Uplift models predict incremental impact rather than total conversion likelihood. They help teams target users who will move the needle, not those who would convert anyway..
Uplift models aim to predict ______.
customer lifetime value
overall conversion rate
the incremental impact of treatment on each user
average order size
A random‑forest uplift model typically requires ______ in the training data.
propensity scores alone
only outcomes
cluster IDs
both treatment assignment and outcome labels
The ‘Qini curve’ evaluates uplift models analogous to ______ plots in classification.
scree
ROC
kaplan‑meier
Pareto
Marketing teams use uplift scores to ______.
target only users likely to be persuaded
estimate total revenue
optimize site speed
segment by geography
A negative uplift score suggests the user might ______ if shown the variant.
convert anyway
be untrackable
react adversely compared with control
return a 404
Two‑model approach for uplift fits separate response models for treatment and control, then ______.
divides raw counts
averages residuals
subtracts their predicted probabilities
bootstraps importance
Causal forests extend random forests by focusing splits that maximize ______ heterogeneity.
time‑on‑site
price elasticity
device type
treatment effect
Unlike standard A/B significance tests, uplift models can prioritize segments even if overall lift is ______.
near zero
monotonically increasing
already significant
perfectly linear
The ‘trans‑te’ method regularizes uplift estimates to reduce ______.
bias in big data
server latency
variance in small leaves
UI clutter
Offline uplift model validation should always use ______ set aside from training.
oversampled positives
a holdout test set
simulated Gaussian noise
the full original data
Starter
Solidify the basics of the topic before running live tests.
Solid
You understand the core ideas—focus on edge‑cases to boost accuracy.
Expert!
Outstanding—you can teach your team and steer experimentation strategy.
