Funnel, Experiment, and Lifecycle Analytics¶
This tutorial covers the stateful analytics families using the demo workspace.
Complete Getting started first so examples/demo has
ingested aggregates.
Funnels¶
The demo workspace configures ih_outcome_funnel (funnel) with the stages
Impression → Clicked → Conversion. Query the stage dropoffs:
uv run valuestream query examples/demo VS_Impression_to_Click_Dropoff --by Channel --grain Day
uv run valuestream query examples/demo VS_Click_to_Conversion_Dropoff --by Channel --grain Day
Experiment Monitoring¶
The engagement processor persists the ModelControlGroup dimension, so test
and control variants can be compared statistically:
uv run valuestream query examples/demo VS_ModelControl_Engagement_Compare --by Channel --grain Day
uv run valuestream query examples/demo VS_ModelControl_Engagement_Z_Test --by Channel --grain Day
variant_compare returns per-variant rates, lift, and standard error;
contingency_test returns the experiment p-value. The formulas are in
Algorithms.
Lifecycle, Sets, and Snapshots¶
Three further stateful processor kinds are available but not configured in the demo catalog:
| Kind | Use |
|---|---|
entity_lifecycle |
CLV, RFM segments, recency/frequency/monetary summaries |
entity_set |
Approximate set overlap and cohort comparisons (theta sketches) |
snapshot |
Periodic aggregate state, e.g. active holdings per period |
To try them, add a processor of the relevant kind to your workspace's
catalog/processors.yaml, bind metrics to its state, validate, and re-run the
source. See Processors for each kind's
configuration and state layout, and the
configuration guide for
the edit-validate-rerun workflow.