Build this after the validation work in Chapter 3 and the serving work in Chapter 6. Deep learning is optional here.
Start with the causal primer. The 25-hour schedule assumes small, prepared data, your earlier environment and a reused service boundary. Finding data, a new delivery integration or approval for a production campaign needs extra time. First finish one supported baseline for a single binary action. Pretrained causal models, generative ranking and bandits are optional depth.
| Session | Hours | Do this | Done when |
|---|---|---|---|
| 1 | 3 | Define action, no-offer control, profit, decision time, and measurement window | A one-page decision contract has an owner and examples |
| 2 | 5 | Audit data, implement features available at decision time, and build rules | You can replay one decision and trace every input |
| 3 | 5 | Compare logistic regression and one boosted tree; build a treatment-effect baseline if data permits | Validation table separates predictive quality from policy value |
| 4 | 5 | Evaluate profit, uncertainty, overlap, fatigue, and important customer segments | The proposed policy beats the appropriate baseline or you explain why it does not |
| 5 | 4 | Serve the policy, implement logs, budget reservation, and fallback | Invalid, repeated, stale, and timeout requests behave correctly |
| 6 | 3 | Write the experiment and rollout memo; rehearse the project story | Someone else can reproduce the result and decide what evidence is missing |