hours | Labour costs
Basics for everyone | 300 | 4 | 1,200 | €54,000
Practice time for regular users | 137 | 12 | 1,644 | €73,980
Multipliers | 9 | 80 | 720 | €32,400
Managers | 30 | 8 | 240 | €10,800
Total | 300 | 12.7 | 3,804 | €171,180

**Assumptions behind the calculation**
- Basics: four hours for everyone on how AI works, its limits, data rules and checking results.
- Regular users: 45.5% according to the IU study, so 137 people, with one hour of practice time a month on their own cases.
- Multipliers: one per 15 regular users, so nine people, with around ten per cent of their working time over six months, assuming 1,600 working hours a year.
- Managers: one per ten employees, with eight hours on how tasks change in the team.
- Costs: 45.00 euros labour cost per hour worked, the 2025 average according to the Federal Statistical Office, excluding trainer fees, licences and learning platforms.

::: diagramm More than two fifths of the learning time budget goes to practice time for regular users
- !Practice time for regular users | 1644 | 1,644 h
- Basics for everyone | 1200 | 1,200 h
- Multipliers | 720 | 720 h
- Managers | 240 | 240 h
Source: Niehues Consulting calculation, September 2026; company with 300 employees, share of regular users from the IU study, other assumptions in the box above

At 1,600 working hours per person, 171,180 euros equal just under 0.8% of annual labour costs. Compared with today's average of around 5.5 hours, learning time slightly more than doubles. It breaks even when each of the 137 regular users saves 27.8 hours a year, around eight minutes per working day at 220 days.

## The maturity of an AI learning organisation shows in learning time, roles and workflows

Maturity shows in learning time, responsibility, workflows and measurement. For a large share of employees the IU study describes stage 1, because almost two thirds perceive no expectation from their company.

**Three stages of the learning organisation for AI**
Criterion | Stage 1: learning as a private matter | Stage 2: learning as an offer | Stage 3: learning in the workflow
Learning time | not planned, learning at the margins of the day | courses can be booked, time approved case by case | fixed budget per role in the capacity plan
Responsibility | nobody | HR organises training | multipliers per unit with a share of time, managers own learning goals
Workflows | AI comes on top of existing work | training separate from work | one workflow per unit redesigned with AI
Expectation | not stated | stated in general terms | described per activity
Metric | none | attendance | learning hours per role and effect in the workflow
Evidence under Art. 4 | missing | attendance lists | budget, content and attendance per role

How AI reaches daily use is described in [Change management for AI adoption](/en/blog/change-management-ai-adoption/), the state of adoption in [AI adoption in German companies](/en/blog/ai-adoption-in-german-companies/).

## Executive teams in mid-sized companies should first approve AI learning time as a budget

The findings translate into six steps, each with an owner, a time frame and a metric:

1. Budget learning time: within four weeks, management and the head of HR set hours per role in the 2027 plan. Metric: share of employees with an approved learning budget.
2. State the expectation per activity: by the end of 2026 every head of unit names which activities are to be done with AI and how results are checked. Metric: share who know the expectation, by short survey.
3. Appoint multipliers: within eight weeks, heads of unit choose one person per 15 regular users with a fixed share of time. Metric: standards created from practice cases per quarter.
4. Redesign one workflow per unit: process owners use AI for twelve weeks in a time-consuming workflow and describe it anew. Metric: processing time per case before and after.
5. Enable managers first: in the first quarter of 2027, HR trains all managers on the changes in their teams. Metric: share of managers who used their eight hours. Programme design is covered in [AI training programmes for executives](/en/blog/ai-training-for-executives/).
6. Measure evidence and effect: HR documents learning hours per person and role for Art. 4 of the AI Act and reports quarterly. Target for year one: fewer than ten per cent of employees without a learning hour.

