9 out of 10 AI Projects fail: why?

It is likely that the headline of this article has caught your attention: we have now spent a couple of years immersed in the AI revolution, and it appears undeniable that this technology is permanently changing the way we write, study, create images, create videos, analyze data, and, of course, work. Its impact seems unquestionable. Nevertheless, according to a new study, the reality is much less promising, particularly when focusing on how some companies are utilizing it.

The study, entitled The GenAI Divide: State of AI in Business 2025 and published in July 2025 by MIT (Massachusetts Institute of Technology), reveals a surprising truth: despite billions of dollars (or euros) invested, 95% of AI pilot projects are not generating any return on investment.

In fact, the majority do not even reach production. Only 5% of initiatives are delivering real value. This divide, which the report refers to as the “generative AI gap,” is not due to the technology itself but rather to a misguided approach on the part of companies.

Massive implementation, little transformation

Although more than 80% of companies have explored using AI tools such as ChatGPT or Copilot and 40% have implemented them, only a few have managed to achieve a significant impact on their businesses.

The report, based on a systematic review of more than 300 publicly disclosed AI initiatives, structured interviews with representatives from 52 organizations, and survey responses from 153 senior executives, reveals that only two sectors, Technology and Media, are showing clear signs of disruption through AI. In the other seven sectors studied (healthcare, energy, manufacturing, financial services, etc.), structural changes remain completely absent.

In the words of the study’s authors, “these tools primarily improve individual productivity, not bottom-line performance. Meanwhile, enterprise systems—whether customized or vendor-provided—are quietly being rejected. Sixty percent of organizations evaluated these tools, but only 20% reached the pilot phase and only 5% made it to production. The majority fail due to unstable workflows, lack of contextual learning, and misalignment with daily operations.”

As a COO interviewed by the study’s authors explained: “There is a lot of talk on LinkedIn about how everything has changed, but in our operations, nothing fundamental has changed. We are processing some contracts faster now, but that is all.”

5 myths about AI in the business sector

The report debunks several widespread beliefs:

• “AI will eliminate the majority of jobs within a few years.” In fact, the authors state that there are no massive layoffs, and there is no consensus among executives regarding how hiring levels will change in the coming years.
• “AI is already transforming business operations.” Adoption rates are high, but transformation remains rare, particularly in terms of its impact on results.
• “Large companies are slow to adopt.” These are actually the organizations launching the most pilot projects, as they have larger budgets and dedicated teams. The issue is not initial adoption, but rather scaling up: it is precisely the largest companies that struggle the most with moving from pilot to production. The study indicates that mid-sized companies were able to progress from pilot to implementation in roughly 90 days, whereas larger organizations required 9 months or more, with lower success rates.
• “The main obstacles are models, regulation, or legal risks.” The real barrier is that the tools do not integrate well with existing workflows.
• “Leading companies develop their own tools.” In truth, internal developments fail at twice the rate of external solutions.

4 lessons on how companies are really using AI

The learning barrier: when AI is not a partner

The report confirms that the primary barrier is not infrastructure, talent, or regulation, but rather the inability to integrate with business mechanics. Thus, the study explains that the majority of current tools do not retain feedback, do not remember context, and do not adapt well to workflow processes. The gap lies not in intelligence, but in memory, adaptability, and the capacity for evolution.

The result is that, for critical and long-term tasks, humans remain the first choice over AI by a margin of 9 to 1. Meanwhile, findings show that AI has already won the battle for simple tasks: 70% of respondents prefer it for composing emails and 65% for basic analysis.

The economy of Shadow AI

Despite the failures of corporate initiatives, 90% of surveyed employees use ChatGPT, Claude, or other consumer tools with their personal accounts, compared with only 40% of companies that pay for official subscriptions. This phenomenon, termed the “shadow AI economy” or Shadow AI, demonstrates that employees utilize AI daily on their own to automate parts of their work. The study indicates that this shadow AI often delivers a better return on investment (ROI) than formal initiatives.

In any case, for companies, this may represent both a security risk and a missed opportunity: there is interest in the use of AI, but it is not being translated into “official” tools.

Where the true ROI is hidden

The report shows that between 50% and 70% of AI budgets are concentrated in marketing and sales, as these functions have visible metrics that are easy to justify. In these areas, the main tasks carried out by AI include:

• Sending AI-generated emails
• Follow-up automation
• Social sentiment analysis
• Personalized content for campaigns
• Competitive analysis
• Lead scoring

However, the highest-yielding cases documented in the report are found in traditionally overlooked functions such as operations, finance, and procurement. The benefits do not result from massive internal layoffs, but rather from the reduction of external expenditure: elimination of outsourcing contracts, decreased use of creative agencies, and reduced reliance on consultants. In the most advanced cases, companies reported annual savings of between 2 and 10 million dollars in administrative processes and reductions of up to 30% in agency spending.

How to bridge the AI Gap

The report identifies two key factors among organizations achieving real results:

• Agentic Systems: a new generation of AI with persistent memory, iterative learning, and adaptability. These systems are active collaborators that improve with each interaction, rather than mere single-use tools. We are just entering the era of AI agents, but they will increasingly become commonplace, thanks to widespread adoption by tools such as ChatGPT. Although it is still early, the authors warn that the window for adoption will close quickly: companies that do not act now risk remaining permanently on the wrong side of the gap.
• Strategic Partnerships: projects carried out in collaboration with external vendors are approximately twice as successful as internal developments. Companies making meaningful progress treat suppliers as partners, demanding customization and measuring success in terms of tangible business outcomes.

Image: ChatGPT