80% of AI projects failed in 2024. Why 90% could fail next—and how to fix it.

80% of AI projects failed in 2024. Here's why that figure will exceed 90%, and some ideas for how to remedy the situation.

‍

Here's a statistic that might make you pause: in 2024, eight out of ten AI projects failed to deliver on their promises. That's according to a comprehensive five-year study by the RAND Corporation, research that should give us all pause as we navigate this AI transformation together.

If you're feeling a mix of concern and curiosity right now, you're not alone. We've seen this pattern up close, working with organizations that poured months of effort and significant resources into AI initiatives, only to find themselves back at square one. But here's what we've learned: failure isn't inevitable, and success has less to do with technology than you might think.

‍

The reality behind the numbers

Let's put this 80% failure rate in perspective. In traditional IT projects, about 40% fail, already a substantial challenge. But AI projects are failing at twice that rate, despite unprecedented investment and attention.

For large organizations, the financial impact is significant: failed AI projects typically cost between 5-15 million Euro. But the hidden costs may be even more damaging, teams losing confidence in innovation, talent becoming skeptical about change, and organizations becoming more risk-averse just when boldness could set them apart.

These aren't just statistics. They represent real people who believed in a vision, worked hard to make it happen, and watched their efforts fall short.

‍

What we're learning from the research

The RAND study isn't based on speculation or quick surveys. Over five years, researchers spoke with 65 experienced AI professionals (data scientists, engineers, and architects) who've been in the trenches, each with multiple projects under their belt. Their insights reveal five patterns that, when combined, explain most failures:

1. When problems and solutions don't align

Too often, business teams and technical teams are solving different problems without realizing it. We see this frequently: a company launches a chatbot to reduce call volume, when the real issue is that their product is genuinely confusing to use. The technology works perfectly, but it's addressing a symptom rather than the cause.

2. The data foundation challenge

AI learns from data, but organizations often discover too late that their data is incomplete, biased, or simply insufficient for what they're trying to achieve. It's like trying to teach someone to cook by showing them only pictures of finished dishes, technically possible, but missing crucial information.

3. Falling in love with technology

There's something exciting about implementing the latest, most sophisticated AI model. But sometimes a simpler solution would work better for the actual problem at hand. We've learned to ask: "What's the simplest approach that could work?" before exploring more complex options.

4. Infrastructure reality check

Even the most brilliant AI model can't succeed if it can't integrate with existing systems, meet security requirements, or handle real-world data flows. These aren't glamorous considerations, but they're often where great ideas meet harsh reality.

5. Attempting the impossible

Some challenges simply exceed what current AI can reliably handle, predicting complex human behavior, navigating rapidly changing regulations, or making decisions that require deep contextual understanding that hasn't been encoded in training data.

‍

Why things might get harder

Several trends suggest the failure rate could climb even higher:

The skills gap is growing: IBM research indicates that specialized AI skills shortages could exceed 50% of demand. More projects are launching, but there aren't enough experienced people to guide them thoughtfully.

Generative AI complexity: While tools like ChatGPT seem simple to use, implementing them robustly in enterprise settings brings new challenges around cost predictability, performance consistency, and integration complexity.

FOMO-driven timelines: Fear of missing out is pushing organizations to launch "internal ChatGPT" projects in unrealistic timeframes. Menlo Ventures found that nearly a quarter of generative AI projects begin without proper discovery phases.

Democratization without guidance: AI tools are becoming accessible to everyone, but without proper training and governance frameworks, this can lead to fragile systems that don't meet enterprise standards.

Evolving regulations: New compliance requirements, like the European AI Act, are adding complexity that many organizations haven't anticipated.

‍

The path forward: what success looks like

Here's the encouraging news: success isn't reserved for the most technically sophisticated or best-funded organizations. The companies that succeed share a different approach, one that treats AI implementation as fundamentally a human challenge.

Start with real problems

The most successful AI projects we've seen begin not with "How can we use AI?" but with "What problems genuinely need solving?" This shift in perspective changes everything that follows.

Involve people from day one

Change is easier when people feel included in shaping it rather than having it imposed on them. The best AI implementations we've witnessed involved affected teams in defining problems, evaluating solutions, and designing new workflows.

Think partnership, not replacement

Organizations that frame AI as a collaborative partner rather than a replacement tool tend to get better adoption, more thoughtful implementation, and ultimately better results.

Spread risk and learn fast

Instead of betting everything on one large project, successful organizations run multiple smaller experiments, learn from each one, and gradually build confidence and capability.

Build learning into the process

Every interaction with AI generates learning, about the technology, about your organization, and about the problems you're trying to solve. The organizations that systematically capture and apply these insights tend to improve continuously.

‍

Finding hope in the challenge

Yes, 80% of AI projects failed in 2024. But this statistic also represents an opportunity. In a world where most organizations are struggling with AI implementation, there's a significant advantage waiting for those who approach it more thoughtfully.

At Spentia, we've come to believe that the key to joining the successful 20% isn't about having the most advanced technology or the biggest budget. It's about remembering that AI implementation is, at its core, about people, understanding their needs, involving them in solutions, and creating environments where human judgment and artificial intelligence can work together effectively.

The future belongs not to organizations that deploy AI fastest, but to those that deploy it most wisely, with respect for both human potential and technological capability.

Research sources: https://www.rand.org/pubs/research_reports/RRA2680-1.html, IBM AI Skills Gap Report, Menlo Ventures State of Generative AI, European AI Act