# Building AI Capabilities in the Organization

Author: Marin Niehues, Niehues Consulting · Published: 2025-09-29 · Updated: 2026-11-16 · AI Adoption & Operations
Source: https://www.niehuesconsulting.com/blog/building-ai-skills-in-the-organisation/

AI capabilities develop in an organization when roles and learning paths are aligned with how the business actually operates. AI literacy describes the ability to use AI systems competently, to judge their output and to understand their risks. Individual training courses are not sufficient on their own. What matters is whether employees apply what they have learned in a real process the next working day, and whether their managers create time and room for it. This article describes a four-step model and treats capability as a delivery topic. This article provides an overview and does not constitute legal advice.

**Building AI capability is a delivery project with clear roles and its own metrics. Its result shows in changed workflows that the business unit applies permanently.**

## Why capability is the bottleneck of AI adoption

Lack of knowledge is the most frequently cited reason in official German statistics for companies not using AI. According to the Federal Statistical Office, around 20 percent of companies with 10 or more employees used AI in 2024, rising to 48 percent among companies with 250 or more employees. The gap between small and large companies is to a considerable extent a capability gap.

A field study of more than 5,000 customer service agents shows how strongly enablement works: with an AI assistant, they resolved around 14 percent more issues per hour on average, and less experienced agents around 34 percent more. The assistant made the knowledge of the best performers available to everyone. The priority for capability building follows directly: the greatest effect arises where AI is embedded in a specific workflow and brings new or less experienced employees up to the level of the unit faster.

## What does AI literacy involve?

AI literacy in an organization involves four abilities: application, judgment, workflow design and steering of use. Many companies equate AI capability with tool knowledge. Employees learn how to operate a language model and are then considered trained. From a delivery perspective, this definition falls short. Capability shows only when a person performs a task in the workflow better than before and takes responsibility for the result.

- Application: use tools confidently in one's own field of work
- Judgment: check results and recognize the limits of the technology
- Design: rebuild workflows so that AI intervenes at the right point
- Steering: take responsibility for the use and benefit of AI in a unit and manage the risks

These four abilities are distributed differently across the levels of an organization. A clerk mainly needs application and judgment. A team leader additionally needs design. Divisional management needs steering. Giving every group the same curriculum misses the actual need and ties up working time that is missing elsewhere.

**Maturity model for AI capability per unit**
- Level 1, isolated: individuals use freely available tools without rules. The next step is a policy with basic training for everyone.
- Level 2, trained: basic training is complete, but use remains irregular. The next step is to rebuild a first workflow with a champion.
- Level 3, in the workflow: at least one workflow has a valid standard. The next step is metrics for usage and quality.
- Level 4, embedded: capability is part of role profiles and onboarding. The next step is to connect further workflows and roles.

## Is AI training mandatory under the AI Act?

Yes. Article 4 of the EU AI Act requires providers and deployers of AI systems to take measures to ensure, to their best extent, a sufficient level of AI literacy among their staff. The regulation prescribes no particular format and no certificate. The yardstick is the technical knowledge and experience of staff and the context in which the systems are used.

In practice, a role-based program with documented participation and traceable content serves the purpose better than a uniform online course. Attendance records do not prove impact, but they are the documentation with which a company shows that it meets its obligation.

## Step 1: Determine capability needs per role

The first step is a role map. It assigns each relevant role to a target profile and answers which tasks of that role change through AI. The basis is the company's prioritized use cases. If they have not yet been defined, start with the assessment described in [Prioritizing AI use cases](/blog/ai-use-case-prioritisation/).

The use cases determine which roles are enabled in which order. Roles with the earliest productive use receive the first learning path. Capability thus develops where the business case requires it, and every investment in enablement can be linked to an expected benefit. The role map also shows where a unit can first win individuals with prior experience as champions.

**The role model at a glance**
- Users: apply AI in their daily tasks
- Multipliers or AI champions: specialists in each unit who support colleagues and collect practical cases
- Managers: decide on use and time budgets and set quality standards
- Specialists: develop and operate applications with internal or external teams

## What are AI champions?

AI champions are specialists within a unit who already use AI confidently in their own work and support colleagues on concrete cases. They are not IT experts. Their value lies in knowing the unit's workflows and answering questions in the language of the function.

For the role to work, it needs three decisions. First, a fixed time budget; ten to twenty percent of working time during the build-up phase has proved workable. Second, a ratio: one champion per ten to twenty users keeps support within reach. Third, a place in the operating rhythm, such as a weekly exchange among champions in which practical cases become standards.

## Step 2: Move learning into the workflow

What is learned on one's own task arrives most reliably in daily work. A learning path therefore starts with a concrete case from the unit, such as preparing quotations or checking invoices, and ends with a changed workflow. A suitable format is a two-week work cycle: a group selects a case, applies AI to it, compares the result with the previous method and records the rule that follows. The cycle ends with a standard that the unit adopts.

This is where training and introduction differ. A prototype that proves its benefit in the real workflow is at the same time the most effective learning format, because participants work on a result that later goes into production. Before business units adopt a solution, they need enablement, clarified responsibilities and adapted workflows. Where this maturity is missing, capability remains confined to individual enthusiasts. The factors that drive adoption are described in [Gaining workforce acceptance of AI](/blog/employee-acceptance-of-ai/). The full path from prototype to operation is set out in the article [From pilot to scale](/blog/from-pilot-to-scale-ai-operations/).

