Customer churn rate, also known as customer attrition, refers to the percentage of customers who stop doing business with a company over a specific time period. Churn rate is an important metric that directly affects revenue and profitability. Businesses track churn rates to gauge their customer retention efforts and how well they are maintaining customer relationships. High churn rates can indicate that customers are dissatisfied with a product or service, while low churn rates can imply customer loyalty and satisfaction. Average customer churn rates vary from industry to industry. In 2022, the median customer churn rate was 13% for private software-as-a-service (SaaS) businesses.

To understand your business's churn, you need to know how many customers you're retaining or losing over a given time period, and why you're losing them. Customer churn models can help you build a holistic retention strategy that addresses the issues behind your churn. Below, we'll cover what you should know about different kinds of churn models, how to choose and build the right one for your business, and what to do with the information you gain from it.

What's in this article?

• What is a churn prediction model?
  
• Why predict customer churn?
  
• Key predictors of customer churn
  
• The four types of customer churn
  
• Types of customer churn models
  
• How to build a customer churn model for your business
  
• How to use predictive analytics to reduce customer churn
  
• Top use cases for churn prediction models
  
• How to measure the effectiveness of customer churn models
  
• How Stripe Billing can help
  
• FAQs about building a customer churn model
  

What is a churn prediction model?

A churn prediction model is a data-driven tool that estimates the likelihood that an individual customer will stop doing business with a company within a defined time frame. Using historical data on customers who have left and customers who have stayed, the model identifies patterns and risk factors such as declining usage, support complaints, or billing issues and applies them to current customers to flag who is most at risk of churning.

Unlike a general churn rate, which reports what already happened at an aggregate level, a churn prediction model looks ahead and scores individual customers, giving businesses a chance to intervene before those customers leave. These models are typically built using machine learning algorithms, such as logistic regression, decision trees, and gradient boosting, trained on customer attributes such as demographics, purchase history, engagement levels, and service interactions.

The output is usually a churn probability or risk score per customer, which businesses can use to prioritise retention efforts, target at-risk segments with personalised offers, and allocate resources towards the customers most likely to be saved.

Why predict customer churn?

Churn rate can have a major impact on a business's revenue, profitability, and reputation. Here's how churn rate impacts business operations.

• Customer acquisition costs: Generally, acquiring new customers is more expensive than retaining existing ones. A reliable churn prediction model helps offset this cost by flagging at-risk customers early, so businesses can direct retention spend where it's most likely to pay off, rather than repeatedly covering the cost of new acquisitions.

• Lost revenue: Losing customers creates a direct loss of income, immediately impacting the business's financial health. For subscription-based businesses in particular, this shows up quickly in monthly recurring revenue (MRR): every churned customer chips away at that baseline, and if the pattern continues across enough customers, it can leave the business without a stable revenue floor to grow from.

• Fewer long-term customers: Long-term customers tend to buy more over time and can therefore become less expensive to serve. They might also buy higher-margin products or services. A higher churn rate means fewer of these long-term customers and can lead to lower profitability.

• Brand reputation: High churn rates can be a sign of customer dissatisfaction, which can harm a company's reputation. Satisfied customers are more likely to refer others to the business, whereas dissatisfied customers might share their negative experiences, deterring potential customers.

• Long-term growth: Sustainable growth requires a stable customer base. Predicting churn before it happens gives businesses a chance to intervene while a customer relationship is still salvageable, rather than reacting only after revenue has already been lost. High churn rates can undermine long-term growth strategies, making it difficult for businesses to expand or invest in new opportunities. This is especially true for companies trying to scale. Without a way to anticipate and reduce churn, growth in new customers can be offset by losses in existing ones, making it harder to build sustainable momentum.

Key predictors of customer churn

One of the most straightforward predictors of customer churn is the level of customer satisfaction with a product or service. Businesses can measure this through surveys, net promoter scores (NPS) and feedback mechanisms. Consistently low satisfaction scores or negative feedback can be a strong indicator of potential churn.

