AI in Enterprise: More Problems Than Solutions

Another day, another report on artificial intelligence (AI) and its enterprise use. This time, the news isn’t good. Analysts at Gartner predict that over 40% of agentic AI projects will be cancelled by the end of 2027. This is a troubling finding, given the rapid proliferation of agentic solutions in the enterprise, which is leaving some users with an agent for every app or platform.

The main causes of these cancellations are escalating costs and unclear business value. According to Anushree Verma, Senior Director Analyst at Gartner, most agentic AI projects are early-stage experiments driven by hype and often misapplied. “Most agentic AI propositions lack significant value or return on investment, as current models do not have the maturity and agency to autonomously achieve complex business goals or follow nuanced instructions over time,” Verma says.

Despite these challenges, Gartner predicts that by 2028, one-third of all enterprise software will include agentic AI, with 15% of replicable, day-to-day decisions made autonomously by the technology. However, this year, 40% of all projects will be scrapped.

Jamie Beckland, Chief Product Officer at APIContext, offers a different perspective. He believes that most AI projects are failing not because the technology doesn’t work, but because the APIs they rely on weren’t built to support agentic usage. “Most enterprise APIs today are still optimized for web and mobile apps, not autonomous agents,” Beckland notes.

This view aligns with a Global Enterprise AI study from SS&C Blue Prism, which surveyed 1,650 CEOs, CTOs, and senior IT leaders. The study found that while 92% of them are using AI to transform business operations, 55% admitted to seeing little benefit. Meanwhile, research from asset management company Bowmore revealed that AI-focused investment funds were losing money year on year, with average returns of -3.3 percent.

Financial commentators are now scenting trouble. The Economist described AI valuations as “verging on the unhinged,” with companies valued on “momentum” rather than metrics. The Financial Times noted that AI returns have yet to justify the investment mania. Bank of America Securities predicts that spending on data centers will jump from $333 billion last year to about $1 trillion in 2030, with 83% of the money going into AI-related investments.

Microsoft’s AI revenue is barely five percent of its predicted total for this year, while OpenAI’s subscription revenues have doubled year on year to $10 billion. However, this is a fraction of the supporting capital expenditure. Microsoft CEO Satya Nadella has emphasized the need for AI to prove its value in areas like healthcare, education, and productivity. “AI must generate measurable value for its vast energy consumption and carbon cost,” Nadella said.

However, AI is largely being deployed as shadow IT by individuals who are using it to cut corners, rather than solve the world’s most urgent problems. Two recent studies highlight this trend. The AI at Work report from Boston Consulting Group (BCG) surveyed over 10,000 users in 11 markets and found that 54% of employees are happy to use unauthorized AIs at work. Younger employees (Gen-Z and Millennials) are the most likely to bypass enterprise restrictions (62%).

A report from KnowBe4, a human risk management and cybersecurity platform, found similar results. It surveyed users in the US, UK, Germany, South Africa, the Netherlands, and France, and found a 60% adoption rate for AI in the workplace. However, only 17% of these employees use apps with their IT and security teams’ knowledge, and 10% input privileged client data.

These findings reveal an inconvenient truth: so-called enterprise AI adoption is still, at heart, individual employees using ChatGPT unsanctioned, with scant regard for data integrity and security. This is hardly a recipe for measurable ROI.

Academic research also highlights AI’s cognitive limitations. A study by Laban et al. found that LLMs struggle in multi-turn instruction settings, with performance dropping by an average of 39% across six generation tasks. This performance degradation is due to a minor loss in aptitude and a significant increase in unreliability.

In conclusion, while AI has the potential to transform business operations, the current reality is that many projects are failing to deliver measurable ROI. The technology must mature and be deployed strategically to realize its full potential.