---
title: "AI check: how far along is your company with AI?"
url: https://punktexe.de/en/ai-check/
language: en
source: generated from 02_website/content/seiten/en/ai-check.html
---

[Home](https://punktexe.de/en/) / AI check

AI check · around 4 minutes

# What can your company already do with AI?

Twelve questions on six capabilities, around four minutes. Result straight away, no sign-up. A self-assessment, not an audit. Your answers are not transmitted to us. You can pass on your result as a link or PDF or send it straight to us.

Check not loading? [Open the check directly](https://punktexe.de/wp-content/uploads/check/ai-readiness-check.html)

**What is assessed**

## Six capabilities, twelve questions

### 01 Use your own data

- Have you defined which company data may be processed with AI and which must stay in-house?

- Can AI work with your own documents and systems without employees copying content into private accounts?

Possible next step: Define data classes (what may go where) and set up a first assistant on your own documents, locally or in the EU.

How it works in practice:

- Define three data classes (public, internal, confidential) and name the permitted models for each class.

- Start with a clearly bounded document collection, such as manuals or templates, and show the source in every answer.

- Process confidential material only locally or after masking it.

### 02 Switch models

- Could you replace the AI model behind an application without rebuilding the application?

- Do you have a second route if your AI provider changes its prices, terms or availability?

Possible next step: Connect applications through a central interface (gateway) and set up a local or European model as a second route.

How it works in practice:

- Connect all applications through a central interface (gateway), not directly to a provider.

- Version prompts and test cases so that a new model can be checked quickly against your own tasks.

- Keep a local or European model ready for operation as a second route.

### 03 Know your costs

- Do you know what your AI usage costs per month, per application or department?

- Are there budgets or limits for each AI application?

Possible next step: Make costs per request and per application visible and set budgets before usage grows.

How it works in practice:

- Measure costs per request, application and department centrally.

- Set monthly budgets with a warning threshold for each application.

- Route simple tasks to smaller or local models and reserve large models for difficult cases.

### 04 Control access rights

- Have you defined who may use which AI tools and models?

- Have you defined which data and systems an AI application may access itself?

Possible next step: Define roles and approvals per model and limit each application's access to what it needs.

How it works in practice:

- Define roles: who may use which tools and models.

- Give each application only the access it needs, with its own key per application.

- Log access and approvals in a traceable way.

Reference: Principle of least privilege

### 05 Be in operation

- Is at least one AI application in daily operation, not just in a pilot?

- Does this application have a responsible person, and is its performance measured?

Possible next step: Take one clearly bounded use case all the way into operation, with a responsible person and a key metric.

How it works in practice:

- Choose a use case with a clear key metric, such as processing time per case.

- Name a responsible person who collects results and feedback.

- Define what a verification agent checks automatically and where a person has to decide. Log the checks and decisions.

Reference: ISO/IEC 42001 (management system for AI)

### 06 Enable people

- Have your employees been trained to use AI safely and sensibly?

- Is there a clear rule on what is permitted and when a person checks the result?

Possible next step: A short training session per role and a clear usage rule. Article 4 of the EU AI Act requires organisations to take measures, to the best of their ability, to ensure sufficient AI literacy.

How it works in practice:

- A short training session per role, with examples from your own day-to-day work.

- A one-page usage rule: what is permitted, what is not and when a person checks.

- Name a contact person for questions and share good examples internally.

Reference: EU AI Act, Article 4 (AI literacy)

Answers: Yes, in place, Partly, No, I don't know. Scoring: Each capability scores 0 to 4 points from two answers (Yes 2, Partly 1, No 0). If one answer is “I don't know”, the status is “to clarify”. Otherwise: 4 points green, 2 to 3 points amber, 0 to 1 point red. Points from all known answers, at most 24. With four or more capabilities at status “to clarify”: “Get an overview first”. Otherwise “Foundations in place” from 18 points, with no red capability and at most one to clarify. “Early stage” up to 8 points or with three or more red capabilities. Everything in between is “Building up”. All red capabilities count equally as the weakest point; if there are none, all amber ones do. No ranking is claimed between capabilities with the same rating. A self-assessment based on your answers. Not an audit, not a security review, not an approval.

**Frequently asked questions**

## Frequently asked questions about the AI check

### What does the AI check assess?

Six capabilities that we consider central to secure and controllable AI operation: using your own data, being able to switch models, knowing your costs, controlling access rights, having at least one application in operation and enabling employees. Two questions per capability.

### Are my answers stored?

Your answers are not transmitted to us. The check runs in your browser; the result is stored only in the link in your address bar and therefore in your browser history. It leaves your device only when you share the link or send it to us by email.

### How do I share my result?

After the check you can copy a result link, save the result as a PDF, copy it as text or open an email draft to Punktexe. The link contains only your answers and the date, no name.

### Is the result an audit?

No. It is a self-assessment based on your answers, not an audit, not a security review and not an approval. It is a good basis for an initial conversation.

## More about our work?

Get in touch, no obligation. How we work with AI ourselves is shown in the [engine room](https://punktexe.de/en/engine-room/).

[Get in touch, no obligation →](https://punktexe.de/en/#contact)