Other factors that can influence a customer's decision to leave a business include:

• Customer support interactions: While some engaged customers may contact support to maximise value, support requests can also indicate problems or dissatisfaction. Analysing the nature and outcome of these interactions can predict churn, especially if issues are not resolved satisfactorily.

• Product or service usage: Monitoring how – and how often – customers use a product or service can offer businesses insights into customer engagement levels. Low or declining usage can signal that a customer is losing interest or failing to find value, which could lead to churn.

• Billing and payment issues: Frequent billing issues or problems with payments can lead to customer frustration and churn. Tracking these incidents and their resolutions can reveal early warning signs that customers are at risk of leaving.

• Changes in buying behaviour: A sudden change in a customer’s usual purchase patterns, such as a decrease in order size or frequency, can signal dissatisfaction or a shift in needs, potentially leading to churn.

• Customer feedback and complaints: Direct feedback or complaints are important indicators of potential churn. Customers who take the time to express dissatisfaction or concerns might consider leaving if their issues are not addressed.

• Engagement with marketing communications: A decline in customer engagement with emails, newsletters, promotions, or other marketing communications can indicate waning interest or relevance, which may precede churn.

• Market and competitive factors: Changes in the market and the actions of competitors can influence churn. For example, if a competitor comes out with a new product or offers a similar service at a lower value, this could lead to increased churn.

• Contract and subscription renewals: For businesses with a subscription or contract-based model, an upcoming renewal period is a key time to assess churn risk. When customers evaluate whether or not to renew, there is a chance that churn can occur.

• Demographic and psychographic factors: Sometimes, changes in a customer’s demographic profile or a shift in their values and preferences can predict churn. For example, a change in financial status, relocation, or lifestyle can influence their decision to continue using a particular product or service.

The four types of customer churn

Not all churn happens for the same reason. Here are the types to know:

Voluntary churn

The customer actively decides to cancel or stop using a product or service, often due to dissatisfaction, better alternatives, or unmet needs. This is the type of churn many churn models focus on predicting, since it's often preventable with the right intervention.

Involuntary churn

The customer doesn't intend to leave, but their access ends anyway, typically due to failed payments, expired credit cards, or billing errors. This type is often addressed through better payment retry systems and billing notifications rather than customer retention outreach.

Deliberate churn

A subset of voluntary churn where the customer makes a conscious choice to leave because the product no longer fits their needs, budget, or goals. This might include downgrading to a competitor or deciding the service isn't worth the cost.

Incidental churn

Also voluntary, but driven by circumstances outside the customer's direct dissatisfaction, such as a change in life situation, financial hardship, or relocation. These customers may have been satisfied with the product but left for reasons unrelated to its quality.

Types of customer churn models

Customer churn models are categorised based on a variety of factors, such as the type of algorithm, how it treats time and the level of prediction detail. Here's a detailed explanation of the types of customer churn models.

• Predictive churn models: These models use historical data to predict the likelihood that a customer will churn in the future. They typically employ machine learning algorithms to identify patterns and predictors of churn, outlined below.

• Logistic regression: This is a statistical model that estimates the probability of a binary outcome (such as churn/no churn) based on one or more independent variables. It's widely used for its simplicity and interpretability.

• Decision trees: These models use a tree-like graph of decisions and their possible consequences. They are easy to interpret but are prone to overfitting.

• Random forests: An ensemble method that uses multiple decision trees to improve predictive accuracy and control overfitting, random forests are more comprehensive than a single decision tree and often provide high accuracy.

• Gradient boosting machines (GBMs): GBMs are an ensemble technique that build trees sequentially, with each new tree correcting errors made by the previously trained trees. GBMs, like XGBoost and LightGBM, are powerful for churn prediction but can be complex to tune.

