Many companies are eager to jump into the world of AI. They see it as a way to stay ahead and boost their business. But recent research shows that most AI projects never make it from a test phase to real use. Companies often rush into deploying AI without fully understanding what it takes to make it work well.

AI Experiments Are Common, Success Is Not

According to a recent survey by the research firm Omdia, many large companies are experimenting with AI. About 58% have between six and 50 AI projects in the testing phase. However, only a small number—around 4%—have more than 100 projects in this stage. Smaller companies with less than $100 million in revenue tend to test fewer than five projects.

The problem is that most of these early efforts don’t succeed. Only 10% of companies reported that more than 40% of their AI projects went into production successfully. Around 37% had 11% to 40% of their projects succeed. But many companies see fewer than 5% of their prototypes make it into real use. This shows that while many companies are trying AI, few are turning those efforts into actual tools they can rely on.

Understanding the Challenges Behind AI Failures

Experts say that failures are often not due to bad AI technology. Instead, they happen because companies don’t fully understand what deploying AI involves. Steven Dickens from Hyperframe Research explains that rushing into AI projects without a clear plan can cause trouble. He suggests focusing on low-risk areas like customer service or internal tasks first.

One key step is developing a good data strategy. Many companies try to jump straight to creating AI tools like chatbots without preparing their data properly. This often leads to costly reworks later. Jack Gold from J. Gold Associates points out that large language models (LLMs), which many companies use, aren’t always tailored for specific organizations. Companies need to add their own data to make these models work better, but that can be hard if their data is scattered or hard to access.

What’s Holding Back AI Adoption?

Many companies are using AI in some ways. A McKinsey survey found that 90% of respondents use AI in their business. It’s especially common in insurance for managing information and in software engineering. Other sectors like services, marketing, and sales also use AI, but areas like manufacturing, engineering, and healthcare are still slow to adopt.

One interesting trend is the use of agentic AI—AI systems that can act on their own and call tools. These are most common in tech companies for tasks like software development and service management. Surprisingly, HR departments are not using AI agents much.

Despite the enthusiasm, many organizations are still in the experimental phase. Tara Balakrishnan from McKinsey says that companies seeing early wins are those reimagining their entire business with AI. But focusing only on cutting costs can limit AI’s potential. To truly succeed, companies should consider scalability, costs, and how much talent they have to manage these projects.

Many AI projects fail because companies try to implement them on outdated systems. Jinsook Han from Genpact highlights that integrating AI into existing workflows and technology stacks is crucial. She emphasizes that humans still need to oversee AI outcomes, especially with more autonomous systems.

Another obstacle is the rapid pace of tinkering. Curtis Northcutt from AI vendor Cleanlab says many companies frequently change their AI stacks, which delays progress. He predicts that fully autonomous AI agents with advanced tool-using capabilities might not arrive until early 2027.

The best approach, say industry experts, is to learn from others who have already faced these challenges. Partnering with experienced companies can help new adopters avoid common pitfalls. They can provide valuable insights that save time, money, and frustration in the long run.

In the end, AI has huge potential, but success requires understanding its complexity, planning carefully, and working with the right partners. Companies that do this will be better positioned to turn AI from a risky experiment into a real driver of growth.

Inspired by

• https://www.computerworld.com/article/4091967/doomed-enterprise-ai-projects-usually-lack-vision.html

Sources

• gmpg.org
• informa.com
• mckinsey.com
• mckinsey.com
• wsj.com