Managers are the second lever in this step. They decide whether learning time is in the calendar and whether a failed attempt counts as insight. Their own learning path differs from user training: it covers deployment decisions and risk assessment.

**Training offer or capability program**
Criterion | Open training catalog | Isolated internal initiatives | With Niehues Consulting
Starting point | Catalog of existing courses | Interest of individual units | Prioritized use cases and role map
Place of learning | Classroom or online course | Self-study | Real workflow of the business unit
Ownership | Learning and development | Varies by initiative | Named owners per unit with metrics
Result | Attendance | Inconsistent practice | Changed workflows and valid standards

## Steps 3 and 4: Embed capability and measure impact

Step 3 turns a learning success into a property of the organization. Capability requirements are written into role profiles and onboarding plans. Responsibility for standards and approvals is assigned to a named role. This settles who spreads new knowledge within the unit.

Step 4 looks at impact on three levels that build on one another:

- Usage: what share of a role uses AI regularly in the defined workflows?
- Quality: how do rework and error rates develop in these workflows?
- Outcome: how do lead time and schedule adherence of the process change?

Attendance rates do not prove impact, yet they remain necessary as evidence of the literacy obligation. The impact metrics belong in the management reporting of each unit, on a monthly rhythm and with a responsible person. If usage does not move, the cause usually lies in the workflow or in leadership and rarely in the quality of the training.

## How long does it take to build AI capabilities?

A first wave per unit takes eight to twelve weeks to an adopted standard, and an organization with several units reaches breadth in twelve to eighteen months. The build-up runs in waves: a pilot group per unit starts, and the next circle follows once the first standard is adopted. With more than five units, two waves run in parallel; otherwise the program ties up the champions for too long.

## How much does AI training for employees cost?

Open AI seminars in the German market are typically charged at a few hundred to around 1,500 euros per person and day, and company-specific training days for up to about twelve participants at around 2,000 to 4,500 euros per day. E-learning basics cost considerably less per person and suit the entry level. Building capability in the real workflow adds consulting support; according to the fee study of the German consultants' association BDU, consultant day rates range from around 1,000 to 2,500 euros.

The larger cost item is working time. Plan one to two days of basic training for users plus the time spent in learning cycles, and the time budget mentioned above for champions. This total belongs in the business case of the use case.

## Niehues Consulting as partner for capability building

[Niehues Consulting](/) supports organizations in building AI capability from the delivery perspective. Founder Marin Niehues builds organizations that deliver and brings AI into operations. His projects range from VC-funded scale-ups to DAX 40 corporations, with assignments in 20+ countries on six continents.

The approach follows the Niehues Delivery Model with the phases Understand, Build, Introduce. Two of its test questions matter most for capability building: Are users enabled and workflows adapted? Which metric shows after 90 days that the solution is taking hold? In AI management, use cases and their sequence are defined, and the capability need per role follows from them. In prototyping and introduction, business units learn in the real workflow until they adopt the solution and operations take it over with internal or external teams. In delivery management and organization design, roles and decision paths are shaped so that what has been learned lasts in daily business.

The entry point is a free [initial consultation](/#kontakt) of 30 minutes. It clarifies which roles to enable first and which use case offers the best occasion for learning.

**The four steps at a glance**
First, a role map determines the need from the prioritized use cases. Second, learning takes place on real cases in the workflow. Third, role profiles and champions embed the knowledge. Fourth, metrics on usage, quality and outcome show whether the effort pays off.

**Sources**
- Federal Statistical Office of Germany (Destatis): Use of information and communication technologies in enterprises, 2024
- Brynjolfsson, Li, Raymond: Generative AI at Work, NBER Working Paper, 2023
- Bundesverband Deutscher Unternehmensberater (BDU): fee study
- Regulation (EU) 2024/1689 on artificial intelligence, Article 4

## Frequently asked questions

### What does AI literacy mean?

AI literacy is the ability to use AI systems competently, to judge their output and to understand their opportunities and risks. The EU AI Act uses the term for the knowledge staff need for responsible use. In organizations it also covers workflow design and the steering of use.

### Is AI literacy training mandatory for employees?

Yes, Article 4 of the AI Act requires providers and deployers of AI systems to take measures for a sufficient level of AI literacy among their staff. The regulation prescribes no particular format or certificate. A role-based program with documented participation is the sensible answer.

### How do you document AI literacy for the AI Act?

Record for each role which content was taught, who took part and which systems the role uses. Add the policy and the standards of each unit. These records show that the company meets its obligation.

### How long does AI upskilling take?

A first wave per unit takes eight to twelve weeks, and breadth across an organization with several units takes twelve to eighteen months. The next circle starts once the first standard is adopted in the workflow.

### How much does AI training for employees cost?

Open seminars are typically priced at a few hundred to around 1,500 euros per person and day, company-specific training days at around 2,000 to 4,500 euros. The larger cost item is the participants' working time. Both belong in the business case of the use case.

### What is an AI champion?

An AI champion is a specialist within a unit who already uses AI confidently and supports colleagues on concrete cases. One champion per ten to twenty users and a fixed time budget have proved workable. Champions turn practical cases into standards for the unit.

### How do you measure the success of AI training?

Success shows in workflow metrics: the share of a role using AI regularly, the development of rework and errors, and the lead time of the process. Attendance rates document the literacy obligation but do not prove impact.

### Who should own AI capability building?

Ownership belongs in the units that realize the benefit. Learning and development and IT provide methods and platforms. A named owner per unit steers the role map and learning cycles and reports to divisional management.
