Doomed enterprise AI projects usually lack vision

Failed implementations of AI technologies are pushing CIOs to step back and try to better understand the technology and its impact before moving ahead, according to analysts and industry experts.

While AI commitment in enterprises remains high, the success rate of prototyping to production is low, research firm Omdia said in a survey released last week.

Proof-of-concept failure “is not usually because of inherent defects in the AI technology being tested, but because enterprises and vendors do not understand the complexity of what AI deployment involves,” Omdia said.

The survey found that AI experimentation is the bastion of cash-rich companies. About 58% of those surveyed have between six and 50 AI projects in experimental phase, and only 4% have more than 100 AI experimental projects. Enterprises with less than $100 million in revenue are prototyping fewer than five AI projects. 

But the failure rates of these early efforts are high. Only 10% of the companies surveyed achieved more than a 40% success rate; 37% saw between 11% and 40% of their projects reach production; and 21% reported a success rate of between 5% and 10%. The rest saw fewer than 5% of their prototype projects reach production.

“This points to a mixed, nuanced picture for [proof-of-concept] progress — a bifurcation rather than universal failure, where many enterprises are successfully transitioning from AI PoCs to production while others are still clearly struggling,” Eden Zoller, chief analyst of applied AI said in a blog post on the Omdia AI Market Maturity 2025 survey.

CIOs and other IT decision-makers are under pressure from boards and CEOs who want their companies to be “AI-first” operations; that runs the risk of moving too fast on execution and choosing the right projects, said Steven Dickens, principal analyst at Hyperframe Research.

Smart leaders are cautious and pragmatic and focused on validated value, not jumping the gun on mission-critical processes. “They are ring-fencing pilot projects to low-risk, high-impact areas like internal code generation or customer service triage,” Dickens said.

CIOs should first develop a data strategy and transform the data pipeline before prioritizing the release of a cool new chatbot or app. “The ones who rush often skip the crucial steps of governance and data preparation, leading to costly reworks later,” Dickens said.

Generic large language models (LLMs) are not optimized for organizations, which need to supplement their own data, said Jack Gold, principal analyst at J. Gold Associates. “In many companies, that is a hard thing to do, as the data may be scattered or not easily accessed to fine tune the models to achieve maximum success,” he said.

In a separate AI market maturity survey released this month, McKinsey found that 90% of respondents rely on AI in some form. The highest use is in the insurance industry for information management and service operations, followed by software engineering in the tech sector. AI is also popular in the services sector for information management, and in marketing and sales operations in the consumer goods market.

The areas where AI use has been light include advanced manufacturing, engineering and construction, and pharmaceutical and medical sectors.

Digging deeper, agentic AI is most widely used in the tech sector for software engineering and service operations. IT and knowledge management agents are popular across a broad range of sectors, while inventory management and manufacturing agents are the least used. Surprisingly, HR agents are not widely used across sectors, either.

“Agentic use cases such as service-desk management in IT and deep research in knowledge management have quickly developed,” the management consulting firm said in its study.

In this experimental period, organizations viewing AI as a way to reimagine business will take an early lead, Tara Balakrishnan, associate partner at McKinsey, said in the study. “While many see leading indicators from efficiency gains, focusing only on cost can limit AI’s impact,” Balakrishnan wrote.  

Scalability, project costs, and talent availability also play key roles in moving proof-of-concept projects to production.

AI tools are not just plug and play, said Jinsook Han, chief strategy and agentic AI officer at Genpact. While companies can experiment with flashy demos and proofs of concept, the technology also needs to be usable and relevant, Han said.

“Everyone can go out to the beach and make sandcastles, but how many are here to stay?” Han said.

Many AI projects fail because they are built atop legacy IT systems, Han said, adding that modifying a company’s technology stack, workflows, and processes will maximize what AI can do.

Humans also still need to oversee AI projects and outcomes — especially when agentic AI is involved, Han said. “Let the machines do what machines do best and let the humans do what humans do best,” Han said.

In a separate survey, AI vendor Cleanlab found very few companies had AI agents in production stage. “Between 60% and 70% of everyone that we chatted with, both in survey and also in sales calls, they were changing their entire stack — their LLM, the AI stack that they built an agent on — they’re just tinkering, every three months at least,” said Cleanlab CEO Curtis Northcutt.

He empathized with constrained CIOs operating under pressure and lacking budget and expertise in specialized areas critical to building solid AI systems. “The reality is that real AI agents that are agentic and have tool calling … is probably [not arriving until] early 2027,” Northcutt said.

Northcutt and Gold said organizations should partner with companies who have been successful already. “They’ve seen lots of failures and can point out the pitfalls as you implement, saving time, resources, and ultimately affecting success rates — especially for first time implementers,” Gold said.