Solution Finder
What is your customer trying to achieve? Pick the problem closest to what they actually said — this translates it into the right methodology, data requirements, and Lab.
You said: “Identify customers most likely to convert”
Conversion Propensity Modelling
AudienceWhy this fits
Conversion Propensity Modelling ranks individual customers by likelihood to convert so effort can be prioritised.
What it answers
Given behavioural signals about a customer, how likely are they to convert in a given window, so marketing can prioritise effort and spend accordingly?
How it works
Behavioural, engagement, and historical purchase variables are fed into a classification model that outputs a conversion probability per customer, which can then be segmented into propensity deciles for targeting.
Data required
- Previous purchase history
- Website/app visit and engagement data
- Recency and frequency of interactions
- Product/category view history
- Marketing engagement history
- Category interest signals
Discovery questions
What conversion event is being predicted, and over what time window?
Propensity models must be trained against a clearly defined outcome and horizon, or the score won't mean what stakeholders assume it means.
How much historical conversion data exists to train against?
Classification models need enough labelled positive and negative examples to learn a reliable pattern.
How will the propensity score actually be used in campaigns?
Determines whether the output needs to be a simple decile split, a continuous score, or feed directly into bidding.
Implementation approaches
Vendors
Risks & limitations
Risks
- Training on too few conversion events produces an unstable model
- Propensity scores can reflect existing targeting bias rather than true likelihood to buy
Limitations
- A propensity score is a probability, not a guarantee — high-propensity customers may still not convert
- Model performance degrades over time without retraining as behaviour shifts