DataTopics: All Things Data, AI & Tech #93 The Most Misunderstood AI Statistic of the Year: Lessons from Tech Expo on Hype, Failure, and Innovation
Dec 19, 2025
Tim Van Erum, Customer Solutions Lead at Data Roots, shares insights from Tech Expo Amsterdam. He debunks the misleading '95% of AI projects fail' statistic, illustrating the importance of experimentation in innovation. The conversation shifts to Reddit's machine learning strategy for community moderation, as well as PepsiCo's AI consoles for business intelligence. Tim emphasizes the need for a culture of curiosity and pragmatic AI adoption, making strong cases for companies like Albert Heijn, who are effectively using AI to address real-world challenges.
AI Snips
Chapters
Transcript
Episode notes
Hype Versus Practical AI
- Conferences split into hype vs. pragmatic AI tracks, and the pragmatic track offered far more actionable value.
- Tim Van Erum preferred grounded talks showing real deployments over abstract multi-agent hype.
Failure Is Part Of AI R&D
- The oft-cited MIT stat that '95% of AI projects fail' is misused and overlooks R&D realities.
- Tim argued experimentation and failed attempts are essential steps toward eventual productive deployments.
Measure Experiments By Investment, Not Count
- Do experiment broadly but treat 'projects to production' ratios as a faulty metric.
- Track scope, time and investment per initiative instead of counting initiatives equally.