• Neural networks: Neural networks are deep learning models that can capture complex nonlinear relationships through layers of nodes or "neurons". They can be very effective, especially with large datasets, but are more difficult to interpret than simpler models.

• Descriptive churn models: Rather than predicting future churn, these models provide insights into past churn behaviour. They help identify trends, patterns, and reasons behind churn, often using clustering techniques or principal component analysis.

• Time series churn models: These models look at how churn rates evolve over time. They can be particularly useful for businesses with strong seasonal patterns or those trying to gauge the impact of specific events over time.

• Cohort-based models: These models analyse the churn rates of different customer cohorts. A cohort might be defined by the date customers signed up, the product they first purchased, or any other major event. This helps identify if certain cohorts are more prone to churn than others.

• Survival analysis models: Also known as time-to-event models, these models measure the time it takes for an event (churn) to occur. They're particularly useful for predicting when a customer will churn.

• Real-time churn models: These models generate instant predictions based on real-time user interactions. They require a strong data infrastructure and are used in scenarios where businesses can take immediate actions to prevent churn.

• Hybrid models: These models combine elements from different types of models to use their strengths and mitigate weaknesses. For example, a hybrid model might use a combination of a predictive model for churn likelihood and a survival analysis for timing.

Choosing the right model type depends on the specific business context, the nature of the customer relationship, the available data and the desired outcome of the modelling effort. It's often beneficial to experiment with multiple models to determine which provides the most accurate and actionable insights for a particular use case.

How to build a customer churn model for your business

Building a customer churn model is a multi-step process that involves analysing how you define churn, what your data shows you and which model is most useful for your business. Here's a detailed guide for creating a churn model that meets your business needs.

• Churn definition: Different businesses have different definitions of churn. For a subscription service, churn might be a customer cancelling their subscription. For an e-commerce platform, it might be a customer who hasn’t made a purchase in a certain time period.

• Data collection: Gather historical data that includes both churners and non-churners. Your dataset should include a variety of features such as customer demographics, transaction history, product usage data, customer service interactions, and any other relevant data that can influence churn.

• Data preparation: Clean the data (handle missing values, remove duplicates, etc.) and transform it into a format suitable for modelling. This might include encoding categorical variables, normalising numerical values, or creating time windows for predictive features.

• Feature development: Develop features that effectively capture the behaviour and characteristics of your customers. This can include aggregating transactional data into meaningful metrics, calculating usage frequency, or deriving other insightful attributes from raw data.

• Exploratory Data Analysis (EDA): Before building your model, conduct EDA to find the patterns in your data. Look for correlations between features and churn, identify outliers, and understand the distribution of key variables.

• Algorithm selection: Choose a machine learning algorithm to predict churn. Logistic regression, decision trees, random forests, gradient boosting machines, and neural networks are common choices. The choice of model depends on the dataset size, feature importance, and need for interpretability.

• Training and testing: Split your data into training and testing sets to evaluate the performance of your model. The training set is used to train the model, while the testing set is used to assess its predictive power on unseen data.

• Model validation: Use metrics such as accuracy, precision, recall, F1-score, and Area under the ROC curve (AUC-ROC) to evaluate the performance of your model. Focus on the metrics that impact your business objectives. For instance, if the cost of false positives is high, you might want to maximise precision.

• Model tuning: Adjust the model’s hyperparameters to improve its performance. Consider using techniques such as grid search, random search, or Bayesian optimisation to find the optimal set of hyperparameters.

• Feature analysis: Learn which features are most influential in predicting churn to guide your customer retention strategies.

• Deployment: Deploy the model into a production environment where it can provide ongoing predictions. This might involve integrating the model into your business systems or setting up a batch process to periodically score customers.

How to use predictive analytics to reduce customer churn

Actionable insights and intervention strategies

• Use the model's predictions to develop targeted retention strategies. For instance, consider offering customers who you have identified as high-risk for churn personalised promotions, proactive customer service, or other incentives to retain them.

