Close-up of a person reviewing charts and reports in an office setting.

AI Without the Hype: 7 Management Decisions to Make First

ByDaniel Fokker

Introduction

There is a pattern I have seen more and more often in recent years. A management team feels pressure to “do something with AI”. A pilot is launched, someone buys a licence, a working group is formed and three months later everyone is mainly tired. Not because AI cannot deliver value, but because the foundation is missing: clear decisions.

And that is exactly why AI governance gained so much ground in 2025. Not as a brake, but as a way to help projects land faster and more safely.

Decision 1: Where is the value, and where is it not?

AI is not a goal. It is a lever. So start with one simple question: which outcome do we want to improve?

Examples of value I often do see:

  • faster turnaround times in customer contact or back office processes
  • fewer errors in reporting and manual checks
  • better findability of knowledge and decisions

Examples you usually come to regret:

  • “we want a chatbot too” without a goal and without an owner
  • “we want to give everyone Copilot” without agreements and training

Management decision: we choose no more than three measurable value cases. Everything else is parked.

Decision 2: Who is the owner, really?

AI initiatives often die quietly because nobody owns them. Then they are “owned by IT”, or “owned by innovation”. Which usually means: owned by no one.

Management decision: every AI case has one business owner who is responsible for results, risk and adoption. IT is a partner, not the ultimate owner.

This sounds simple. In practice, this is often the decision that suddenly becomes uncomfortable. That is usually a good sign.

Decision 3: Which data may AI touch?

AI without data is a demo. AI with the wrong data is an incident.

The problem is rarely model quality. The problem is that the organisation has not clearly defined what is “sensitive”, what is “internal” and what is “public”. This connects to the wider security discussion around the growing “AI exposure gap” when adoption moves faster than basic control.

Management decision: we define three to four data classes and attach simple rules to them. For example:

  • Public: may be used in a public AI tool
  • Internal: only in approved tools, no copy-paste into public models
  • Confidential: only in protected environments, with logging
  • Highly sensitive: do not use, full stop

Decision 4: What are the rules for use, and how do people learn them?

AI literacy is not a “nice to have”. In the EU AI Act, AI literacy is explicitly part of the first obligations that are being phased in.

But beyond regulation, the point is simple: if you do not give people clear rules, they will invent their own. That always happens. The only question is whether you organise it deliberately or let it happen by accident.

Management decision: we publish a short, one-page AI code of conduct, plus a 45-minute training. Not to turn everyone into an expert, but to prevent damage.

Decision 5: How do we safeguard quality, audit and responsibility?

If AI “makes something up” in an internal memo, that is annoying. If AI makes something up in a decision, a case file or customer communication, that becomes expensive.

That is why CIOs are increasingly investing in guardrails and audit trails, precisely to make speed possible.

Management decision: we choose a quality level for each use case:

  • low impact: light checks, fast iteration
  • medium impact: human review, logging of prompts and outputs
  • high impact: human in the loop, traceability, periodic review

This is governance light: small, but serious.

Decision 6: How do we manage the portfolio, and when do we stop?

AI pilots have a magical talent: they can continue forever without anyone daring to stop them. Because “we are still learning”. True. But learning without a decision moment is a hobby.

Management decision: every AI case gets:

  • a start date
  • a success metric
  • an end date or stop criteria
  • a moment when the management team decides: scale, adjust or stop

If you do not do this, you automatically build an AI zoo. Everyone has one little experiment. No one has responsibility.

Decision 7: What minimum governance do we need to create momentum?

Here is the funny part. Many teams are afraid that governance will slow them down. But in 2025, the movement is the opposite: governance is used to increase speed because it reduces debate, risk and rework.

Management decision: we establish a small “AI cadence”:

  • biweekly review of ongoing use cases
  • decision-making on new intake
  • a short risk check on data, privacy, reputation and security

No heavy committees. No politics. Just rhythm.

Closing: why this works in existing organisations

You do not have to throw away your existing structures, frameworks and processes. You need to adjust them so AI has a place without creating chaos. That is exactly where I am often brought in: when delivery is required and direction is needed at the same time.

If this topic feels familiar, start with these seven decisions. Only then come tooling, vendor selection and architecture.

A short invitation: do you want to get clarity on this in one conversation? A 30 to 45 minute intake is enough to remove the noise.

Similar Posts