# AI governance for business leaders

A 10-minute guide for business leaders on governing AI, so it gets from pilot to production. Who owns it, whether your data is ready, and which guardrails it needs.

Has your AI pilot impressed everyone, but never made it into daily work? The model is rarely the reason. AI stalls when the data isn't ready, the rules live in a prompt and nobody owns the result. This guide is for the business leaders who decide on AI. It shows what to sort out first, and how to start AI governance.

## Why this matters now

AI already runs in your business. The rules for it now apply, and the organisations that govern it well get the most from it.

### The rules for AI now apply.

Since 2 August 2026, the EU AI Act requires AI that talks to customers to say that it is AI. Since 15 August 2026, the Dutch Cyberbeveiligingswet makes the boards of organisations in its scope approve their security measures. The AI Act's rules for high-risk systems follow on 2 December 2027.

### AI is already in the tools you use.

Software suppliers are switching on AI features inside the tools your teams work with every day. Often nobody has decided whether that is allowed, or what happens to the data.

### A demo is easy. Production is hard.

An AI demo runs on sample data. Production runs on your real product, price and customer data, with the controls, owners and fallbacks that any business-critical system needs.

## Our AI manifesto

Five principles guide how we use AI, so that it is practical, responsible and built for the long term:

- Founded in engineering
- Designed for people
- Innovation guided by learning
- Transparent and responsible
- Positive impact

[Read our AI manifesto](https://www.labdigital.nl/culture/ai-manifesto)

## Two reasons AI stays stuck in the pilot

**Problem 01**

### The data is not ready

A demo runs on a clean sample. Production runs on your real product, price and customer data, with its gaps, duplicates and old records. Many AI projects stall at the moment the sample data is replaced with the real thing.

Treat data readiness as your first AI project. For each use case, find out where the data comes from, who owns it and how complete it is, before you choose a model. Then give AI one way in to that data. A GraphQL API is one approach we use, so AI reads the same data as your website and apps.

Build the proof of concept first, and find the data problems when you try to go live.

**Problem 02**

### The rules are written in a prompt

A language model interprets instructions. It does not follow them the way a program does. A prompt that says "never show margin data" is a request, not a control. The risks that matter most are small mistakes that repeat at machine speed.

Put the limits in the system itself. The AI only gets the tools and data it needs, it can only run actions you have approved, and a person checks the result where it matters. We give AI agents those approved actions as MCP tools, generated from GraphQL operations a developer has reviewed.

Give the AI broad access, and describe in the prompt what it must not do.

![Two Lab Digital colleagues planning a strategy session](https://www.labdigital.nl/media/658E6K6cPkikodWncBq2v8-lab-digital-210622-129-9d4855.jpg)

## Five questions before you scale AI

We ask the same five questions of every AI idea. It does not matter whether the idea is a chatbot, an agent that processes orders or a tool for your own teams.

Each question has a good answer. If you cannot give that answer yet, that is the work to do before you start writing prompts.

The guardrails behind these answers are built into your systems. Our guide to [AI architecture](https://www.labdigital.nl/insights/topics/ai-architecture) explains how.

### 1. Value: does it add value you can measure?

A good answer: you know what it saves or earns, and how you will measure it.

### 2. Data: is your data correct, accessible, structured and owned?

A good answer: product, price and customer data are complete, up to date and owned by someone.

### 3. Guardrails: what happens when it goes wrong?

A good answer: the AI can only do what you allow, a person checks where it matters, and you can switch it off.

### 4. Governance: who owns the process once AI runs it?

A good answer: one owner, clear rules, and a plan for when it gets something wrong.

### 5. People: is your organisation ready?

A good answer: the people who use it helped shape it, and know what changes for them.

## Governed AI is AI you can trust

Independent research points the same way as our own projects. The organisations that get value from AI are the ones that have decided who owns it, what it may do and where it runs.

**McKinsey**

### CEO oversight of AI governance is linked to bottom-line impact

In McKinsey's 2025 State of AI survey, CEO oversight of AI governance was one of the factors most closely linked to the profit organisations reported from generative AI.

- **28%** of organisations using AI say their CEO oversees AI governance

[Read the McKinsey report](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-how-organizations-are-rewiring-to-capture-value)

**IBM**

### Unapproved AI makes breaches more likely and more expensive

IBM's 2025 Cost of a Data Breach report found that breaches linked to unapproved 'shadow' AI cost more, and that most breached organisations had no rules for AI.

- **1 in 5** organisations had a breach linked to shadow AI

- **63%** of breached organisations had no AI governance policy, or were still writing one

[Read the IBM report](https://www.ibm.com/reports/data-breach)

**Cisco**

### The most AI-ready organisations scale more of their pilots

Cisco's 2025 AI Readiness Index, a survey of more than 8,000 AI leaders, found that only a small group is fully ready for AI. That group leads on every dimension, including data, security and governance.

- **13%** of organisations are fully ready for AI

- **3x** more likely to scale their AI pilots

[Read the Cisco research](https://investor.cisco.com/news/news-details/2025/Cisco-AI-Research-The-Most-AI-ready-Companies-Outpace-Peers-in-the-Race-to-Value/)

## AI that is live, not a pilot

Libra Energy, part of Rensa Family Group, sells solar and battery equipment to installers. Since summer 2026, an AI agent has turned emailed requests from installers into shopping lists they can purchase.

**Case study**

### AI-assisted order intake at Libra Energy

Installers email their requests for a quote as PDF or Excel files. An AI agent reads each one, matches the lines to the product catalogue and creates a shopping list. An account manager checks the list before the reply goes out. The agent uses the same gateway and permissions as the webshop, so it can do only what the webshop can do.

- **<1 min** to turn an emailed quote into a shopping list, down from ~30 min

[Read the case study](https://www.labdigital.nl/work/libra-energy)

## How to start AI governance

Governance is not one more initiative. It is the framework the rest of your AI work runs in. If you already work to NIS2 or ISO 27001, you know most of the method: know your assets, name an owner, assess the risk, handle incidents and check your suppliers.

1. Name an owner. One person in the business owns AI governance, with sign-off from the board.
2. Make a register of your AI systems. Include the AI features your suppliers have switched on in the tools you already use.
3. Check the data each AI system uses. Is it correct, up to date and owned by someone?
4. Write an AI policy. Cover acceptable use, who can access what, and how you assess suppliers.
5. Run governance alongside your first AI projects, not after them.

Whoever owned a process before AI should still own it afterwards. Your security team can lend the method, but the business keeps the decision.

## Find out where your organisation stands

Our AI readiness session is a four-day, fixed-scope review of your data, platform, team and governance. You get a shortlist of the AI use cases worth building first and a 90-day plan with owners.

[See the readiness session](https://www.labdigital.nl/services/ai-readiness-session)
