Revenue-modeling agent
Pooled CLV
is a lie you tell yourself.
ltv decomposes lifetime value into acquisition value and retention value, per cohort — because a single blended CLV number hides an 8x spread in retention across your own customer base. Proven against synthetic ground truth: 1.6% error vs. 48% for the naive pooled estimate.
[higgsfield: a single gold line forking into two diverging paths against a black field, one path faint and dashed, engraved line art]
ltv · cohort decomposition · 12 cohorts, 30-month horizon
Why teams install
What ltv earns back
Pooled CLV is off by 48%
On a synthetic cohort mix with retention spanning 0.70–0.97, the naive blended-average estimate misses ground truth by 48.28% MAPE.
‹0f041c30›Cohort decomposition lands within 1.6%
Splitting acquisition value from retention value, per cohort, tracks the generative ground truth to 1.61% MAPE — estimated from observed transactions only.
‹0f041c30›The naive number isn't just imprecise — it's the planted failure
Swapping the analyzer to aggregate mode is the literal regression test: MAPE jumps to ≈49% and the gate goes red.
‹0f041c30›How it works
Three steps. No faith required.
01
Observe
Simulate or ingest month-by-month transaction history per cohort: one acquisition event, N recurring-revenue events.
02
Decompose
A Kaplan-Meier-style survival curve plus active-month revenue, estimated independently per cohort from observed data only.
03
Gate
Cohort decomposition must clear ≤3% MAPE against ground truth AND the naive pooled estimate must be provably worse (≥15% MAPE).
Proof band
One packet. Every claim accounted for.
packet rig/ltv · 12 cohorts claim "cohort decomposition tracks ground truth within 3% MAPE; naive pooled CLV is ≥15% wrong" decompose 1.6091% error · ceiling 3% naive 48.2791% error · floor 15% (proof it's a real problem) accuracy PASS — margin 1.39pp under ceiling baseline PASS — naive fails by 33.28pp over its own floor themis SEALED — 0f041c30
Every ltv run emits a packet like this — hash-chained, verifier-signed, and queryable from your own dashboard. How ProofPackets work →
Objections
Answered before you ask
Isn't this just a cohort retention chart?
A retention chart shows you the curve. ltv turns it into a decomposed dollar figure — acquisition value plus a discounted retention stream — and proves that decomposition beats the naive pooled number on your own data.
How do you know 1.6% is actually accurate and not overfit?
The analyzer never sees the generative parameters — only observed transaction history, exactly like real billing data. The 1.6% is measured against synthetic ground truth specifically to make that check possible.
Why trust the naive-baseline comparison?
Because the naive path is the literal planted-failure regression test: swap analyzer mode to aggregate and MAPE jumps to ≈49%, sealed into the same audit trail as the passing run.
Buyers
Who runs ltv
VP Finance
subscription SaaS
Head of Growth
marketplace, 12 acquisition cohorts
Revenue Analyst
PE-backed rollup
Pricing
Start free. Pay when it pays.
Generalist ships today. Domain Expert adds vertical memory + priors. Custom is built to your stack.
Free
Priced after a short conversation
3 cohorts, 12-month horizon. Full decomposition output.
Team
Priced after a short conversation
Unlimited cohorts, 30-month horizon, CSV / warehouse import.
Pro
Priced after a short conversation
Dedicated verifier, custom survival models, SSO.
Revenue models touch financial data. ltv's evidence bundle (per-cohort MAPE, gate thresholds, decomposition method) is built for the same audit trail SOC 2 wants. Formal attestation lands Q4 — the evidence pack is available to finance teams now.
Runs well with
