Implementing AI is not an IT project

A director calls me. “We ran an AI pilot. The model works. But it is not going live.”

I hear this story at least twice a month. And it is almost never a technical problem.

The model is trained. The accuracy is good. The demo was impressive. But there is no policy. No DPIA. No communication plan for employees. No idea how it fits within existing processes. And so it stalls.

Where it goes wrong: the classic trap

The pattern is familiar. IT or an innovation team kicks off a pilot. A use case is chosen, data is gathered, a model is built. After three months, a working prototype is up and running.

Then come the questions nobody asked:

• Does this fit within our privacy policy? Has a DPIA been carried out?
• Do we comply with the EU AI Act? Which risk category does this fall into?
• Who is accountable if the model makes a wrong decision?
• Have employees been brought on board? Do they understand what the model does, and what it does not?
• How do we monitor the output? Who pulls the emergency brake?

The answer is usually: “We will sort that out later.” But later never comes. The pilot dies a quiet death in a Confluence page that no one opens again.

AI is an organisational change

The heart of the problem: AI is treated as a technology project, when it is an organisational change.

Technology is the easy part. Building a model or connecting an API is something a good team can do in weeks. But preparing the organisation for AI? That touches policy, processes, culture, compliance and communication. And that takes time, attention and leadership.

Compare it to introducing a new ERP system. Nobody would implement SAP without change management, training and process adjustments. But with AI we think: “It is a clever bit of software, we will just bolt it on.”

That does not work.

What does work: three pillars from day one

In every successful AI implementation I have guided, from enterprise to SME, three things were present from the start:

1. Governance: the rules of the game before you begin

Governance is not bureaucracy. It is clarity. Who is allowed to deploy which AI? For which tasks? With which data? And who is accountable?

At Rabobank (5,000+ engineers) I saw what happens when you set up governance only after the fact: projects that run months behind, legal reviews that grind pilots to a halt, and teams that grow frustrated because they do not know what is allowed.

Start with an AI policy, a risk classification and a register. It does not have to be perfect, it just has to exist. Three documents. An afternoon's work. And you avoid months of delay further down the line.

2. Compliance: build in the GDPR and AI Act, do not bolt them on afterwards

The EU AI Act is not a distant prospect, the first rules are already in force. Prohibited AI practices have been punishable since February 2025. High-risk applications (think credit scoring, HR screening, medical decision-making) require documentation, transparency and human oversight.

The GDPR sets additional requirements for automated decision-making. A DPIA (Data Protection Impact Assessment) is mandatory when you use AI for profiling or for decisions that affect people.

This need not be a showstopper. But it has to run in parallel with development, not after it. If you discover after three months of building that your use case is a high-risk application, you are effectively starting over.

3. Adoption: bring employees along from start to finish

The best AI solution is worthless if no one uses it. Or worse: if employees distrust it and work around it.

AI literacy is not a luxury, it is a precondition. Teams need to understand what AI can do, what it cannot do, and why it is being deployed. Not with a PowerPoint after launch, but with hands-on workshops and guidance during the project.

From the boardroom to the shop floor. The leadership needs to understand the risks they are taking. The team leader needs to know how the daily work changes. The employee needs the confidence that the AI helps them, rather than replacing them.

From practice: a pilot that stalled, and how it succeeded after all

A mid-sized financial institution wanted to use AI to automate document analysis for mortgage applications. The technical team built a working model in eight weeks. Accuracy: 94%. The demo went well. Management was enthusiastic.

Then came the questions from compliance. How do we handle personal data in the training set? Does this count as high-risk under the AI Act? Where is the DPIA?

The result: the pilot was put on hold. Four months of standstill while legal, privacy and compliance caught up on what should have happened beforehand.

The breakthrough only came when governance and adoption were brought in in parallel:

• A DPIA and risk classification were carried out after all, in two weeks
• An AI policy was put in place with clear boundaries for document processing
• Mortgage advisers got workshops on how the model works and where its limits lie
• A feedback loop was set up so employees could report errors

The checklist: AI as organisational change

Before you start an AI project, ask yourself these questions:

• Governance: Is there an AI policy? Does everyone know what is and is not allowed?
• Compliance: Which risk category does the use case fall into? Is a DPIA needed?
• Adoption: Are the people who will work with it involved? Do they understand it?
• Accountability: Who owns it? Who monitors it? Who escalates?
• Communication: Do stakeholders know what is coming and why?

If you answer “no” or “I am not sure” to more than two questions, you are not ready to build yet. Start there. The technology can wait, the organisation cannot.

Conclusion: do not start with the technology

Implementing AI is not an IT project. It is an organisational change that touches technology, policy, compliance and people. The organisations that get this, the ones that build in governance, compliance and adoption from day one, are the organisations where AI actually runs in production.

The rest have pilots. Nice demos. And a Confluence page that no one opens again.

Want to talk through how to implement AI as an organisational change? Book a no-obligation call.