UPLIFT command is very similar to the
AB_TEST command, where we are calculating how effective an A/B test has been.
UPLIFT goes on step further than finding statistical significance, by calculating the effectiveness of the A/B test.
UPLIFT calculates the uplift or
treatment_effect of applying a treatment to an entity (i.e. a user, customer, etc).
Common use cases for uplift modelling include:
- Measuring the effectiveness of advertising campaigns on KPIs like sales
- Measure the effectiveness of a new feature using a KPI like product engagement
By measuring how effective a treatment is to specific individuals, you can then prioritise who to target with new features, or who to spend your ad campaign money on!
UPLIFT command takes two inputs: the
target column (the column we wish to compare, e.g. conversion, churn, click-through-rate), and the
treatment column (i.e. the group they belong to - A/B, gender, location... whatever you wish!).
By default, the treatment column is assumed to be
treatment, and needs to be specified otherwise.
UPLIFT will use all the input columns to estimate the
Under-the-hood, we use causal machine-learning models to do this.
The output will contain one new column,
treatment_effect, i.e. the conditional average treatment effect, or how effective a treatment is expected to be for a particular individual.
treatmentcan be used to specify the column defining the group (A/B, advertising campaign, gender, etc). Default is
Returns the conditional average treatment effect (
treatment_effect) for each row of data, indicating how effective the treatment would be on that individual.
A/B test to see if a new feature increases usage.
SELECT * FROM user UPLIFT(total_hours_active_per_week, treatment=feature_group)
Calculate the effectiveness of a marketing campaign on individuals.
SELECT * FROM user UPLIFT(total_sales, treatment=saw_campaign)
Statistical test to see gender affects the types of products users buy.
SELECT * FROM user UPLIFT(product_category, treatment=gender)