Embedding AI skills in job profiles and performance reviews

The introduction of generative AI and language models in the workplace almost always starts informally. Employees discover on their own how a model can summarize emails, speed up code, or structure market analyses. However, as soon as an organization wants to structurally secure these productivity gains, this informal approach quickly runs into sharp limits. As long as the use of AI systems is not explicitly included in job profiles, competency matrices, and review cycles, adoption remains dependent on individuals, compliance risks arise, and there is no objective performance standard.

Formalizing AI skills is not a purely administrative process; it directly affects job evaluation, terms of employment, and day-to-day work practice. Anyone who expects employees to use language models safely and efficiently must define what good craftsmanship looks like, which responsibilities come with it, and how performance is evaluated. This article covers how an organization translates AI competencies into concrete job requirements, how these fit within existing HR frameworks, and how managers assess performance without resorting to superficial metrics.

Why informal adoption stalls without formal anchoring

When the use of AI remains optional, two persistent patterns emerge within teams. On one side, a group of early adopters emerges who produce significantly faster with the help of AI, but often without documented quality checks or source attribution. On the other side, a group of employees keeps avoiding the technology out of uncertainty about copyright, privacy, or fear of mistakes. The result is a widening gap in productivity and working methods within the same job group.

Read the article on mapping the impact of AI on roles and tasks to determine in advance which work processes will fundamentally change. Without formally recalibrating tasks, employees continue to be assessed against outdated time norms, even though their actual work has shifted from writing to editing, verifying, and synthesizing.

In addition, the absence of formal frameworks directly leads to information risks. See the analysis on tackling shadow AI within teams to prevent employees from using unapproved tools without oversight. When a job profile doesn't specify which tooling is permitted and which validation steps are mandatory, an organization will find it difficult afterward to hold employees accountable for improper data use or uncontrolled hallucinations in customer documents.

The four competency levels for AI literacy

To keep job profiles manageable, it is unwise to draw up a separate skill requirement for every individual AI tool. Software changes quickly, but the underlying cognitive and methodological skills remain stable. A workable competency model distinguishes four increasing maturity levels that can be directly linked to junior, mid-level, and senior job levels.

Using these four levels immediately clarifies which role requires which skill. An administrative employee or content writer typically only needs levels 1 and 2, while a data analyst or process architect must master level 3, and a lead engineer or compliance officer operates at level 4.

Adapting job profiles: from vague buzzwords to concrete tasks

Many organizations make the mistake of including vague phrases in job postings and profiles such as "experience with modern AI tools desired" or "AI-minded." Such wording offers neither the candidate nor the manager any real guidance during recruitment or performance reviews. A sound job profile translates the technology into specific responsibilities and result areas.

Consult the overview on AI skills in non-technical roles for current market insights into required competencies. In non-technical domains such as marketing, legal, HR, and customer service, the task portfolio shifts primarily from manual content creation toward sharp query formulation, synthesis, and quality control.

In concrete terms, this means the task description of, for example, a business analyst changes:

• Old wording: "Responsible for manually preparing process analyses and market reports based on interviews and desk research."
• Recalibrated wording: "Structures raw transcripts and process documentation using approved language models; comprehensively checks generated syntheses for consistency, missing preconditions, and factual accuracy before decisions are made."

By explicitly placing responsibility for verification with the employee, the model is prevented from being used as an excuse for mistakes ("the model gave this answer"). The human professional remains ultimately responsible for the end product.

Separating responsibilities: operational roles versus governance tasks

When rewriting profiles, an organization must draw a sharp distinction between using AI within a task and managing or auditing AI systems. Adding heavy compliance or management obligations to operational roles leads to overload and delay.

See the guide on AI governance roles and responsibilities to properly assign oversight responsibilities. While the operational employee is responsible for data hygiene and output verification, responsibility for vendor audits, model selection, DPIAs, and logging lies with specific governance roles such as the security officer, privacy counsel, or the product owner of the AI platform.

When this separation is clearly established in job profiles, psychological safety is created. Employees know within which vetted frameworks they are free to experiment, and when a use case needs to be escalated to a governance owner.

Designing objective assessment criteria and KPIs

Assessing employees who work intensively with AI brings new challenges. Classic productivity metrics, such as the number of articles written, tickets closed, or lines of code per day, can easily be gamed with generative AI without actual business value increasing. Managing by quantity gets you prompt-generated filler.

