Key takeaways
- Why technical capability usually isn't the real bottleneck: it's business alignment, framework, and having the right people at the table throughout the machine learning lifecycle
- The AT&T fraud detection journey, from a small team writing manual rules, to automated machine learning with H2O, to more than 20 models in production and an almost 80 percent drop in fraud
- The three recurring blockers, time, trust, and talent, and how automation and interpretability address each one
- What the AI Pulse engagement includes: H2O Driverless AI licenses, an embedded data scientist, tool training, and a design thinking Analytics Studio workshop with business stakeholders
- How model deployment flexibility works: training and scoring can run on your own cloud, on premises, or the partner's cloud, with ModelOps handling monitoring and governance
- The practical difference between H2O-3, open source and code first, and Driverless AI, built for governed enterprise deployment
Overview
H2O.ai's Mark Backer and Ed Santana, from an H2O Latin America partner, explain why AI projects stall between proof of concept and production, opening with the story of Netflix's recommendation engine contest, whose winning algorithm was too complex to ever deploy. They walk through the machine learning lifecycle end to end and an AT&T fraud detection case study where moving from manual rules to automated machine learning grew the model count from a handful to more than 20 and cut fraud by almost 80 percent. The session closes with the partner's AI Pulse program, a three month package pairing H2O tooling with dedicated data science talent and a design thinking framework to move faster.