## Niehues Consulting links AI adoption with learning time, roles and workflows

[Niehues Consulting](/en/) supports executive teams and heads of department in building organisations that deliver reliably. In delivery management, a delivery manager or product owner steers the programme until the change has arrived in operations. Where AI enters the workflows, a business case and a prototype are added, followed by introduction with learning time and multipliers.

The approach follows the Niehues Delivery Model with its phases Understand, Build and Introduce. For building skills it asks three questions. In Understand: who uses AI today, for what and with how much learning time? In Build: which roles, budgets and workflows do operations require? In Introduce: are users enabled and workflows adapted? Founder Marin Niehues, a management consultant from the delivery side, builds organisations that deliver. His projects range from venture-backed scale-ups to DAX 40 groups, with assignments in more than 20 countries on six continents. In the free 30-minute initial consultation you assess together how much learning time your organisation needs and in which workflow it takes effect first: [book a consultation](/en/#kontakt).

**What decision-makers can take from the IU study**
Employees are confident they can learn, their companies rarely provide the time. 43.9% had no learning time for AI in a year, almost a quarter of them because the company did not want it. A learning time budget per role costs a company with 300 employees around 171,000 euros of working time a year and pays off when regular users save around eight minutes a day. The budget also produces the evidence that Art. 4 of the AI Act requires in its version since July 2026.

**Sources**
- [IU International University of Applied Sciences: KI in der Arbeitswelt: Wie gelingt der Change?](https://www.iu.de/forschung/studien/ki-und-arbeitswelt/), 24.09.2026
- [IU International University of Applied Sciences: press release on the study](https://www.iu.de/news/iu-studie-zeigt-ki-veraendert-den-berufseinstieg--zukunftskompetenzen-werden-bislang-zu-wenig-gefoerdert/), 24.09.2026
- [ingenieur.de: report on the IU study](https://www.ingenieur.de/karriere/arbeitsleben/ohne-ki-kenntnisse-kein-berufseinstieg-mehr-als-die-halfte-der-beschaftigten-glaubt-das/), 24.09.2026
- [Regulation (EU) 2024/1689 on artificial intelligence, Art. 4 in the current version](https://artificialintelligenceact.eu/article/4/)
- [Lewis Silkin: The Digital Omnibus on AI enters into force today](https://www.lewissilkin.com/insights/2026/07/27/the-digital-omnibus-on-ai-enters-into-force-today-102nedo), 27.07.2026
- [Federal Statistical Office: labour costs 2025, press release no. 148](https://www.destatis.de/DE/Presse/Pressemitteilungen/2026/04/PD26_148_624.html), 29.04.2026
- [Section 98 Works Constitution Act (BetrVG)](https://www.gesetze-im-internet.de/betrvg/__98.html)

::: faq Frequently asked questions

### When was the IU study "KI in der Arbeitswelt" published?

IU published the study on 24 September 2026. The survey ran from 31 March to 14 April 2026 with 2,000 employees aged 16 to 65.

### How many employees use no working time for AI training?

43.9% used no working time for AI training in twelve months, a further 14.3% say the topic does not concern them.

### Why do employees not train in AI?

Employees without learning time most often say the training is not necessary (47.4%). They are followed by reasons the company controls: it does not want the training (24.0%), provides no information (21.8%) or offers nothing suitable (21.6%).

### How much learning time for AI does a company need?

For a company with 300 employees our calculation comes to around 3,800 hours a year, 12.7 hours per person on average and around 171,000 euros in labour costs.

### Is AI training mandatory under the AI Act?

Since 2 February 2025, Art. 4 of the AI Act has required providers and professional users of AI systems to take measures on the AI literacy of their staff. Since 27 July 2026 these measures must support the development of AI literacy, without guaranteeing a specific level for any individual. This assessment does not replace legal advice.

### What does the IU study mean for mid-sized companies in Germany?

For mid-sized companies in Germany, building AI literacy is a decision by management on learning time, roles and workflows, because employees are willing to learn. Usually missing are a budget in the capacity plan, multipliers with a fixed share of time and a clear expectation per activity.