Why AI Projects Fail: Start with the Business Problem, Not the Technology

Many organizations are rushing to adopt artificial intelligence (AI), but according to Troy Norcross, AI Governance Advisor and Founder of SERTeam, they are approaching the challenge from the wrong direction.

In a recent episode of Petri Dish, Norcross explained that too many companies are starting with AI and looking for places to use it, rather than identifying a business problem first and then determining whether AI is the right solution.

“What’s my business challenge? What’s my problem? What’s currently wrong? And is AI possibly part of the solution?” Norcross said. “They’re coming with AI is the answer instead of starting with what’s the business problem?”

As AI adoption accelerates across industries, that distinction could mean the difference between measurable return on investment (ROI) and costly failure.

The rise of AI theater

According to Norcross, today’s AI boom resembles previous technology waves, including blockchain and cloud computing. Many organizations are eager to demonstrate innovation but struggle to translate experiments into business outcomes.

He refers to this phenomenon as “AI theater.”

“We had innovation theater,” Norcross explained. “And we see the exact same thing now. I call it AI theater.” He described organizations where different departments independently experiment with AI, creating “AI sprawl” while employees bring their own tools to work through “shadow AI.”

The result is often a fragmented approach to adoption, with no clear governance, strategy, or measurable business value.

Why AI adoption remains difficult

Despite the enthusiasm surrounding generative AI, organizations continue to face major implementation challenges.

One obstacle is that AI systems are fundamentally different from traditional software. Most business applications produce deterministic results. AI models, however, generate probabilistic outcomes.

“AI is completely different than the majority technologies in one very specific way,” Norcross said. “The outcomes are probabilistic instead of deterministic.”

That uncertainty makes business leaders uncomfortable, particularly when decisions depend on predictable and repeatable outcomes.

Cultural resistance is another significant challenge. Employees are often less worried about losing their jobs to AI than they are about changing established ways of working.

Governance is not optional

While much of the AI conversation focuses on data, Norcross believes governance deserves equal attention.

He argues that organizations need guardrails that enable experimentation without introducing unnecessary risk.

“Without governance, AI can really go off the road really fast,” he said. “Good governance, it’s not handcuffs, it’s guardrails to let you start experimenting and rolling out AI in your organisation and go fast safely.”

This becomes particularly important when businesses integrate AI with sensitive corporate data, customer information, or regulated environments.

What does AI-ready actually mean?

Many organizations claim they want to become “AI ready,” but Norcross says the term should have a practical definition.

He identifies three core requirements:

1. Executive-level commitment and communication
2. An AI usage policy
3. Appropriate AI infrastructure and controls

“The first thing is, and most importantly, is senior leadership buy-in and communication,” he said. Leadership teams must clearly communicate that AI will be used to improve the business and that adoption is a strategic priority.

Beyond leadership support, organizations need clear policies covering approved tools, acceptable use, and data protection. Finally, they need technical controls such as AI gateways that manage access to models and datasets while preventing sensitive information from leaving the organization.

Why AI projects fail to deliver ROI

When organizations complain that AI initiatives are not producing returns, Norcross believes the cause is often straightforward.

“They’re just doing anything and everything to say they’re doing AI instead of starting with a tangible specific business problem,” he said.

Hackathons, innovation experiments, and isolated proofs of concept can help familiarize employees with AI. However, they rarely produce measurable business value on their own.

For organizations seeking ROI, Norcross recommends building a business case first, identifying expected outcomes, and only then selecting the appropriate AI technologies.

Watch or listen to the full episode here: