Use propensity scores and uplift modeling to target customers who are most likely to be incrementally influenced by your marketing. Shift from response rate to true causal lift to grow CLV efficiently.
A propensity score is the probability of receiving a treatment ______ observed covariates.
given
versus
ignoring
despite
In marketing, targeting by uplift focuses on customers with the largest ______ effect.
placebo
incremental
observational
seasonal
Matching or weighting on propensity scores aims to ______ covariates between treated and control groups.
discard
amplify
balance
duplicate
Inverse probability weighting uses weights of 1/PS for treated and 1/(1−PS) for controls to create a ______ sample.
pseudo‑randomized
clustered
overfit
undersampled
A key assumption for valid causal estimates is ______: every unit has a non‑zero chance of treatment and control.
post‑treatment adjustment
unit homogeneity
perfect collinearity
overlap (positivity)
Propensity methods cannot remove bias from ______ confounders.
synthetic
instrumented
measured
unobserved
To avoid bias, do not include variables that are affected by the treatment; these are called ______ variables.
post‑treatment
instrumental
collider‑free
pre‑treatment
A common way to evaluate uplift targeting quality is with the ______ curve or its Qini coefficient.
ROC
uplift
PR
calibration
Optimizing for CLV with propensity scores typically uses an outcome like ______ rather than clicks.
pixel fires
views per session
impressions
revenue or margin over a horizon
When PS is extremely close to 0 or 1, a robust practice is to apply ______ or trimming to stabilize estimates.
duplication
reverse scoring
weight caps
noisy labels
Starter
You’ve got the core ideas—reinforce balancing, overlap, and why unobserved confounding is a risk.
Solid
Well done—apply matching/weighting and validate with uplift curves or Qini.
Expert!
Superb—optimize on CLV outcomes, cap extreme weights, and avoid post‑treatment bias in production.
