Predicting response and willingness.
WHAT A PROPENSITY MODEL PREDICTS
Likelihood of a particular action: purchase, upgrade, response.
WHAT IT IS USED FOR
Targeting, prioritisation, and resource allocation.
WHAT THE COMMON MISTAKE IS
Targeting those most likely to act.
WHY THAT IS WRONG
Some would have acted anyway, and contacting them changes nothing.
WHAT UPLIFT MODELLING PREDICTS INSTEAD
The change in likelihood caused by the intervention.
WHAT IT REQUIRES
Experimental data: a treated group and a control group.
WHY THAT DATA IS NECESSARY
Causation cannot be estimated from observation alone.
WHAT THAT MEANS PRACTICALLY
Hold back a control group, always.
WHAT PRICE ELASTICITY MODELS ESTIMATE
How demand responds to price.
WHAT MAKES IT DIFFICULT
Prices were set for reasons, which confounds the relationship.
WHAT HELPS
Deliberate price variation, where commercially acceptable.
WHAT TO BE CAUTIOUS ABOUT
Personalised pricing, which raises fairness and regulatory questions.
WHAT TO MEASURE
The incremental effect, against the control.
WHY
Total outcome among the targeted group says nothing about the intervention.
