Guide to Customer Churn Prediction

You can read more about feature engineering here.

Now that you have a good set of signals and the data to support it, you will need to turn that into specific features and training data.

Training data: Examples of churn and happy customers

Training data forms the foundation of your prediction model. Think of it as a comprehensive spreadsheet containing all your features and whether each customer ultimately churned or stayed. This historical data teaches your model to recognize patterns that predict future churn.

For effective training, you need:

• A significant sample size (minimum 10,000 examples recommended)
• Data from both churned and retained customers
• Clean data without outcome leakage
• Examples across different customer segments, plans, and time periods

A crucial consideration is preventing data leakage. For instance, if you're predicting churn 28 days in advance, your training data shouldn't include any information that wouldn't be available 28 days before a customer churns. This ensures your model learns from legitimate predictive signals rather than artifacts in your data.

Once you have a training data set, you’re ready to train a model.

Models: Machine learning to actually make predictions

While the technical details of machine learning models are beyond this article's scope, understanding your options is important. Here are common approaches with their tradeoffs.

The table below outlines some common ways you can achieve good results when creating a churn prediction model. However using a tool like AutoML – which finds the appropriate model for you – will make for a much simpler process at a slightly increased training cost. 

Each approach offers different benefits in terms of accuracy, training speed, and explainability. For many businesses, starting with simpler models like Random Forests or Gradient Boosting provides a good balance of performance and interpretability.

Explainability: Going beyond a likelihood and knowing why

Understanding why a model predicts churn is often as important as the prediction itself. Models provide a score between 0 and 1, indicating churn likelihood, but stakeholders inevitably ask, "Why is this customer likely to churn?"

Feature importance analysis can answer this question, showing which factors most influenced the prediction. This information is crucial for taking effective action - knowing a customer is likely to churn is only useful if you understand what you can do about it.

Using predictions in the real world

How to get the information to the right people

Now that you have a model that returns a prediction and a way to explain the why behind that prediction, you are probably going to want to do something with this model. 

Getting predictions is only half the battle - you need to make them actionable. Consider how different teams will use this information:

• Customer Success teams might want alerts in Slack for high-risk accounts (see also: Logo Retention)
• Sales teams might need churn risk scores in their CRM (Salesforce or HubSpot)
• Marketing teams might use predictions to trigger automated campaigns in Customer.io
• Product teams might adjust the user experience based on churn risk

In all of these cases you are going to need to:

1. Access to live data for ongoing predictions
2. Regular model retraining to maintain accuracy
3. Integration with your existing tools and workflows
4. Clear processes for acting on predictions

Google Cloud, AWS and open source frameworks have guides on how to take a model and serve it or run batch predictions against a set of data. 

In most cases you will want to not only continuously run these predictions as real world data changes, but also retrain your model often to account for drift in the underlying data, seasonality, new pricing plans and other changes. 

What plays you can run with churn predictions

Customer Success: Churn Plays

Using tools like Salesforce, HubSpot, or Intercom, CS teams can:

• Prioritize outreach to high-value accounts showing churn risk
• Create automated workflows for different risk levels
• Schedule proactive check-ins based on churn indicators
• Develop targeted success plans for at-risk customers
• Track intervention effectiveness through customer health scores

Integration example: Connect Upollo with Salesforce to automatically create high-priority tasks for CSMs when enterprise accounts show increasing churn risk.

Sales Team Plays

Using CRM platforms like Salesforce and HubSpot, sales teams can:

• Identify expansion opportunities in accounts with low churn risk
• Prioritize account handoffs based on churn prediction scores
• Target accounts showing signs of growth for proactive upsell
• Focus on consolidating multiple individual seats into team licenses
• Prepare renewal strategies based on account health indicators

Integration example: Connect Upollo with HubSpot to automatically create opportunities when accounts show potential for team expansion or enterprise upgrade.

Multi-Seat Consolidation

When multiple users from the same company are on individual plans, sales teams can:

• Identify total company spend
• Calculate potential savings with team/enterprise plans
• Reach out to the most engaged users as potential champions
• Present consolidated billing benefits

Growth-Based Outreach: Monitor account growth signals to:

• Track increasing user counts within teams
• Identify departments using the product separately
• Spot companies approaching usage limits
• Flag accounts ready for enterprise features

Renewal Strategy: Use churn predictions to inform renewal approaches:

• Start renewal conversations earlier with high-risk accounts
• Prepare tailored expansion plans for healthy accounts
• Identify additional stakeholders needed for renewal success
• Time upsell conversations based on account health

Integration example: Use Upollo with Salesforce or HubSpot to automatically update account health scores and trigger different sales motions based on churn risk and expansion potential.

Putting it all together

Churn prediction has become a crucial tool for modern subscription businesses, helping retain millions in revenue through proactive customer engagement. While building a custom prediction system requires significant expertise and resources, modern solutions like Upollo make this capability accessible to businesses of all sizes.

The key to success isn't just in the technical implementation - it's in creating a systematic approach to acting on these predictions. Whether you build in-house or use a specialized platform, focus on making predictions actionable and measuring their impact on your customer retention.

Luckily, products like Upollo turn this from an ordeal into something you can be up and running with in minutes. Upollo connects with all of your existing tools like CRMs, subscription and billing systems, product events tools and more to predict churn, conversion and expansion. 

Upollo learns from billions of data points to have constantly updated models that learn from users and customers all across the globe. Keen to try it out?