Published by Forbes.com on September 13, 2022
Forbes Councils Member
Understanding customer churn is critical to sustaining a healthy business. Knowing how and why customers leave and what can be done to win them back underpins efforts to target new markets and product requirements. As a data-driven process, churn management has matured into a mission-critical task that companies depend on to make sense of the market.
But no two companies are alike. An acceptable churn rate for one may be catastrophic for another. Even identifying when churn has occurred can be challenging: It may be sudden and deliberate or gradual, implicit and unconscious. For example, a SaaS subscriber may let a service expire for unknown reasons and then rejoin a month later. Does that count as churn? If so, how is it quantified?
Every company is bound to have some level of churn. No matter how great your company’s products or services are, people and markets are always changing. The key is finding the sweet spot where you’re not forced into an economically unsustainable model of finding new customers to replace old ones. This challenge isn’t new. But today’s digital landscape allows us to take a more granular look at customer behavior—one that calibrates the retention versus acquisition question to individual behaviors.
Although it’s deceptively simple to calculate turnover with a simple equation—divide customers lost by the total number of customers in a given period—learning how, when and why churn is occurring is more nuanced. You need a comprehensive view of the entire customer experience. One way to navigate that complexity is with data and artificial intelligence (AI).
Let’s look at a few inputs that can be used to model a churn prediction algorithm.
• Customer Demographics And Psychographics: Examples include age, location, income, job type and family status.
• Transactions: Examples include subscriber history, SKUs, product returns and purchase frequency.
• Pricing: Examples include promotional offerings, price changes, fees and shipping rates.
• Economic Factors: Examples include unemployment rates, interest rates and seasonality.
• Competitor Activity: Examples include advertising, pricing and geographic proximity.
• Customer Behavior: Examples include the frequency of usage, level of engagement, service calls and preferences.
• Customer Journeys: Examples include product discovery, learning curve and satisfaction.
Some of this data is easier to come by than others. Particularly challenging are the qualitative features like a customer’s mood during a customer service call or survey or how a platform update influenced their engagement. But AI can help with that, too. Consider a few ways marketers are using AI to collect and quantify customer activity.
• Identifying Trigger Events: Customers are not usually forthcoming with their motivations for leaving, but historical data can be aggregated to identify potential triggers, including price hikes, service outages and customer service interactions. AI systems can categorize behaviors surrounding these changes and A/B test to delineate risk clusters.
• Analyzing Customer Sentiment: Natural language processing (NLP) can analyze customer interactions via emails, text messages, service reviews and phone calls and identify sources of friction. This data can be leveraged to update technologies, modify products or retrain customer service.
• Defining Explicit Vs. Presumed Churn: Not all churn is absolute. Sometimes, customers fade away, making their departures difficult to interpret. By mapping historical churn data—such as fewer purchases or interactions—AI systems can identify models of behavior based on gradual or “presumed” churn, feeding results into corresponding risk categories.
The process of converting data into a 360-degree view of a company’s churn rate begins with classification: Data scientists use machine learning (ML) algorithms to segment clients into risk categories based on weightings of input variables, such as subscriber history, platform usage and customer service interactions. Using that data, analysts tune rules or probabilistic model structures into ML, rating different combinations of inputs to produce a value. Those values indicate a given customer’s churn risk. But before a model is released, it should be tested against competing models to find the most accurate one.
That’s a simplified way of explaining how companies use AI to predict churn, but the real value is in what they do with those predictions. Some AI systems are equipped with recommendations for the “next best action” (NBA)—the steps a company should take to win back at-risk customers. Some examples of possible NBAs a company can take based on customer data include:
• Personalized pricing for customers likely to churn within certain periods (e.g., one month or two months).
• Personalized messages to prompt at-risk customers to use a product.
• Adjusted credit limits for customers with a reduced behavioral risk score.
• Automated messages informing customers of an issue or providing account details for those likely to call a service center, thus avoiding expensive calls.
• Best next product offering for a customer who’s likely to buy it or additional product features to increase engagement.
This ongoing dance is the holy grail of customer relationship management. Like churn management, it’s best informed by data. Data scientists can frame the NBA as a reinforcement learning problem—an approach to predictive modeling that rewards or penalizes a model based on performance—or as a batch analysis program that looks at large collections of data and behaviors using a plethora of predictive modeling techniques.
In the context of an NBA, the idea is to optimize long-term profitability over short-term losses. Because customers have nothing to lose when they’re considering leaving, their short-term interests should be privileged over those of the company. Framing the problem in these game theory terms leads to decisions on NBAs that prioritize customer retention over competing outputs.
Customer behavior isn’t as erratic and unpredictable as some believe. We only think it is because individuals can be fickle, and the human brain isn’t adept at finding patterns in massive datasets. With AI, though, companies can unearth those patterns, including why customer churn is happening and what can be done to stymie it. Some churn is inevitable, but there’s no reason to accept the mystery behind it. AI tools can enable a much more systematic approach to churn: how to better understand, predict and manage it.
View online