Superseded, 27 September 2026. This page is the record of Call #8 (10 Sep, paper v5.0) as presented, and keeps the figures and vocabulary of that date. In paper v5.2 (27 September 2026): the 3+ product relativity is 0.9683 on 9.44 expected years against 6.53 for mono-line (was 0.9425 on 7.92 against 2.72); the ten-year renewer is $1,093.62, breaking even in year 2 (was $782.14, year 4); the fully-allocated ten-year renewer is $590.36 (was $278.88). The current state is on the project hub.
Call: Thursday, September 10, 2026 · 4:30 PM ET · Call #8 of the bi-weekly series
Project: Customer Lifetime Value Pricing (CAS Ratemaking Working Group)
Phase: 6 — Paper, Toolkit, Executive Summary (Aug 20–Sep 15) · period of performance ends Sep 30
State: paper v5.0 (59 pp, conceptual) + technical supplement v5.0 (51 pp) — same pinned numbers, not one changed; no toolkit arithmetic changed · executive summary v5.0 · toolkit cas_clv v1.1, MPL 2.0 release subset · review register: Robinson 18 + Mondello 6 + Kozlowski 93 written, plus the verbal rounds
Previous call: #7, Thu Aug 27 — call notes · pointers from the recording
Researcher: Pramod Misra (Georgia Institute of Technology; 5G Vector partner: Neha Tiwari)
CAS staff / coordinators: Elizabeth Smith, Heather Davis
POG: Mondello · Robinson · Paik · Werner · Kozlowski
Master log: docs/calls/README.md
One thing to say before the agenda. On 27 August I promised the second draft for 1–2 September. It went out on 9 September. The reason is a choice, not a slip: Ron's marked-up review was due the same weekend and arrived on the 7th with 93 comments, and I chose to wait for it rather than send a draft that would have been superseded within days. Reviewers asked on the 27th for roughly a week with a draft before each call; this time you have had one day. If you would rather answer Q1–Q3 by email after reading, say so and the call becomes a walk-through.
Nothing here is new. Robinson's 18 comments (13 Aug) and Mondello's 6 (17 Aug) were closed at v3.0–v3.2 and accepted on Call #6; the seven round-2 items from 21 August were closed at v3.3 and presented on 27 August, including the retention-adjusted-loss-ratio defects that moved Exhibit 11 (3+ product relativity 0.9646 → 0.9425, expected life 7.92 years against 2.72 mono-line). All of those dispositions stand unchanged in v5.0. The review register is still the one place every point lives.
v5.0 is a rewrite for two readers, in two documents, not a re-computation. The trigger was Call #7: Geoff Werner — "who are we really writing for?" and "the code's all lost on me, but I could be the person who reads this and hands it to a team" — and Mark Mondello — "first I want to understand it very simplistically." Ron's review then showed, sentence by sentence, where v2.0 failed that reader.
Why two documents. Call #7 asked for simpler, not shorter, and v4.0 answered inside one 87-page binding. Geoff said "capture my attention so I hand it to them and say Do this"; Mark said "first I want to understand it very simplistically, then the complicated coding part." v5.0 executes the same decision in two bindings, built from the same pinned numbers — not one changed: the paper (59 pp) is conceptual and in plain language, with no formulas or software vocabulary in the body and 17 figures (eight new schematic diagrams — the lifetime value as a stack of yearly margins, two views of the same customer, three ways to report value, when the expenses fall, from customer history to a filing exhibit, the four-fifths rule, from one term's loss ratio to a lifetime loss ratio, valuing a book of business); the technical supplement (51 pp) carries the formulas, derivations, full tables and every former Reproduce this line under the paper's own section numbers (Part I), with the notation, data dictionary, sensitivity grid, estimation details, reproduction guide, all 13 hand-worked exhibits and the full 78-term glossary as Appendices A–G (Part II). The drift gate reports 0 unmatched literals across the paper (822), the supplement (1,094), the executive summary and the quick-start. This is an amendment adopted on your review, not drift; Q5 (reserve) asks whether the split serves you.
