clvkit

Skip to main content

Back to top

Ctrl+K

Search<br>Ctrl+K

System Settings

Light

Dark

GitHub

PyPI

Search<br>Ctrl+K

System Settings

Light

Dark

GitHub

PyPI

clvkit#

Four parameters. Six decimal places. No trace to read.

Buy-till-you-die models for a business where nobody cancels a subscription, so<br>you never observe a customer leaving — you only ever observe them not coming back<br>yet. Six lines take a transaction log to a per-customer CLV table, in 2.0 s<br>of wall clock on the CDNOW sample. Python 3.11+, four dependencies, no compiler.

uv add clvkit # or pip install clvkit

Install

One line, four dependencies, no compiler. Get it into your environment.

Install

User Guide

The key ideas, the four models, and the calls clvkit makes on your behalf.

User Guide

API Reference

Every class and method, documented from the source.

API Reference

Examples

Three worked runs on real data, each answering a question a business asks.

Examples

It lands where the paper said it would#

The BG/NBD model comes from a 2005 paper by Fader, Hardie and Lee in Marketing<br>Science. They fit it on the CDNOW 1/10 systematic sample, 2,357 customers with a<br>39-week calibration period, and printed the four fitted parameters. clvkit fits<br>the same model on the same data and recovers the same four numbers, to the last<br>printed digit.

CDNOW is the yardstick, not the boundary. clvkit fits any transaction log that<br>carries a customer id, a date and an amount, so point it at your own data and the<br>four models run the same way.

What it is

Parameter

Published, 2005

clvkit

Gamma shape for how the purchase rate varies across customers

0.243

0.242595

Gamma scale for that purchase rate

4.414

4.413602

Beta shape for the dropout probability after a purchase

0.793

0.792922

Beta shape for that same dropout probability

2.426

2.425907