Call: Thursday, July 2, 2026 · 4:30 PM EST · Call #2 of the bi-weekly series
Project: Customer Lifetime Value Pricing (CAS Ratemaking Working Group)
Phase: 2 — Data Architecture & Synthetic Data (CAS window Jun 25–30) · Phase 3 opened Jul 1
Researcher: Pramod Misra (5G Vector partner: Neha Tiwari)
CAS staff / coordinators: Elizabeth Smith (), Heather Davis ()
Project Oversight Group (reviewers): Mark Mondello (QBE), AJ Robinson (Allstate), AJ Paik,
Geoff Werner (Werner Advisory), Ronald Kozlowski (RTK Services)
Master log of all calls / updates / POG inputs: docs/calls/README.md
Phase 2 — Data Architecture & Synthetic Data (window Jun 25–30) — ✅ complete
- docs/02_data_architecture.md + cas_clv.data: synthetic generator mirroring the Applied Epic
agency-management data model — multi-year, multi-product (auto / home / commercial BOP)
customer / policy / claim histories, fully deterministic under set_seed(42).
- No proprietary agency data enters the repo (synthetic-first, per the kickoff decision); public
sources remain available for calibration.
- QA gate green throughout: pip install -e . · pytest · ruff · mypy.
Phase 3 — CLV Model Development (window Jul 1–14) — 🟡 in progress
- First model in place: BGNBD_CLV — native, scipy-only BG/NBD (Fader–Hardie–Lee 2005) with
each policy renewal treated as a purchase occasion; discounted expected renewals × net annual
margin. This is the probabilistic base of the POG-selected hybrid headline model, with
fit() / predict_clv() / sensitivity_analysis() on the common interface.
- Fully auditable implementation (no dependency on the unmaintained lifetimes package);
engineering rationale in docs/03_dependency_decisions.md.
Adopted into the framework this week (docs/01_clv_framework.md §2.1): every model reports CLV
in two variants, side by side:
| Variant | Cash flows | Primary use |
|---|---|---|
| Unconstrained | Renewal book as-is: P − L − E (X = 0) |
Filing-defensible base; the "actuarially justified" side; rate-side analysis in strict jurisdictions (NY §2304) |
| Constrained | P − L − E + X, where X = cross-sell + upsell |
Business-planning view under explicit cross-sell / upsell program assumptions; marketing, distribution, agency management |
X term concrete as cross-sell (products
added to the household — Verhoef–Donkers 2001 "potential value") plus upsell (coverage-tier /
limit upgrades — the Markov transition dynamics).predict_clv() returns both columns; sensitivity_analysis()
sweeps cross_sell_value and upsell_value as explicit grid dimensions.POG input requested: confirm the labels ("constrained / unconstrained" vs e.g. "filing base / business-potential"). The decomposition stands regardless; only column names change.
ParetoNBD_CLV (native, on the shared RFM pipeline) — the Schmittlein (1987) benchmark beside BG/NBD.MarkovCLV + HMM latent states (coverage-tier / product-addition transitions; Dong–Frees 2022 template).notebooks/01_data_exploration.ipynb and the start of the model comparison table.docs/01_clv_framework.md (incl. new §2.1) · docs/02_data_architecture.md · docs/03_dependency_decisions.mdsrc/cas_clv/ (data, features, _btyd, models) · tests/ (15 tests, QA green)docs/literature/00_literature_review.md · research/papers/ (16-item corpus + manifest)