Most machine learning models never reach deployment, and Gooder AI co-founder Eric Siegel argues that better algorithms won't fix the problem.
Eric Siegel is the co-founder of Gooder AI and author of The AI Playbook and Predictive Analytics. With more than 35 years in machine learning, a PhD from Columbia University, and experience teaching AI at Columbia, Siegel has spent his career connecting predictive technology to business outcomes. His first book has been adopted by hundreds of universities, and he has delivered 180 paid keynote speeches. He also founded Machine Learning Week, a conference series running since 2009.
In this episode of the Ignite Podcast, host Brian Bell welcomes Eric back to challenge the industry's obsession with generative AI. Siegel argues that most organizations should invest at least as much in predictive AI as generative AI, particularly when improving existing business operations. While generative AI attracts disproportionate investment and attention, predictive models can deliver substantial returns by improving decisions about customer retention, fraud detection, credit scoring, and other operational processes.
The deeper problem isn't model sophistication. It's deployment. Siegel explains why predictive AI projects stall when data scientists report precision, recall, and accuracy instead of translating predictions into financial outcomes. Meanwhile, generative AI faces a different obstacle: reliability. Even a system that performs correctly 95% of the time may be commercially unusable when its remaining errors carry serious consequences.
His proposed solution combines both approaches. Predictive reliability layers can identify high-risk interactions and route them to humans, potentially making partially autonomous systems viable where fully autonomous systems fail.
In Today’s Episode We Discuss
- 00:00Eric Siegel's 35 Years in Machine Learning
- 03:09Decision-Making Models and Predictive AI
- 06:41Why Better Data Beats Complex Models
- 09:24The AI Playbook and the Future of Predictive AI
- 14:15Audiobook Production and the New Paperback Edition
- 17:30Why Enterprise AI Projects Fail
- 20:25Reliability Layers and Human-in-the-Loop AI
- 24:14The Six-Step BizML Framework
- 30:56Can Foundation Models Replace Predictive AI?
- 35:27Who Should Read The AI Playbook?
- 37:58Why AI Benchmarks Fail to Measure Business Value
- 41:40AI Doom, AGI, and the Intelligence Explosion Narrative
- 46:19Eric Siegel's Predictive Analytics Rap
- 47:10The AI Playbook's Central Message
Siegel breaks down his six-step BizML framework, explains why technical model scores fail to communicate business value, and describes how routing risky AI decisions to human reviewers can rescue otherwise undeployable systems. He also challenges AI extinction narratives, recalls building a home sound booth to narrate his audiobook, and reveals the provocative original title MIT Press rejected.
Ultimately, the value of intelligence lies not in how impressive its capabilities appear, but in whether those capabilities improve decisions in the real world.
Pull Quotes
"The thing that tends to pay off more isn't com in increasing the complexity of the analytical method, but rather improving the data."
"my P doom is zero."
Follow Eric Siegel on LinkedIn: LinkedIn
Follow Eric Siegel on Twitter: X
Visit Our Website: Website
🎧 Watch, listen, and follow on your favorite platform: Website
Pre-order my book LP: The Insider's Guide to Investing in Venture Capital Funds