This document provides observations and recommendations for reconciling machine learning with business needs. Some key points made include:
- In many cases, machine learning is not needed to solve a problem and simpler solutions like collecting missing data can work better.
- The data companies already have is sometimes useless for machine learning problems. Domain expertise alone also often means less than expected.
- Not understanding technical constraints can cause machine learning projects to fail. Always create a proof-of-concept first before full development.
- It is important to establish causality through proper testing like A/B testing, as this validates models and addresses financial risks of implementations.
- Framing learning problems is challenging due to issues like lous