A recent MIT study revealed a startling truth: 95% of enterprise AI projects fail to deliver measurable impact, resulting in what the study calls a ‘GenAI Divide’. If you’re social’s timelines are like mine, you’ve seen plenty of posts written about this research and the reasons for failure. I want to focus on something different—what it takes to succeed.

Let’s dive into what is causing failures and then look into what is driving success in AI, and how organizations can position themselves on the winning side of the divide.

Why AI Projects Fail (Short)

The failure rate of AI projects is not driven by inadequate technology—it’s caused by poor implementation. The MIT study identified key obstacles:

  1. Learning Gap: Most AI tools don’t adapt, learn, or retain knowledge, making them ill-equipped for complex workflows.
  2. Workflow Misalignment: Tools fail to integrate with existing systems and feel overengineered or brittle.
  3. Pilot-to-Production Chasm: While 80% of projects reach pilot stage, only 5% scale successfully due to barriers like poor adaptation and resistance to change.

These issues keep organizations stuck on the wrong side of the GenAI Divide—running experiments while failing to achieve meaningful transformation.

Why AI Succeeds

The organizations that succeed don’t just adopt AI tools—they adopt learning-capable systems and strategic approaches that tackle these common barriers head-on. Successful AI projects share a few essential characteristics:

1. Demand Systems That Learn

AI can’t just deliver answers—it needs to improve over time by learning from every interaction, retaining context, and dynamically adapting to workflows. The projects that cross the divide prioritize “agentic” systems with persistent memory and continuous improvement capabilities.

2. Start Small and Scale Gradually

Winners focus on targeted high-value workflows instead of broad transformations. By achieving visible wins—like automating document processing, summarizing calls, or simplifying contracts—these organizations use the momentum of success to expand AI adoption step-by-step.

3. Empower Frontline Champions

Rather than relying on centralized AI teams, successful companies enable line managers and employees on the ground to identify use cases. Supporting grassroots efforts, like shadow AI users experimenting with tools like ChatGPT, helps uncover real value in workflows.

4. Prioritize Workflow Integration

Tools that seamlessly connect with existing systems (e.g., Salesforce, SharePoint, or internal databases) and eliminate friction are far more successful than those requiring extensive customization.

5. Leverage External Partnerships

Organizations that succeed often work with trusted external vendors who specialize in learning-capable systems. These partnerships have a 66% success rate compared to just 33% for internal builds.

A Confirming Conclusion: We’re Already Doing This

What’s empowering about the MIT research is that it confirms and validates the strategies we’ve already made central to our AI projects. Over the past years, we’ve worked on implementations with the Ministerie van Financiën, Tweede Kamer, Ministerie van Justitie en Veiligheid, and Ministerie van Volksgezondheid, Welzijn en Sport.

Each of these projects focused on:

  • Custom solutions aligned to the needs of end-users.
  • Secure and dynamic systems that evolve with workflows.
  • Incremental transformations to ensure adoption success.

This research reinforces our strategic approach—success is within reach for organizations that focus on learning systems, gradual scaling, and collaboration.

Final Thoughts

The MIT study provides a lot of insight into why most AI projects fail, but more importantly, it shows us what we need to do to succeed. By demanding systems that learn, starting small, empowering employees, and fostering external partnerships, organizations can cross the GenAI Divide and unlock the potential of AI.

For us, applying these principles has already driven real results, and I’m excited to share more about our journey in future posts. Stay tuned—and let’s make AI success the norm, not the exception.