A well-thought-out assessment system therefore uses a balanced set of criteria resting on three pillars:

• Quality and error reduction: How accurate is the final deliverable? Are hallucinations caught before publication or sending? Is there a demonstrable quality standard in editing work?
• Speed and turnaround time: Has the turnaround time for complex syntheses, analyses, or drafts decreased, while maintaining or improving the quality level?
• Process and knowledge sharing: Does the employee document effective prompt structures and working methods? Are discovered vulnerabilities or model deviations reported to the team?

During the performance review, the manager does not assess whether someone types prompts all day, but whether the employee masters the tooling as a professional lever. The central question is: does the AI skill enable the employee to reach reliable results faster at a higher level of abstraction?

Linking to learning goals and development plans

Including AI requirements in job profiles only makes sense if employees get real opportunities to build the required skills. A review should never come as a surprise; it should be the capstone of an ongoing development path.

Read more in the guide on Setting up AI training for employees to directly link learning trajectories to assessment criteria. Instead of a one-off workshop on prompting, the organization embeds permanent learning cycles in the personal development plan (PDP). Concrete milestones are recorded in it, such as:

• Completing the internal course on data security and ethical AI use within three months;
• Redesigning two weekly manual reports into a structured, AI-supported workflow;
• Conducting a peer review of colleagues' validation methods within the same discipline.

This turns the review cycle from a static control instrument into a development-oriented mechanism that encourages employees to continuously sharpen their digital professional skills.

Legal and employment law frameworks in the Netherlands

Changing job profiles and introducing new technology-related assessment systems is subject to formal laws and regulations in the Netherlands. Organizations cannot push through these changes unilaterally without consulting works council bodies.

Under Article 27 of the Dutch Works Councils Act (WOR) the Works Council (OR) has the right of consent for decisions on establishing, amending, or withdrawing:

• A personnel assessment system or remuneration system (Art. 27(1)(g) WOR);
• Regulations in the field of training and education (Art. 27(1)(f) WOR);
• Facilities aimed at or suitable for observing or monitoring the presence, conduct, or performance of employees (Art. 27(1)(l) WOR).

When an organization deploys monitoring software to record how many prompts employees send or how intensively AI software is used, this falls directly under the OR's right of consent. In addition, the GDPR requires that no automated decision-making about employees take place without human intervention. A performance review may therefore never be generated automatically from performance statistics produced by software platforms.

Pitfalls and misuse in personnel assessment

When operationalizing AI in HR processes, specific pitfalls arise in practice that can seriously damage trust in the workplace:

1. Measuring activity instead of impact: Counting API calls, tokens, or generated documents leads to fake productivity. Employees optimize for the metric instead of the business outcome.

2. Unequal access to tooling: Assessing employees on AI literacy while part of the team has no license for advanced models or has to work with slow, outdated interfaces leads to unfair evaluations and frustration.

3. Lack of clarity about liability for mistakes: If a model makes a subtle error in a legal or financial document, friction arises when it hasn't been agreed in advance who bore the duty to check. The job profile must explicitly state that the employee is accountable for the approved content.

4. Bias in evaluations: Managers who themselves have little experience with AI sometimes overestimate what the technology can do and expect unrealistic time savings. Conversely, skeptical managers may unfairly view employees who use AI as "lazy." Both extremes call for management training before performance reviews begin.

Practical action plan for HR and managers

To carefully embed AI skills in the organization, management goes through the following six consecutive steps:

1. Take stock of the actual tasks: Determine, per job family, which core tasks change due to the availability of language models and where the greatest gains in quality and efficiency can be achieved.
2. Link the desired competency level: Assign each role an appropriate level from the four competency levels (basic, applied, advanced, or strategic).
3. Rewrite result areas and job requirements: Replace vague buzzwords with explicit responsibilities around prompt structuring, validation, source attribution, and data hygiene.
4. Involve the works council in good time: Align proposed changes to assessment systems, training arrangements, or monitoring with the OR at an early stage, in accordance with Article 27 WOR.
5. Equip leadership: Train managers to recognize effective AI use, so they can probe validation processes and actual value creation during annual reviews.
6. Evaluate and recalibrate annually: Given the pace at which models and integrations develop, job profiles and competency requirements should be briefly reviewed each year for relevance and workability.

By systematically following these steps, an organization transforms the use of AI from an unregulated experiment into a professional, measurable, and legally secured core competency of its workforce.