🔗  Cumulative Incremental Gain (CIG)

Cumulative Incremental Gain (CIG)#

The running sum of incremental gain as more of the ranked population is treated.

Important

✨ AI-generated content. This page was written with the assistance of an AI language model and is provided as a learning aid. Despite careful review, it may still contain mistakes, omissions, or out-of-date information. Whether you are new to the topic, a team lead, or a senior practitioner, treat it as a starting point rather than an authoritative reference: read it critically and independently verify anything you act on (code, commands, figures, and factual claims) against official documentation and primary sources before relying on it.

What it is#

Cumulative incremental gain (CIG) is an uplift-modelling measure of the total additional outcomes — purchases, sign-ups, conversions — won by targeting customers ranked by uplift score, relative to random or no targeting. It is the quantity plotted on the Y-axis of a Qini curve.

How it’s built#

Four steps: rank customers by predicted uplift score, highest to lowest; split them into deciles or percentiles; for each segment compute the incremental gain as treatment responses minus control responses; then take the running sum of those gains down to the segment of interest.

The formula#

\[\text{CIG}(p) = \sum_{i=1}^{pN} \left( y_i^{\text{treat}} - y_i^{\text{control}} \right),\]

the treated-minus-untreated outcome accumulated over the top proportion \(p\) of a population of size \(N\).

A worked example#

Across 10,000 customers, targeting the top 20% yields 400 treatment versus 300 control conversions — a gain of 100; extending to the top 40% adds 750 versus 550 — a further 200. The cumulative incremental gain at 40% is therefore 300 conversions.

Reading the Qini curve#

Plotting CIG against the percentage of the population targeted gives the Qini curve. A steep early slope means the model finds persuadables first and targeting pays off quickly; a flat curve means the model is no better than random. The shape reveals the optimal targeting fraction — where gain stops rising — and lets you compare models by how much incremental value each delivers.


Theme: Causal Inference & Uplift  ·  All terminology



See also

Source article Adapted (context, re-expressed) in our own words from: Cumulative Incremental Gain (CIG) (insightful-data-lab.com).

Tags: purpose: reference topic: terminology level: advanced