Product Conjoint Analysis

A reusable end-to-end workflow for conjoint analysis, distilled from a working case study (safety glasses on Amazon, N=160). It covers data acquisition, attribute engineering, experimental design, logistic regression estimation, and the four canonical insight transforms (importance, WTP, choice probability, cost-benefit ROI).

The skill is opinionated about decisions that beginners commonly get wrong — choice of reference levels, when to split models for collinearity, why price needs special handling for importance, when WTP is meaningful, etc. Read this file fully on first use, then consult the references/ folder when you hit a specific decision point.

Out of scope

Do not use this skill for: simple A/B testing, pure descriptive statistics, recommendation systems based on collaborative filtering, or single-attribute pricing studies.

The 5-phase architecture

Every conjoint study — regardless of domain — follows this structure. Always show the user this map at the start so expectations are clear.