GuideNovember 20266 min readAI & work

Twelve questions for boards and managing directors

A framework for your next meeting on AI, with pointers on what good answers look like

At a glance

  • For decisions on AI, senior leadership mainly needs a framework for testing initiatives and recording the basis of its decisions.
  • The twelve questions follow the five areas of the AI delta and set out, for each one, what a good answer and a warning sign look like.
  • Seeing where answers are missing matters more than having an answer to every question, because that's where the next task lies.

Boards and managing directors don't need to know how a language model works. They should know which questions to ask, and be able to tell whether an answer holds up. This guide contains twelve questions, grouped by the five areas in which companies close the AI delta, from exploring to enabling. For each question we set out what a good answer looks like and what counts as a warning sign.

The framework is meant for the management board of a public company just as much as for the managing directors of a mid-sized business. Supervisory or advisory boards can use it to get a report on the state of AI initiatives.

Why good questions are part of due care

German company law requires the management board of a stock corporation to apply the care of a diligent and conscientious manager (Section 93(1) AktG). Under the same provision, there is no breach of duty if, when making a business decision, a board member could reasonably assume they were acting on the basis of adequate information and in the company's best interests. For the managing directors of a GmbH, Section 43(1) GmbHG requires the care of a prudent businessperson.

With AI initiatives, adequate information is the hard part. The technology changes quickly, vendor claims rarely relate to your own data, and some consequences only show up in operation. Questions that are asked before a decision, with the answers recorded, are therefore more than good practice. They document the basis on which the decision was made.

Then there's the AI Act. Since the Omnibus amendment in July 2026, Article 4 requires providers and deployers to take measures to support AI literacy. For senior leadership, that's a good reason to start with its own knowledge.

This section is a general overview and does not constitute legal advice.

The twelve questions

Explore

1. What can AI do today in our most important processes, and how do we know? A good answer points to our own tests on our own data, with a date and a result. A warning sign is an assessment based solely on demos or vendor claims.

2. Which development over the next twelve months would change our plans? A good answer names two or three specific developments and the person tracking them. A warning sign is "we're following that closely" without a name or a date.

Decide

3. Which three initiatives offer us the greatest benefit, and how do we measure it? A good answer names three initiatives, each with a business metric and a business case whose assumptions are disclosed. A warning sign is a long list of ideas of equal rank.

4. Which initiatives are we deliberately not pursuing? A good answer is a short list with reasons, decided and communicated. A warning sign is being told there aren't any, because everything is meant to happen at some point.

5. Which AI systems do we use, and in what role under the AI Act? A good answer refers to a register that also covers AI features in purchased software, with role, risk class and owner. A warning sign is a register that only knows the systems built in-house.

6. Who may decide that a system acts without human approval? A good answer names who decides, the criteria for the classification and a probationary period before an agent is allowed more. A warning sign is "the project team decides".

Build

7. What's stopping our successful pilots from becoming the standard? A good answer names the specific obstacle for each pilot, such as a missing interface, along with measured quality and cost per case. A warning sign is being told that only budget is missing.

8. How do we make sure confidential data stays protected? A good answer describes rules on which data may go into which tools, enforced technically and checked regularly. A warning sign is a policy nobody knows is being followed.

Embed

9. Who runs the solutions once the project ends? A good answer names a team with a budget for operation and a person in the business who answers for the benefit. A warning sign is "we'll sort that out after the pilot".

10. How is AI changing the tasks in our most important areas, and how are we involving employees and the works council? A good answer describes for each solution who will do what in future and gives a timetable for informing the works council before the process is fixed. A warning sign is the assumption that nothing changes for employees.

Enable

11. Who in senior leadership is responsible for AI, and can that person assess initiatives themselves? A good answer names a person with a clear mandate who works with AI regularly themselves. A warning sign is responsibility delegated entirely to IT, with senior leadership only receiving reports.

12. Which skills do we build ourselves, and which do we buy in? A good answer makes this decision role by role and links it to learning on real tasks, with documented delivery. A warning sign is mandatory training for everyone, followed by nothing.

How to use the framework

The questions work well for a meeting in which senior leadership reviews the state of its AI initiatives. The process has five steps.

Figure 1How to use the framework in the meeting
  1. 01CirculateSend the questions to the owners of the initiatives in good time before the meeting.
  2. 02Answer in writingShort answers with evidence, no slide presentation.
  3. 03Mark open pointsIn the meeting, record for each question whether the answer is evidenced, partly evidenced or open.
  4. 04Assign tasksAn owner and a deadline for every open question.
  5. 05RepeatAsk the same questions a quarter later and compare the answers.

Illustration: Paice

Three ratings are enough. An answer is evidenced if it points to something that can be checked: a test result, a register, a named person, a measured metric. It is partly evidenced if the direction is right but the proof is missing. It is open if nobody can answer or the answer consists of intentions. The written answers and the ratings go into the minutes. That way, as a by-product, you document the basis on which senior leadership made its decisions.

Seeing where answers are missing matters more than having an answer to every question. That's usually where the next task lies. If the open questions cluster in one of the five areas, that also shows where your company is held back most.

How to tell it's working

After two or three rounds, the answers should get shorter and the evidence more specific. The number of open questions falls, and the remaining ones shift from insight towards adoption. That's a good sign. It means initiatives are finding their way into operation and senior leadership is discussing the questions that arise there.

Sources

Philipp Pöschke

Author

Philipp Pöschke

Founder, Paice

Philipp Pöschke founded Paice in Bremen in 2026. He is personally accountable for every engagement and works hands-on on analyses, prototypes and solutions. He is the author of the articles in Insights.

Get in touch