AI is no longer something we just talk about. Most organizations have either started experimenting – or are about to – with how AI can create business value.

But something tends to happen fairly quickly.

What initially looked like a “typical” development project starts behaving differently. Plans become uncertain. Expectations diverge. And what seemed simple in a demo turns out to be significantly harder in reality. After more than 20 years working at the intersection of business and technology – and in recent years increasingly with AI initiatives, from workshops to full implementation – I’ve seen a clear pattern:

AI projects rarely fail because the technology is too complex. They fail because we try to run them like something they’re not.

Here are four challenges that often determine whether you succeed or get stuck along the way.

1. You can’t define AI requirements the same way as traditional development

In many projects, teams start the way they always have: defining requirements, scope, and expected outcomes early on. That works well until AI enters the picture.

In one project, I worked with a client to explore how AI could support a complex business process. Early on, the organization wanted clear answers:

“What will the model be able to do?”
“How accurate will it be?”
“What output will we get?”

The problem? No one could give definitive answers. At best, we could define hypotheses and goals. Instead, we had to shift the entire way of working; from planning to exploration, from requirements to experimentation. We ran workshops, built simple prototypes, and tested ideas. Some worked immediately. Others failed completely. This is where many AI projects struggle. They require a more experimental, iterative approach, something that can feel uncomfortable for organizations used to predictability and control.

2. ROI is harder to calculate when the landscape keeps changing

In leadership teams, one question always comes up early: “What’s the business case?” It’s a fair question. The challenge is that AI doesn’t follow the same rules. In one engagement, we assessed AI as part of a product strategy.

We had a clear initial ROI calculation but within months, the assumptions had changed:

• New models were released
• Costs shifted
• Some use cases became significantly easier to solve

Our original business case quickly became outdated.

This happens often. Organizations want long-term, stable investment decisions while the technology evolves at a pace that makes decisions obsolete before they’re even implemented. A more effective approach is incremental: start small, test, learn, and scale once value becomes clear.

3. “We can build it ourselves” – but should we?

One of the biggest shifts in recent years is how easy it has become to get started. In many organizations, someone – often highly driven – quickly builds a solution using an API or an AI tool. And it can be impressive.

That’s usually where the discussion begins:

“Do we really need external support?”

Sometimes, the answer is no.

But more often than not, challenges appear when the solution needs to scale or integrate into real operations. Common issues include:

• insufficient data quality
• poor integration with existing systems
• rising costs as usage increases
• late-stage security concerns

At that point, teams often need to step back and rebuild more structurally. The gap between a prototype and a production-ready solution is larger than many expect.

4. Lack of a shared understanding of what AI actually is

One of the most common challenges isn’t technical – it’s organizational alignment.

Within the same initiative, you often see three perspectives:

• Leadership sees opportunity and wants to move fast
• Developers see complexity and risk
• The business struggles to understand practical impact

In one AI initiative I worked on, we spent significant time upfront aligning on what AI is, and just as importantly, what it is not. That alignment proved critical.

When organizations share a common understanding, decisions improve. Priorities become clearer. Execution speeds up. Without it, the opposite happens.

What does this mean in practice?

AI creates enormous opportunities – but it also requires us to rethink how we work. In my experience, the organizations that succeed are not the ones starting with the most advanced technology.

They are the ones that:

• embrace iteration
• accept uncertainty early on
• build shared understanding
• know when to move from experimentation to structure

AI projects are not like other projects. And that’s exactly what makes them both exciting and challenging to get right.