The Propensity to Refer® data model is a customer segmentation tool which predicts each customer’s propensity to refer at the point of purchase.
How does it work?
Each customer will be segmented into one of two main groups; those with a High Propensity to Refer and those with a Low Propensity to Refer.
The model is updated every weekend to ensure it's using the newest and most relevant information possible, and adapting to the latest customer behaviour
These groups will then be presented different experiences optimised for conversion, The High Propensity to Refer group will be shown a referral offer while the Low Propensity to Refer group will be shown an alternative action based on their business goal for the experiment.
Example drivers used to predict the customers' behaviour
| Driver | Impact on prediction |
| Currency | Very high |
| Number of past orders | High |
| Marketing opt-in status | High |
| Operating system | Medium |
| Revenue from most recent order | Medium |
| Number of past shares | Low |
Alternative actions are based on 3 business goals:
- Showing a discount on next order to Increase repeat revenue
- Showing a Newsletter sign up to Increase marketing database
- Showing a different value incentive for each segment to Optimise referral and increase referral revenue
Some helpful terminology:
- High Propensity to Refer (High PTR): The customers who are most likely to share/refer. They're typically shown a referral offer.
- Low Propensity to Refer (High PTR): The customers who are least likely to refer. They're typically shown an alternative offer such as NPS, IFA or choice of offer with an IFA
- Control: This is a group of customers who should be in the Low Propensity to Refer group but are instead shown the same referral offer as the High Propensity to Refer group. This is for two reasons:
- We can see how well showing customers with a low Propensity to Refer an alternative action to referral is working. This includes showing additional value generated through the action as well as any revenue potentially lost by not showing referral.
- We can ensure the future predictive models are including how the Low Propensity to Refer group would behave with referral.
How do we calculate additional value of the Propensity to Refer®?
In order to calculate the additional value of Propensity to Refer® we compare the performance of the Low Propensity to Refer group (who are shown an alternative action to referral) against the Control group.
How are your customers being segmented?
Here is an example aggregated view of the proportions of propensity groups being calculated within a customer base. (Proportions may differ slightly at the experiment level).