• Segment high-risk customers based on their characteristics or reasons for potential churn and tailor interventions accordingly.

Monitoring and feedback loop

• Monitor the model's performance on a continuous basis. Adjust and retrain the model as necessary to respond to changing patterns in customer behaviour or business operations.

• Establish feedback mechanisms to capture the outcomes of retention efforts and regularly monitor the effectiveness of your retention strategies. Use this data to further refine the predictive model and intervention strategies.

Collaboration and organisational integration

• Share learnings and recommendations from the predictive analytics process across relevant departments to create a cohesive retention strategy.

• Integrate data-driven insights into decision-making processes to address customer needs proactively and reduce churn.

Top use cases for churn prediction models

Churn prediction is valuable across nearly any industry with recurring or repeat customer relationships, but it's especially helpful in sectors where customer acquisition is expensive or revenue depends on ongoing engagement.

How to measure the effectiveness of customer churn models

It’s important to measure the effectiveness of your customer churn models to ensure you don’t make business decisions based on false predictions. The following metrics can help you evaluate how well your churn models are performing.

• Accuracy: This is the percentage of total predictions that the model got right. While accuracy is a starting point, it doesn’t always give a complete picture, especially if the number of churners and non-churners is imbalanced.

• Precision: Of the customers whom the model predicted would churn, how many actually did? Assessing this metric is important because the cost of false positives (predicting churn when it doesn’t happen) can be high.

• Recall: Of all the customers who churned, how many did the model correctly identify? This is important if you want to capture as many true churn cases as possible, even if it means tolerating some false positives.

• F1-score: This metric combines precision and recall into one number for a balanced view.

• Area under the ROC curve (AUC-ROC): The ROC curve plots the true positive rate against the false positive rate at various threshold settings. The AUC measures the entire two-dimensional area underneath the entire ROC curve. A model with perfect predictions has an AUC of 1.0, while a model that makes random guesses has an AUC of 0.5.

• Confusion matrix: This is a table that shows the number of true positives, false positives, true negatives, and false negatives. It helps you see the types of errors your model is making.

• Lift: This metric demonstrates how much better the model is at predicting churn compared to random guessing. A lift greater than 1 indicates the model is better than random guessing.

• Business impact: The true test of the model’s effectiveness is its impact on business metrics. Are you able to reduce churn rates by acting on the model’s predictions? Are customer retention strategies informed by the model’s insights leading to increased customer lifetime value (LTV) or enhanced customer satisfaction? To evaluate the business impact, you can run controlled experiments (such as A/B testing) to compare the outcomes with and without the interventions suggested by the churn model. Observing changes in churn rates, customer satisfaction scores, and profitability over time will help ascertain the model’s real value.

How Stripe Billing can help

Stripe Billing lets you bill and manage customers however you want – from simple recurring billing to usage-based billing and sales-negotiated contracts. Collect and retain more revenue, automate revenue management workflows and accept payments globally.

Stripe Billing can help you:

• Offer flexible pricing: Launch quickly with built-in usage-based and hybrid pricing models – including flat-fee plus overage, credits, and more. Support for coupons, free trials, prorations, and add-ons is built in.
  
• Experiment and iterate on pricing: Respond to user demand faster with no-code tools to adjust usage-based rates, manage pricing cohorts, and inform pricing decisions with granular usage and spend analytics.
  
• Align pricing to customer value: Meter and charge by the usage dimensions that deliver the most impact, and define pricing in ways that directly reflect how customers gain value.
  
• Increase revenue and reduce churn: Improve revenue capture and reduce involuntary churn with AI-powered Smart Retries and recovery workflow automations. Stripe recovery tools helped users recover over US\$6.5 billion in revenue in 2024.
  
• Boost efficiency: Use additional Stripe solutions for tax, revenue reporting, and data to consolidate multiple revenue systems into one. Easily integrate with third-party software.
  

Learn more about Stripe Billing or get started today.

FAQs about building a customer churn model