Step 03 · Learn
Train, evaluate and explain — classical and generative together
The right technique for the decision: propensity models, forecasts, optimisers, and retrieval-grounded assistants. Every model is compared against the current human baseline.
Mini process diagram
Baseline
Train
Evaluate
Explain
Approve
Model development with a human baseline as the bar.
01
Baseline first
A model only ships if it beats what the team does today.
02
Explainability
Feature attribution and scenario views built for the operator, not the data scientist.
03
Evaluation harness
Repeatable offline and online evaluation, including GenAI answer grading.
6 weeks
Time to first model
Required
Models beat baseline
Offline + online
Evaluation
Other delivery steps