| What changed | Where | Why (who asked) |
|---|---|---|
| The one-page case, Figure 1. Two in-force customers with the same current-term margin — \$979.28 and \$988.56 — and different renewal probabilities, 95.1% and 78.4%. Five-year values \$3,701.20 and \$3,196.62; gap \$504.58, ratio 1.16, opened entirely by retention. Beside it, a flat-retention table (90% vs 80%, every cell checkable: \$203.00 × 0.7290 ÷ 1.08⁴ = \$108.77) | §1.1 | Werner, Robinson, Mondello — the early "so what" (R3-1) |
| What this paper is, and is not — a boxed statement naming the two uses (rating decisions; valuing a book) and ruling out any willingness-to-pay content | §1.2 | Werner, Kozlowski (R3-4, R3-5) |
| Two documents. Paper (59 pp) and technical supplement (51 pp); Section 1.5 Where the detail lives; every §3 subsection opens In plain terms; every section closes with a one-line Further detail pointer into the supplement, where every former Reproduce this line now sits. §5 collapses the four model-family subsections into 5.1 Four ways to estimate how long a customer stays (hybrid 5.2, comparison 5.3); §11 is a one-line-per-limitation table plus three paragraphs; §12 is Where the numbers come from | §1.5, throughout; supplement Part I | Werner, Mondello (R3-2); Kozlowski RK-88 |
| "Price optimization" removed except in the §2 literature citations; §3.6 retitled cost-based factors versus demand-based factors | §3.6, sweep | Werner, Kozlowski (R3-5; RK-47) |
Two expense views. Marginal stays the default; a named fully-allocated view via Economics.fixed_expense_per_term (\$75 illustrative), worked on the §3.4 ten-year renewer: \$782.14 marginal vs \$278.88 fully allocated, difference \$503.26 |
§3.4; hand-worked in supplement Appendix F.4 | Kozlowski RK-39 (#112) vs Robinson (#111) → Q2 |
| §9.6 Valuing a book of business. In-force 3,458 customers: \$13,842,714.22 marginal vs \$12,973,898.83 fully allocated, gap \$868,815.39, 93.72% retained | §9.6 | Kozlowski RK-34 (#95), Werner (R3-4) → Q3 |
| L16 — ex-ante identifiability. A high-value customer must be identified from covariates and history, not in hindsight | §11 | Kozlowski (R3-7) |
Two glossaries from one source (scripts/glossary_data.py): paper Appendix C, 29 core terms; supplement Appendix G, all 78. docs/quickstart.md (3-pp PDF). Revision history is the change memo (CAS_CLV_Changes_Since_v2.0_to_v5.0.pdf, 32 pp); paper appendices are now A Two policyholders, term by term, B sample filing exhibit, C glossary |
paper App. A–C; supplement App. A–G; change memo | Werner (R3-3); Kozlowski RK-11, RK-33 |
| Sentence length, jargon, concept-before-section-number, definitions at first use, ultimate/reported loss terminology, Table 8 redesign, attribution of the design decisions to the author | whole paper | Kozlowski, applied as global rules (memo §"What changed") |
Not changed: the six model families and their fits, every exhibit and figure of v3.3, the relativities, the disparate-impact screen, the §9 comparison. The only new numbers since v3.3 are the opening case and the fully-allocated example, both produced by the toolkit and under the drift gate; v5.0 adds none. The two-customer figure and the flat-retention table are worked on the deep-dive page, §2–§3.
Register ids RK-2 … RK-91 map to his Word comment numbers in order (three pairs marking the
same phrase share a row); the point-by-point response is
2026-09-09_review_response_RTK.html. He reviewed Abstract–§5.3;
§6–§15 are unmarked, and the memo invites a second pass on exactly those — §9.6 and §6.2 first,
because they are his use case and the exhibit his July expense question built.
Three comments changed what the paper says, not how:
| Word # | He asked | What v5.0 does |
|---|---|---|
| #24 (RK-8) | "This part makes it sound like CLV isn't worthwhile … Then why write this paper?" | The abstract stated the negative result first and alone. It now says where CLV's economic content is — acquisition, retention allocation, book valuation — and that re-ranking in-force customers is the one use where it adds little. |
| #95 (RK-34) | "I was thinking that this is more of a valuation exercise … one company buying another's book." | Both uses named in §1.2; new §9.6 with the in-force and whole-book totals under both expense views. → Q3 |
| #169 (RK-82) | "Synthetic data is better than real data? How can it be…" | No. The §4.4 paragraph now leads with that prior, states the one thing synthetic data permits (true retention is known, so recovered retention can be scored), gives the result, and says everything else about it is weaker. |
One is a disagreement between reviewers and goes to the room: #112 (RK-39), fixed expense — Ron says someone has to pay for it and leaving it out overstates value; AJ (#111) said it stays out of a customer-level CLV. v5.0 carries both views; Q2 picks the default. Two items are held open honestly: #153/#154, whether the 22% underwriting margin is defensible (to be verified against the payload, not asserted), and #122, whether Mark and AJ follow the §3.5 bases paragraph (that is Q4).
