Skip to content
BENIBIX Learn

Propensity Model vs Uplift Model

People think the customers most likely to buy are the ones to send an offer. They are not. A propensity model scores how likely someone is to act. An uplift model scores how much an offer changes that. Many top propensity scores belong to people who would buy anyway. Uplift finds the ones the offer really moves.

Marketing🛠️ MarTech, Data & AI

Propensity Model

A model that guesses who will buy, leave or click. Each buyer gets a score.

A propensity model is a model that gives each customer a chance of doing something, like buying, leaving or clicking. It learns from past customers who did or did not do it. Each person gets a score, often from 0 to 100%. Marketers then act on the top of the list. They might send an offer to people likely to leave. The score is a guess, so check it against what really happens.

Example
A phone company scores every customer on the chance they leave next month. It calls the top 5% with a better deal.
Why it matters
The money goes to the people most likely to act. It does not go to everyone.
Full entry
Marketing🛠️ MarTech, Data & AI

Uplift Model

A model that finds who acts because of an offer. Some people would act anyway.

An uplift model predicts how much an offer changes what a person does, not just what they will do. It sorts customers into four kinds. Some buy only if asked, some buy anyway, some never buy, and an offer puts some off. Only the first kind is worth paying to reach. The model learns from a test where some people got the offer and some did not.

Example
A shop's best buyers would mostly buy without a coupon. An uplift model picks the 20,000 people the coupon really sways instead.
Why it matters
A discount sent to someone who would buy anyway is money lost. This finds the people it really moves.
Full entry

In one line

Propensity Model
A model that guesses who will buy, leave or click. Each buyer gets a score.
Uplift Model
A model that finds who acts because of an offer. Some people would act anyway.