Predictive scoring estimates which of your customers are likely to churn, so you can act before they shop less often. Each contact gets a churn likelihood score, along with an Act before date, an Activity level, Estimated time left, and Customer Lifetime Value (CLV). The scores are calculated by a machine learning model trained on your Voyado data. They're updated once a month and shown on the member's contact card.
What predictive scoring calculates
Predictive scoring is essentially a churn score. It estimates which customers are likely to churn, and it also includes several calculated values.
The likelihood of churn is a number between 0 and 1. The closer a customer's score is to 0, the less likely the customer is to churn. From this value, we calculate the following:
- Act before date
- Activity level
- Estimated time left
- CLV
Churn in retail
Churn is tricky to define in the retail industry. There's often no definite action that ends the relationship. Instead, a customer's shopping frequency declines until it's very low.
The model predicts this decline in frequency and presents a likely timeline in which the customer still shops at their normal frequency. This gives you the Act before date. After this date, if nothing has happened, the customer is likely to shop less often.
How the scores are calculated and updated
The scoring script runs in Databricks and follows these stages:
- The script creates a data set from the data lake.
- Python code containing a machine learning (ML) algorithm trains the scoring model on the data, then scores all contacts in the data set.
- The results are exported as a CSV file to a designated folder in the data lake.
- A job reads the file and imports the values into Voyado, on the member's contact card.
The algorithm is a gradient boosting algorithm that uses decision trees as learners. The model can be trained on any data that comes from Voyado.
The model retrains every month to keep it accurate, which means the scoring values are updated once a month.
Model settings
You need to decide on the following settings before we can set up the module.
| Setting | Description | Default value |
|---|---|---|
| Margin of sales rate (%) | Your margin on items sold, used to calculate CLV. A contact's CLV is based on the margin of income (the profit). If you don't want to base CLV on your margin on goods sold, enter 100%. A contact's CLV is then based on turnover, which is the total amount that contact has purchased. | 25 |
| Maximum years to cap Customer Lifetime Value calculation at | The maximum number of years to use in CLV calculations. Common values are 5 or 10 years. A contact's lifetime value is based on a lifetime within that timeframe. If a contact's remaining lifetime is predicted to be shorter than the maximum, that shorter lifetime is used as the basis for the CLV calculation. | 10 |
| Cost of capital rate (%) | The interest rate on capital used in the CLV calculation, which reflects the depreciation of a contact's value. | 2 |