| # | You raised (27 Aug) | What shipped |
|---|---|---|
| R3-1 | Werner, Robinson, Mondello: an early "so what" — two customers, one year vs five, static, flat 90% retention fine if caveated | §1.1, Figure 1 and the flat-retention table, first thing after the introduction |
| R3-2 | Werner, Mondello: two audiences; write in a business register | Two documents — paper (59 pp, plain language, no formulas in the body) and technical supplement (51 pp); Section 1.5 Where the detail lives; In plain terms openers and Further detail closers; every Reproduce this line now in the supplement |
| R3-3 | Werner: spell out abbreviations; explain the four-fifths rule from scratch | Done at first use throughout; §7.2 rewritten for a reader who has never met the rule, with a new schematic figure; glossaries — paper Appendix C (29 core terms), supplement Appendix G (all 78) |
| R3-4 | Kozlowski, Werner: can this value a book of business? Say so if it serves both | §1.2 names both uses; §9.6 gives the book-value reading → Q3 |
| R3-5 | Werner, Kozlowski: this is not price optimization; drop the term or say so early | Term removed except §2 citations; §3.6 retitled; §1.2 box |
| R3-6 | Kozlowski, Werner: rewrite the §4.4 "synthetic data permits a validation real data cannot" sentence | Rewritten (same disposition as RK-80 … RK-84) |
| R3-7 | Kozlowski: can the high-value risk be identified ex ante? | Limitation L16; the exam paper he referenced is not yet identified — a citation from Ron would close it |
| R3-8 | Mondello, Robinson, Werner: second draft + spreadsheet reconciliation, one email, links in the email not Teams | The 9 September package: v5.0 paper and technical supplement, both memos, the reconciled workbook, all links, one message |
| R3-9 | (Researcher's own open question, unanswered on the call) how to describe the planner; do the extra scenarios stay | Decided in the absence of a steer: described once in §9.6 and once in §12; scenarios stay |
Four decisions on the decision poll, plus one reserve question; the substance behind each is on the deep-dive page §4–§5. Q1–Q3 need the room; Q4 is a ratification carried from 27 August; Q5 is a reserve on the v5.0 split.
clv_base / clv_with_growth with the gloss the customer as
they are today vs plus expected future purchases (★) / rename to single-line / multi-line
(Robinson's Call #7 suggestion; wrong for the 250 three-plus-product and 1,545 two-product
households whose base value already spans several lines) / current-holdings / with-growth.Reserve, for 24 September: where to present. Geoff offered on the 27th to moderate or co-present at RPM 2027, Indianapolis; the poll's reserve question lists session formats.
CAS side
- Submission venue — Variance vs E-Forum. Still open; drives the reference template and the
abstract's target length (v5.0 abstract: 247 words, unchanged from v4.0).
- ScholarOne + github.com/casact provisioning — still open since ~Jun 19; gates the public
toolkit release.
- RPM 2027 — deadline and route (project deliverable vs individual proposal), open since ~Jun 30.
- Recurring call invite lapsed end of August — Elizabeth to reissue (raised 27 Aug).
POG - Q1–Q4 on the poll, Q5 if time allows. Ron: a second pass on §6–§15 if time allows, §9.6 and §6.2 first. - Call #4 poll Q3 (may a tenure relativity enter a filed exhibit) remains unanswered; §7 takes a position a regulator could contest (L13).
Researcher
- Fold the Q1–Q4 answers into the four "provisional" passages; act on Q5 if the room asks for
anything to move back into the paper; verify the 22% margin (#153/#154).
- Delivery target: the 24 September call, with final acceptance edits inside September — the
contractual period of performance ends 30 September.
- dist_release/ gate in a fresh venv; §IX handover package to CAS.
New this call: 2026-09-10_poll.html ·
2026-09-10_worked_examples.html ·
2026-09-09_review_response_RTK.html — the response to Ron's
93 comments · docs/calls/2026-09-09_email_package.md — the one-message package (draft) ·
docs/calls/phase7_v4_call8_data.py — generates both HTML pages from the payload ·
docs/quickstart.md · scripts/glossary_data.py + scripts/build_glossary.py (paper Appendix C,
supplement Appendix G) · paper/technical_supplement.md (v5.0, new) ·
register §5b — round 3.
Updated: paper/paper.md (v5.0) · paper/executive_summary.md (v5.0) ·
paper/build/CAS_CLV_Paper_v5.0.{docx,pdf} (59 pp) ·
paper/build/CAS_CLV_Technical_Supplement_v5.0.{docx,pdf} (51 pp) ·
paper/build/CAS_CLV_Executive_Summary_v5.0.{docx,pdf} (3 pp) ·
paper/build/CAS_CLV_Changes_Since_v2.0_to_v5.0.{docx,pdf} (32 pp) ·
paper/build/CAS_CLV_Quickstart_v5.0.{docx,pdf} (3 pp) ·
docs/01_clv_framework.md · paper/figures/ (17 figures — Figure 1 the two-customer case, eight new
schematic diagrams, all renumbered by placement) · AI-usage log (contract disclosure).
Every figure in this brief is produced by python docs/calls/phase7_v4_call8_data.py from
paper/paper_numbers.json (keys so_what, book_valuation), under set_seed(42); the v3.3
figures quoted in §2 are those of the 27 August materials.