What does the customer churn prediction template provide?

This article explains what the customer churn prediction template provides in the AI & Analytics Engine.

For more information about what a template is, read this article.

Background

The customer churn prediction template provides a complete machine learning (ML) solution from start to finish, for businesses seeking to reduce customer attrition by predicting their customers' likelihood to churn.

The solution is designed to predict the risk of churn for active customers in advance, according to the desired period configured to suit the business' need. For example, if a business requires 30 days time to mitigate churn, the prediction can be made for currently active customers churning after 30 days.

The template supports two different types of churn use cases, based on two common business models.

Note: For more information about the types of churn supported within the customer churn template, read this article.

 

Similar to all business templates, it is highly accessible to business users without needing any technical expertise in data engineering, ETL, ML pipelines, and etc. Once the models are built, users can examine whether their performance is satisfactory and also understand how they predict.

Predictions generated by the customer churn template:

Using the customer churn template, users simply need to provide a connection to their live database, and obtain periodic predictions sent to chosen destinations on a set desired schedule, in the form of a tabular dataset containing:

  • The current active customers (who have not churned)
  • Their predicted likelihoods of churn within or after on pre-configured periods
  • The features (variables) that were used to make the churn prediction.

Prediction outputPrediction output from an app built using the transactional churn use case within the customer churn prediction template: who are the currently active customers and what their churn probabilities are

Model Insights

The Engine also provides all insights for models trained using the churn template, as with any other classification & regression model on the Engine trained with the “build from scratch” option.

Users can see how well the model performs on test data consisting of the last 30 days of all data with known outcomes:

Performance insights: Leaderboard summaryPerformance insights: Leaderboard summary

Performance insights: MetricsPerformance insights: Metrics

 

Performance insights: Precision-recall curvePerformance insights: Precision-recall curve

 

Users can also obtain further insights like feature importance and prediction explanation:

Feature importance valuesFeature importance values: provides insights such as the most important factors that contribute to churn

Prediction explanationsPrediction explanation: provides how different feature values impact a particular prediction

They can also explore how feature values impact the models' predictions using the what-if tool:

What-if tool

What-if tool provides a way to see how the predictions would change if the values of some contributing factors change.

Template inputs

The template also provides a way to get a complete ML pipeline by entering only business-specific objectives and domain-knowledge inputs, which does not require any deep technical knowledge of data science/machine learning concepts.

image-20230621-034040

 

Inputs for transactional churn
Inputs for subscription churn

 

Inputs for subscription churn

The inputs to the template differ based on the two types of churn supported by it.

As an example, below are the required inputs for the transactional option:

  • Historical datasets of transactions and,

  • Additional dataset with extra information about the customers (e.g. age, gender, education level, etc.)

Once the datasets have been defined, the template requires additional configuration details:

  • Identification of specific columns: which column stores customers' unique identifiers (ID), which column stores the transaction timestamp, and which column stores the transaction amount.

Column mapping

  • Definition of churn: how churn is assessed in their business based on activity threshold (e.g. 30-day period with < 100 dollars in total monetary value)

Defining churn

  • Domain-knowledge input: where signals that foretell impending churn can potentially be found, in other words the parameters of the time windows relative to the prediction date

Define contributing factors

  • Choice of model training: Either let the Engine automatically identify the top models and train them straightaway, or pick them manually.

No other input or configuration is needed from the user. The template automatically does the heavy lifting by building the right data science & machine learning solution based on the user’s business objectives – without requiring the need for any technical concepts. 

Automation provided by the template

The churn template completely automates the process of aggregating and joining the various input datasets in a suitable manner, by: 

  1. Providing full automation of feature engineering and data wrangling process.
  2. It then immediately proceeds pick the right ML problem type (classification) and the target column
  3. It then creates and processes the necessary feature sets, and trains the models according to the selection in the template
  4. At the end, it will allow the user to run the entire pipeline in prediction mode using the chosen trained model, and automatically obtain periodic predictions. One-time predictions can also be triggered with a simple point-and-click UI interface

App setup

Once the data and the business-problem definitions are provided, the Engine automatically processes everything and provides a trained model to predict churn.

ML pipeline

The full behind-the-scenes ML pipeline built by the Engine to suit the user’s business objectives. The Engine takes care of the tech expertise required to build ML solutions