Product Management and AI
summary
Breaks AI problems into four concrete model types and shows when to use classifiers, detectors (SSD Yolo), regressors for counting rotisserie chickens and clustering—built a VIPRE Siamese-based visual product recognizer that handles ~30,000 products in the IRL and churns 300–400 new SKUs daily. Warns that model chains are brittle (three 90% steps → ~72% end accuracy), emphasizes CI/CD gates, real-time training limits, annotation error sources (fatigue, bias) and incidence-rate math that makes naive “90% accuracy” claims misleading.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:00Models
- 4:11Chains
- 5:33Learning
- 7:19Downtime
- 10:50Life Cycle
- 12:10Ground Truth
- 13:44Sources of Error
- 16:23Groundwork
- 16:46Accuracy
- 19:34Precision
- 20:28Confusion Matrix
- 22:27Good Bad Statements
- 23:28Requirements
related talks
26mAI and ML for Product Management
27mAI Versus Traditional Product Management
23mDelivering business success with AI
01h 09mWorkshop: Product development for identifying and leveraging ML opportunities
37mPanel Discussion: Developing with AI
27mBuilding AI applications: Successes and Lessons
13mAI as a Strategic Asset: Building Competitive Advantage
40mData Products and AI: A Conversation on Success Factors and Hurdles