AI Product Analytics

An end-to-end product analytics system for an AI assistant: activation, adoption, usage intensity, cohort retention, commercial segmentation, and randomized onboarding experiment analysis.

The project deliberately separates descriptive signals from causal claims. Synthetic event data makes the work reproducible and safe to publish.

Questions answered

• What percentage of new users reach meaningful AI activation?
• How quickly do accounts adopt the feature?
• Which signup cohorts retain through weeks 1, 4, and 8?
• How does adoption vary by commercial segment?
• Did guided onboarding improve 7-day activation?
• What uncertainty surrounds the estimated experiment lift?

Metric framework

Exposure → first query → activation → weekly retention → account adoption
                              │
                              └── experiment outcome (7-day activation)

Activation requires at least three successful AI queries within seven days of signup. Weekly retention requires at least one successful query in the given week after signup. Account adoption requires at least 20% of eligible users to activate.

Run

make all

This generates synthetic source data, builds the warehouse and marts, runs eight quality tests, calculates experiment statistics, and creates dashboard/data.js. Open dashboard/index.html for the product scorecard.

What this demonstrates

• Event modeling and behavioral funnels
• User and account-level metric grains
• Cohort retention analysis
• Product adoption segmentation
• Randomized experiment analysis with confidence intervals
• Clear separation of correlation and causation
• Reproducible data quality and CI

Read docs/methodology.md, docs/metric_contracts.md, and docs/decision_memo.md.