Churn Is a Top Revenue-Leakage Problem for Banks:
Is Deep Learning the Answer?

The impact of churn within the financial services industry is striking.

BCG research found that attrition affects 30% to 50% of a corporate bank's client base and spans all products and segments. Corporate banks annually lose 10% to 15% of gross revenues to attrition. The BCG analysis indicated that full relationship exits account for an average of 1% to 2% of gross revenue loss. Other forms of attrition, such as partial defections (clients close a few accounts or stop buying some products) and volume cuts (clients rein in the amount of business conducted across the portfolio as a whole) do the most damage. According to the study, relationship exits that are more gradual account for 9% to 13% of gross revenue loss per year.

The numbers are similar on the retail side. The Accenture Banking Customer 2020 Survey found that over the last five years 18% of retail bank customers switched providers completely and 27% added new providers. The study revealed that over the same period the combined switch rates (customers who switched completely from one bank to another or partially by adding new providers to existing ones) increased globally by 8%. Each point of attrition reflects as much as 2% of net income loss for banks.

When one considers that financial institutions spend an average of as much as $300 to attract each new retail customer — and more to attract commercial customers — the cost of replacing customers can quickly add up. And there's a distinct possibility that many of those customers won't last. Thus, the process of replacing customers continues in a repetitive cycle, draining revenues and operational efficiency.

Customer and company factors

The impact of customer churn on financial services is considerable for many reasons. Banks not only contend with traditional life consequences that result in churn (such as death, divorce and displacement), but also an increasingly competitive landscape altered by mobile technologies, digital transformation and greater customer expectations. The advent of electronic banking, online virtual banks and non-bank providers has de-personalized the banking experience, resulting in less loyalty and more attrition.

Most banks have enough data to reduce the instance of churn which, when decreased even 5%, can boost net profits as much as 80%. However, many fail to either analyze their data accurately or quickly enough to reduce attrition rates. Employing expensive data scientists and classic machine learning techniques can take months to analyze the bevy of data sources required to determine the feature characteristics for churn — during which time customers may leave.

The good news: Artificial intelligence data science is now practical

The BCG research also revealed that by employing predictive analytics and arming relationship managers with data-driven insights and enablers, corporate banks can reduce total attrition by 20% to 30% — a result that would nearly double most banks' average revenue growth.

But breaking the attrition cycle requires more than simple awareness of customer behavior. In today's world of digital transactions, understanding customer behavior requires analysis of big data — which includes transaction data, customer profiles and customer micro-segmentation data — to understand leading indicators for future customer behaviors.

Analyzing these data sources in real time at scale is now possible using artificial intelligence (AI) data science, which readily parses through these data elements to detect the features that can accurately predict churn and prescribe how to prevent it from occurring.

Using AI to drive customer analytics empowers financial institutions to recognize complex patterns faster — in real time. By using deep learning AI, banks can uncover previously unknown patterns and figure out what those patterns mean in regard to the customer experience.

This ability creates opportunities to pinpoint customer needs and wants, anticipate their actions, and prevent customer churn by surprising customers with custom offers, experiences and service. Organizations can take specific actions that positively impact the individual customer — creating a unique, personalized experience that builds loyalty.

Why all the hype about deep learning?

Deep learning is being positioned by many experts as the future of artificial intelligence. Most people are aware of its enhanced capabilities to recognize patterns in astounding amounts of big data over lengthy time periods. Those that follow AI are familiar with the high accuracy rates of its predictions based on those patterns. But there are actually three advantages of using deep learning:

1. Superior accuracy. Compared to classic machine learning, deep learning has a much higher rate of accuracy — up to 98% for image recognition as demonstrated in the 2016 ImageNet Large Scale Visual Recognition Challenge.
2. Complex data doesn't break it. Deep learning can encompass more variables than classic machine learning and is ideal for the numerous data sets required to understand financial customer behavior.
3. Identifies what you don't know. Most importantly, deep learning techniques are responsible for their own "feature detection," the process of ascertaining which characteristics of data are relevant to a particular use case such as churn. Feature detection in conventional machine learning requires humans to predetermine the features and relationships involved. Deep learning algorithms can identify new patterns or anomalies in raw data, therefore helping banks avoid “blind spots." Thus, organizations can readily understand what factors are affecting churn for customers, micro-segment them according to those factors, and deliver strategies to retain them.

Empowering banks with deep learning

Because most banks maintain extensive customer profiles and log millions upon millions of transactions, banking operations are ideally suited for deep learning. It can determine which customers are likely to churn and why, and provide insights into which campaigns proved effective in mitigating churn — all for a fraction of the amount banks typically lose to churn each year.

Deep learning can predict, at a granular level, the customers that will cancel subscriptions, stop using a product or service, and/or respond to competitive offers. By identifying churn in advance, companies can take actions that will change outcomes.

Deep learning can be a powerful tool for data scientists, IT teams and the banks for which they work. It can serve to reduce the manual labor involved in analyzing financial data for customers, enabling banking professionals to focus on long-term, value-adding strategies for reducing churn and increasing customers. Deep learning methods give banks new insight from their data and fuel the prescriptive analytics to minimize customer churn, increase employee efficiency and drive operational success.