Customer churn is inevitable. Customers will abandon products and services after years of using them. And they will leave and switch to a competitor even if they’ve been satisfied customers. The reality is that in today’s increasingly dynamic marketplace, consumers are spoiled for choice when it comes to brands, and in many cases, face low barriers to make a change. In addition, with the continued digital revolution, advertisers have more ways than ever before to reach them at the moment it really matters.
As companies monitor their customer churn and ask the questions “Why? What happened?” the answer is never simple. Customer churn can be related to price, performance, the competition, product, service, support and any combination thereof.
What marketers know is that customer departures and defections can rarely if ever be attributed to a single reason based on a single episode at a single point in time. Customer churn is often the result of many factors that continually change over time. Unfortunately, by the time most marketers realize that a customer is at risk of churning, they don’t have enough time to act in ways that will preempt it.
But that doesn’t stop marketers from investing heavily in trying.
Nearly every CMO is focused on how to keep customers longer and increase their lifetime value. Key to customer retention success is understanding customer churn, and being able to analyze, predict and influence the behavior of customers by deploying the right marketing actions at the right time. The challenge, however, is that the analytics underpinning retention marketing programs relies on churn models that are often static, which results in retention marketing that is too late in influencing the customer.
So how frequently do most churn models score customers? And if it were more often, would a marketer’s ability to accurately predict churn improve?
A recent study by Amplero looked at whether churn scoring frequency drives performance when it comes to predicting churn. A large B-to-C brand traditionally only scored users against a churn model one time per month, and then used that score throughout the month as the basis for retention efforts. However, as the chart below shows, the percentage of churners correctly identified by the traditional model decreased by 20 percent as the scores became stale over the course of the month. As time passed after the scoring took place and before the scores were used for retention efforts, new events were ignored, and the performance of the model in accurately predicting churn decreased.
To test the effects of frequency, a new churn model was configured to operate on a daily cadence, so that scores remained fresh. In addition to driving performance on Day One that was twice as high as the traditional model that scored customers monthly, data and scores were refreshed daily resulting in sustained performance lift. The performance of the new model relative to that of the increasingly stale alternative model improved with the increasing delay.
Importantly, as a churn score becomes increasingly out-of-date, the amount of time left to interact with a churner is decreasing, and may become negative, representing a missed opportunity.
Models that capture the most recent event history allow for more precise prediction of at-risk customers, and ultimately lead to more relevant and timely context-based marketing actions.
When marketers rely on stale churn models, they miss important signals about their customers. Keeping data fresh and increasing the frequency of analysis are key to maintaining an accurate customer assessment and the first step to maximizing retention marketing and increasing the customers companies keep in the long-term.