Executive summary
A framework for deciding whether deep learning fits a use case and delivering it against a fair baseline.
What is inside
What the framework covers
The framework defines the task, screens fit on six criteria, builds a simple baseline, pilots a deep model on the same held-out data, weighs errors avoided against cost, checks explainability and regulatory risk, and sets monitoring.
- Task definition and success metric
- Six-criterion fit screen
- Baseline built before the deep model
- Comparison on a held-out evaluation set
- Errors avoided, hours released and payback
- Explainability, fairness and regulatory checks
- Monitoring, drift and the review record