All companies in the consumer goods market and the enterprise sector have to deal with customer churn, as it can ultimately impact the company’s revenue figures and thus influence strategic decisions.
According to the authors of Leading on the Edge of Chaos, a 2% increase in customer retention (or a decrease in churn) is equivalent to a 10% reduction in costs. It’s no wonder that SaaS companies (and companies that care about customers) pay a lot of attention to churn prediction. In addition, according to various sources, it is 6-7 times more expensive to acquire a new customer than to keep an old one.
„There are only two sources of competitive advantage, the ability to learn more about our customers faster than the competition and the ability to put that learning into action faster than the competition.“
– Jack Welch, former Chairman and CEO of General Electric
Customer Churn refers to the rate of customer departure in a business or in simpler terms, the speed at which the customer leaves your business or service. Examples of customer churn include
- Cancelling a subscription
- Closing an account
- Non-renewal of a contract or service agreement
- Decision to shop at another store
- Use another service provider
Churn can occur for many different reasons and churn analysis helps identify the cause (and timing) of that churn, opening up opportunities to implement effective customer retention strategies. Here are 6 proven steps to ensure you’re focused on retaining your customers – we’ll just focus on step 2 and parts of step 3 for this article. This isn’t about blaming the product or anyone else for churn, but rather understanding the customer better and developing a strategy to improve customer retention.
- Recording customer behavior, transactions, demographic data and usage patterns
- Use these data points to predict customer segments that are likely to churn
- Create a model to model the company’s risk tolerance in relation to the likelihood of churn
- Develop an intervention model to investigate how the extent of the intervention might affect churn percentages and customer lifetime value (CLV)
- Implement effective experiments across multiple customer segments to reduce churn and promote retention
- Iterate this process (cognitive churn management is a continuous process and not a once-a-year exercise)
We also believe that the risk analysis – decision making – marketing segmentation approach is a generic structure that can be used for many business problems and not just churn analysis.
A predictive churn model is a simple classification tool: look at past user activity and check who is active after a certain period of time, then build a model that probabilistically identifies the steps and phases at which a customer (or segment) leaves your service or product.
A predictive churn model gives you the awareness and quantifiable metrics to work with in your customer retention efforts. This gives you the ability to sample customers‘ habits and intervene before they make the decision. Without this tool, you would be acting on general assumptions, not a data-driven model that reflects how your customers really behave.
Without a deep understanding of your customers and their behavior, it’s difficult to retain them, so the first step in building this model is to understand your customer behavior from customer data points. What kind of data do we need to identify the triggers that ultimately led to them leaving your company.
Customer information
- Postal code
- Income class
- Gender
- Profession
- Do you have children in the household?
- How do they find your website/product?
- Do they open your newsletters and other trigger emails or click on any links?
Products
- Art des Product
- Product variety
- Voucher usage
- Product preferences or combinations
Purchase history
- Frequency of purchase
- Date of last purchase
- Time of day/season of purchase
- Value of purchases
- Payment methods
- Credit / branch credit
Customer interactions
- Servicefragen
- Shop visits / Online
- Complaint Resolution
- Priority of the complaint
- How do they complain – email or phone or Twitter?
- Frequency of complaints
These are just a few examples to get you started. It is important to know as much as possible about our customers to know what event causes them to drop out and look for the next competitor. The more relevant data you collect, the more accurate your model will be. Once you collect this customer data in a central data sink and start analyzing, you will start to see trends that will give you insights into customer churn. The consolidated data of all churners will also help you group behavior and establish patterns.
DATA PREPARATION
Once you have collected enough data for your analysis, the next step is data preparation. This is the most time-consuming but important step in data analysis. As the famous saying goes, „Garbage In, Garbage Out“. Your analysis will be as good as the data it is based on.
You can use these three criteria to ensure good data quality:
- Complete
- Clean
- Exactly
Complete – Do you have all relevant dimensions? What percentage of the data has missing values or zeros? You can fill in some of the missing values through data exploration, e.g. values of ‚State‘ based on the customer address or ‚Product Category‘ based on the item number etc.
Clean – Do you have multiple values for the same dimension? e.g. NRW/Nordrhein-Westfalen/Nordrhenwestfalen; different abbreviations for a product name etc. If so, you can perform data cleansing to ensure consistency.
Exactly – Are there negative revenue values or €0 revenue for some transactions?; date conflicts; ‚NONE‘ or ‚N/A‘ values etc. You can decide after consulting with stakeholders whether to include or exclude such bad data in the analysis.
In predictive and diagnostic analytics, there is one more step in data preparation – creating the target variable. In case churn analysis is performed, it could be a binary column like „will cancel?“ You can enter the values for this variable by analyzing the historical data. For example, a value of 1/TRUE for customers who canceled their subscription and 0/FALSE for those who renewed it.
EXPLORATIVE ANALYSE
*Example implementation at the end of the article*.
Unlike traditional statistical modeling (like linear regression), machine learning predictive models are generated by a computer algorithm. A critical skill for building the churn model is being able to ask as many questions as possible. This allows you to test, retest, and qualify your assumptions and data before moving on to implementing a model. Here are some basic questions to get you started.
1. What is the degree of correlation between the available data points and the fluctuation?
From this question, you can derive an answer like „of all the customers who left in the last quarter, 80% of them filed a complaint over the weekend,“ or „of all the customers who left in the last two quarters, 56% of them never used the reporting tools in our software.“ Your goal is to use this exploratory question to uncover one or more correlation patterns in specific data points about churn. While you’re at it, ask the same questions for current customers as well and see if your general assumptions are true or if they’re just random clues. If your assumptions are true, you’ve discovered a trend that needs attention.
Using machine learning models can also show you some connections that are normally invisible to the human analyst.
2. At what stage in the product life cycle did customers cancel?
When did these customers leave? At the end of the initial trial period, when the subscription expired, or was a feature too difficult to use, or was it an external event? How long did they use our software before leaving? Did they make a call to customer service and remain dissatisfied? What are some of the key trends for the customers who did not leave and renew their subscription?
Continue to explore each individual data point and its relationships – machine learning models usually do this much better than humans and find unexpected patterns.
3. What impact do the different customer segments have on lifetime value (LTV)?
The result of a good model is to find triggers and behaviors that are synonymous with an increase in churn or increased retention and decline. For example, if every customer in a certain segment had an LTV of €100 and your model shows that X number of them are moving away, then you can easily highlight the negative impact on revenue numbers due to this churn and help business leaders quantify and prioritize customer retention strategies.
What comes after analysis
If you can’t find at least three highly correlated attributes, you’re not asking the right questions about your data or perhaps you don’t have the right data sets.
Once you’ve identified the customer segments and their behaviors that lead to churn, you can develop recommendations for your team to increase customer retention. Designing, implementing, and deriving results from smart experiments is a topic for another article, but here are some quick ideas:
For a telecommunications company that is seeing a migration of high-income populations that use more text messaging than actual phone calls, a niche plan can be created that targets that segment.
For an online project management service that experiences rapid customer churn during the trial period, it may be useful to offer explicit training during the onboarding of new customers to make the trial period even more interesting.
