Customer Churn and Why Your Model Isn’t Working
Many companies are facing a challenging business problem. They have invested a lot of time, effort and money into sales and marketing to attract the right prospects, nurture them into leads, and convert them into paying customers.
However, they are starting to notice that they are losing these same customers, long before they become profitable.
I am sure you have read the stats; according to Harvard Business Review, acquiring a new customer is anywhere from five to 25 times more expensive than retaining an existing one.
Why? Because existing customers have already built a relationship with you. They know you, and as a result, they’re more likely to buy new products or continue to pay for existing ones. With new customers, you have to start the long and expensive sales and marketing cycle from the beginning.
That’s why it’s so important to understand which customers are likely to leave and determine what you can do to stop them – customer churn prevention. With the right data, it is possible.
Using Data to Predict Customer Churn
Over the years, we have frequently been asked to help our clients solve this problem. With predictive analytics, you can build a churn model based on trends and previous behaviors of customers who have left. This allows you to identify patterns and determine the likelihood of an existing customer leaving before they do so. By knowing this far enough in advance, you can then take action to reduce your churn rate and increase your profitability.
Churn modeling can be very effective in identifying at-risk customers; but in my experience, it frequently fails to stem the flow. There are two main reasons for this:
- An ill-conceived model
- An ineffective strategy
Framing the Customer Churn Problem
Far too often the predictive model that is designed isn’t the right one. Before diving in, there are a few things you need to ask to ensure you are framing the problem correctly.
Do you have the data necessary to build an accurate model?
After all, the strength in your predictive model is dependent on the strength of your historical data. To predict the future, you need a good view of customer behavior in the past. For example, say you wish to predict who is likely to leave in the next three months. You need to identify a group of customers that have churned in the past three months and a group that did not, and be able to reconstruct a view of customers three months ago. Ideally, you also have snapshots of data which let you see changes in that customer data over the months preceding that three-month window. For businesses with seasonality in behavior, you would need that data for a minimum of a year of data in addition to the three months.
How does customer churn work in your business?
Probably the biggest challenge in framing the problem is truly understanding how churn works in your business. How do you measure churn? For example, in a subscription business, you are perhaps most interested in who doesn’t renew on their anniversary date. You would likely then design your model to look only at renewing customers and predict their likelihood of renewal. Or perhaps renewal is relatively high once they are entrenched and you are more interested in defection during the first year.
What are your customer “moments of truth”?
Moments of truth, like renewal date or expiry date, make framing the model much easier. In businesses where customers buy products and are under no obligation to buy more, the definition of churn is even more challenging. It is often necessary to understand the expected purchase cycle for a product to determine how you will know when a customer has churned. This applies to any industry. For example, someone doesn’t need their residential windows washed every week, but they might need them done annually.
Another example is in banking. Perhaps you want to understand churn on checking accounts. Is the date the account closes the real churn date? Or in fact, did true churn happen many months earlier when they stopped having their paycheque deposited and severely reduced the number of monthly transactions they made? Determining this will have a huge impact on the success of your retention initiatives. After all, the point of churn modeling is rarely to predict churn alone, your real goal is to avert it.
I Know Customers are Going to Leave, Now What?
This is where another challenge comes in. Churn models often ‘fail’ because there is a poor plan about what to do with them. The key question is, now that you know who is at risk, what are you going to do about it?
Can you identify someone early enough to retain them? In the banking example, predictive analytics could easily identify that those who have rarely used their checking accounts in the last quarter are likely to churn. But by the time they have displayed the behavior, it is almost guaranteed that they have opened another account somewhere else and have already transferred their paycheque and direct debits – making it nearly impossible to win them back.
Some churn cannot be averted no matter how early you predict it – for life insurance or reverse mortgages churn often occurs when the customer is deceased – not something most companies can do much about! So predicting it may be moot.
Even when it is possible to predict in a timely manner, research is required to understand what would be necessary to avert it. The model sometimes has the answers; perhaps customers who pay a lot of service charges are likely to leave. Offering a flat rate package with no incremental charges may address this pain point. Or perhaps those you bought in on a particularly discounted promotion leave when they receive their first full-price bill.
Taking Action to Increase Profits
Getting results from your data means you need to understand what it is telling you and what it isn’t. Your data won’t tell you who is definitely going to leave or even why. Instead, it will provide you with a likelihood based on the customer’s previous patterns and behaviors. Using this, you can rank customers and focus your efforts on who is most likely to leave.
Once you have established this model you can start adding other information and variables to dig deeper into your results. For example, there may be customers who have a high risk of churning but aren’t profitable to you, either because they do not spend enough or because they cost too much. Similarly, if the cost of saving a customer is too high due to the offer itself or the amount of time and resources it will take to convince them to stay, you may end up eliminating the value of retaining them.
Your success with a model will be dictated by what you learn and what you plan to do with that learning. Sometimes your efforts should be directed at saving those at risk today; in other cases, it may mean rethinking how you acquire and service customers to avoid future churn.
Are you able, and willing, to take steps to try and improve your retention rate, and do you have the budget and resources available to do so? If not, the information you get might be interesting, but it won’t have an impact on your results.
It Is Worth It
As we’ve seen, understanding churn can have a big impact on your business results. Bain & Company found that a 5 percent increase in customer retention produces more than a 25 percent increase in profit.
A churn model can be a key part of your strategy, but alone it will not be enough. You need to take the time to not simply predict churn but truly understand it and design programs to address it; the work may be daunting but the rewards are significant.
Learn more about how Shift Paradigm is redefining the Data and Analytics landscape here.