Interactive scenario planner

Set your own assumptions and read CLV back · Call #4, August 6, 2026 · all six model families · runs entirely in your browser

This is the page presented on 6 August 2026, kept as the record of that call. It has since been superseded by the interactive CLV scenario planner, which exposes every lever predict_clv() has — including retention, the two growth hazards, LAE, the fixed-dollar expense fields and both tenure bases — adds a single-customer mode that reproduces a worked example term by term, and verifies its own arithmetic against Python-generated golden vectors on load.

What this is

“One of my thoughts was that when this work is completed, we can create a web user interface where it gives you different options. You select, in your particular scenario, what’s the loss ratio, what’s the average premium, what’s the expenses and all. Based on that, you can get possibilities of this being your most probable value of lifetime. And as you want to change it, you can also change it.” — Pramod Misra, Call #3 (2026-07-16)

This is that interface, at the preliminary-toolkit stage of the tranche-2 bundle. Move any lever; all three CLV variants and the year-by-year build-up update immediately. Nothing is sent anywhere — the model coefficients are embedded in this page and the arithmetic runs locally, so it works offline and from a PDF-adjacent context.

Toolkit users who want per-customer rather than book-average answers call predict_clv() directly; this page is the book-average view for people who do not want to run Python.

Your assumptions

Model family Loss ratio improves with tenure

Average annual premium

Per customer, across all products held

Loss ratio

Incurred loss ÷ earned premium

Acquisition expense ratio

Charged once, at inception (sunk for in-force)

Renewal expense ratio

Charged on every future term

Horizon

Years of future renewals valued

Discount rate

Applied as (1 + r)^−t

Cross-sell value — CLV of the secondary product

Credited once, at the modelled time of sale · with-growth variant only, excluded from filing

Upsell value — permanent annual margin step-up

Applies in every term from the upgrade year onward · with-growth variant only, excluded from filing

Customer lifetime value — book average, per customer

Base — filing basis

—

P − L − E only. Defensible in an indication.

With growth — planning headline

—

Adds cross-sell (once) + upsell (recurring).

Net of acquisition — new business

—

With-growth minus acquisition expense.

Year-by-year build-up

Discounted contribution to with-growth CLV Negative contribution

The same numbers as a table

Year	Expected renewals	Loss multiplier	Effective loss ratio	Discount factor	Contribution	Cumulative

How it computes, and how to check it

CLV = Σ_t=1..H (1+r)^−t × [ (1−e) · A_t · s_P − m_t · B_t · s_P s_L + X · N_t ]

where, for the selected model, A_t is premium-weighted expected renewals in year t, B_t loss-weighted, N_t unweighted; s_P and s_L scale those to your premium and loss ratio; m_t is the tenure loss multiplier (1.0 when the toggle is off).

Why weighted aggregates rather than a simple book mean. Renewal counts and premiums are positively correlated, so mean renewals × mean margin is not the mean of the products — that shortcut was out by 4–14% here. Carrying premium- and loss-weighted aggregates makes this page reconcile to predict_clv() to the cent.

Press Reset to book reference and you should read exactly these, which are the same figures the toolkit produces on the seeded book:

Model	Base	With growth
Hybrid (headline)	$2,886.91	$3,494.10
BG/NBD	$2,805.31	$3,498.71
Pareto/NBD	$2,759.37	$3,394.29
Cox PH	$2,879.44	$3,498.62
Weibull AFT	$2,688.91	$3,257.01
Markov	$2,574.99	$3,128.86

BG/NBD's $2,805 / $3,499 are the same two numbers as case study 3 on the July 16 page — the $693 filing-vs-planning gap. Figures are consistent across calls by construction, not coincidence.

Honest limits

Book average, not per customer. Individual CLV dispersion is enormous (the 1-year cohort spans $146–$4,650 at P25–P75). This page cannot tell you about one policyholder — see the cohort exhibits.
Premium and loss ratio scale proportionally across the whole book. A change in mix rather than in level needs the Python API.
Retention is fixed at the fitted book rate. The retention-shock lever is deliberately absent here: shocking retention properly requires refitting, and the annuity-ratio approximation used in the sensitivity grid is not something to hand to a slider. Retention scenarios live in cas_clv.scenarios.
The tenure toggle uses the planning basis (level effect). The revenue-neutral filing basis is book-neutral by construction, so it would not move any number on this page — its effect is a redistribution between cohorts, shown in §5 of the deep dive.
Synthetic book. Coefficients come from 5,000 seeded synthetic customers, not a real carrier. The NAIC cross-carrier reproducibility test is still outstanding.

Coefficients embedded from python docs/calls/phase4_cohort_scenario_data.py §F, set_seed(42). Reference: premium $4,013.37 · loss ratio 0.5603 · acquisition 0.40 · renewal 0.20 · cross-sell $150 (credited once) · upsell $100 (recurring) · horizon 5 · discount 8%.

Companion materials: call brief · decision poll · cohorts & scenarios deep dive

← Project hub