This page is the operational surface for the Customer Health Score initiative authored by Angela Fultz Nordstrom. It blends the existing ML stack (churn risk, usage tier, expansion probability) with per-account CS-Operations cadence (onboarding, health checks, QBRs, Insider strategic calls) to produce a composite 0–100 score per account, plus a backtest of the renewal forecast itself. Below is what's already shipped, what's in warmup, and what still requires new data integration.

The composite score is intentionally a transparent weighted sum, not a learned model — when CS asks "stop counting QBR cadence so heavily" we flip a number, not retrain. To retune:

1. Edit WEIGHTS at the top of etl/build_account_health_score.py. Must sum to 100.
2. Re-run python -m etl.build_account_health_score — finishes in <5 s on the active customer cohort (~1K accounts).
3. Refresh this page; the bands and at-risk drivers recompute against the new weights on the next API hit.

To add a 9th component (e.g. survey CSAT once Alchemer lands): implement a _component_csat() expression that returns a 0–100 polars column, register it in the WEIGHTS dict, and add the label to _DRIVER_LABELS so the top-3-drivers column picks it up. Cohort-only components return None for out-of-cohort accounts so the weighted average normalizes correctly — see _component_insider for the pattern.