According to this article by Frederick F. Reichheld in the Harvard Business Review, the average company loses about half its customers in a five-year period. When customers see a loss of value, they churn. The ultimate goal of a great loyalty strategy is to increase the perception of your solution’s value in the eyes of your customer, exactly when you need to. To achieve that, you need to be able to predict churn, many in the industry turn to leading (or lagging) indicators to indicate where efforts need to be focused.
Some common leading indicators for churn include:
- Numerous complaints. Analyse the number and the severity of support issues raised, and how quickly your team resolved them.
- Irregular payments or expired payment methods. Monitor if customers pay you regularly and on time.
- Low response rate. Observe customers’ email-open rate and/or replies to your agents in a live chat.
But identifying whether these incidents occur before your customers churn, requires further analysis. A higher churn rate might occur from these happening in combination, or there might be indicators in your database that are unique. If these unique indicators arise in high volumes, you should pay attention to these more than the common ones prescribed by the industry. You should also be aware that just because your customers are exhibiting the indicators does not guarantee they will churn – your database of customers is unique to your company and so are their behaviours. Don’t merely look for indicators that are common across multiple industries and companies.
What’s the alternative?
The key to accurately predicting churn is unbiased data science.
If your customer base is small, you could likely analyse it manually to look for markers of churn – leading and lagging indicators as defined by marketing professionals everywhere. But algorithms pick up patterns that we humans miss, and preconceived notions don’t blind them. E.g., we might dismiss an event that we consider unrelated and therefore could not possibly have an effect on churn. But an algorithm can identify correlations and causation based on patterns and historical data. Analytical models can determine which factors have been indicators of churn in the past in your database.
If you have a large customer base, data analytics and machine learning is your new best friend! An extensive database means you just can’t work through the data to identify churn indicators manually, and analytical models will give you an opportunity to identify patterns that indicate churn.
How do I get started with using data analytics to predict churn?
Understanding your customers better with data analytics always comprise of these four steps:
- Understand the business problem you want to solve
- Acquire the data needed
- Develop analytical models
- Incorporate analytical insights in an actionable business strategy
You’ve already identified what question you want to answer: which customers are going to churn? Next step is collecting the data you need. You’ll need data of customers who have churned in the past, as well as data of your current active customer base. The sources of the data can include transaction data, marketing and communications data, credit bureau data, demographic data, website data, search data and customer service data. The indicators could be contained in any of these sources – or even a combination of them. The more data that’s available, the more accurate the prediction will be.
Once you have all the data you need, you run a robust analytical process to analyse your data and identify which of your customers are likely to churn.
We’ve developed Genius, a Machine Learning as a Service that uses our Genius Artificial Intelligence engine to not only identify customer most likely to churn but also which are worth keeping and which worth re-activating.
Contact us if you would like to discuss how we can help you optimise the use of your business data.
