Module 8: End-to-End Pricing Pipeline

You have spent seven modules building the individual components of a pricing model. Module 3 taught you how to fit a Poisson GBM with the correct exposure offset. Module 4 showed you how to extract SHAP relativities. Module 5 gave you conformal prediction intervals with a provable coverage guarantee. Module 6 handled thin cells with Bayesian credibility. Module 7 optimised rate changes subject to loss ratio and volume constraints.

This module connects everything into a single pipeline that runs from raw data to a pricing committee pack. The pipeline is not a technical showcase. It is an organisational discipline. It exists because the individual components, however well-built, fail when they are connected incorrectly.

By the end of this module you will have run a complete UK motor rate review in one Databricks notebook: data ingestion, feature engineering, walk-forward validation, hyperparameter tuning, frequency and severity modelling, SHAP relativities, conformal prediction intervals, rate optimisation, and an audit record that satisfies the FCA's Consumer Duty reproducibility requirements.

This is the capstone of the core course, but it is not the end. The pipeline we build here has deliberate gaps that Modules 9-12 fill. Module 9 adds demand-aware pricing so your rate changes account for customer retention. Module 10 automates interaction detection so you do not miss non-linear effects your GLM cannot express. Module 11 covers what happens after deployment - monitoring the model in production and detecting when it drifts. Module 12 adds spatial territory factors using postcode-level data. Each of these modules slots into the pipeline at a specific point. Download the notebook for this module

---