Why most AI products fail: Key findings from MIT’s 2025 AI Report

MIT published a report on the current state of AI in business this week, one that highlights a big disconnect between AI investment and business returns.

The MIT report, The GenAI divide: The state of AI in business 2025, points out that US business has so far invested between $35 billion and $40 billion in generative AI with very little to show for it, with 95% of generative AI pilots yielding no business impact or tangible P&L outcomes and only 5% of organisations successfully integrating AI tools into production at scale. The report proved so influential that it stoked nascent fears of an AI investment bubble and the price of tech stocks like Nvidia tumbled after its publication.

The learning gap

The core problem is a “learning gap”, say its authors. This means that rather than poor models or insufficient infrastructure, the failure lies in enterprise systems that don’t adapt, don’t retain feedback, and don’t integrate into workflows, and AI tools often become static “science projects” rather than evolving systems. "Most GenAI systems do not retain feedback, adapt to context, or improve over time," the report said.

Aleksas Drozdovskis, a member at The International Academy of Digital Arts and Sciences, commented: “The report's finding that 95% of organisations are getting zero return from their GenAI pilots doesn't surprise me. In my experience, most GenAI pilots (unless they're for well-understood use cases like customer service productivity) are disconnected from real workflows and lack meaningful feedback loops. It often seems that teams launch these proof-of-concepts to "tick an innovation box" rather than genuinely solve pressing business problems.”

Individual productivity improvements

Tools like ChatGPT and Copilot are widely adopted in businesses, the report said, but they are used primarily to improve individual productivity, not P&L performance. Many employees are using shadow AI tools independently, with over 90% of workers reporting that they use personal AI tools despite low corporate adoption. This shows a fundamental tension, according to the report, as this shadow usage creates a feedback loop: employees know what good AI feels like, and are less tolerant of static enterprise tools.

The report also pointed out: “ChatGPT's very limitations reveal the core issue behind the GenAI Divide: it forgets context, doesn't learn, and can't evolve. For mission-critical work, 90% of users prefer humans. The gap is structural, GenAI lacks memory and adaptability.”

The report said that “enterprise-grade systems, custom or vendor-sold, are being quietly rejected”. It found 60% of organisations evaluated such tools, but only 20% reached pilot stage and just 5% reached production. Most failed due to brittle workflows, lack of contextual learning, and misalignment with day-to-day operations. The report quoted one unnamed CIO as saying: “We've seen dozens of demos this year. Maybe one or two are genuinely useful. The rest are wrappers or science projects."

Successful pilots

Successful pilots were found to have approached AI procurement differently, according to the report’s authors. They demand deep customisation, hold vendors accountable to business metrics and understand that partnership, not just purchasing is needed. The report said: “…one insight was clear: the most effective AI buying organisations no longer wait for perfect use cases or central approval. Instead, they drive adoption through distributed experimentation, vendor partnerships, and clear accountability. These buyers are not just more eager, they are more strategically adaptive.”

Disruption is currently only apparent in two industries – technology, and media and telecom – the report finds. For other industries – the report names professional services, healthcare and pharma, consumer and retail, financial services, advanced industries, and energy and materials – generative AI has had little impact beyond experimentation.

Guido Lonetti, Head of Product at fintech dLocal, says that the company has fully deployed a GenAI Assistant platform that has transformed how it operates. He said: “At dLocal, there is no debate about the value of AI. Beyond boosting productivity and operational efficiency, AI allows us to build products that solve our merchants’ problems 10 times better, smarter, and faster. The real challenge in becoming an AI-first company lies not in technology, but in people. Every leader at dLocal must make a mindset shift, re-imagining their day-to-day work through the lens of AI.”

Key takeaways from the report for product people are:

• Focus on integration because success comes from embedding AI into workflows, not just deploying models
• Monitor shadow AI usage because it shows unmet needs
• Design AI to adapt and improve and retain context
• Partner with proven providers

The report's authors say it is based on a multi-method research design. This included a systematic review of over 300 publicly disclosed AI initiatives, structured interviews with representatives from 52 organisations, and survey responses from 153 senior leaders collected from four major industry conferences.