Beware the risky whale: How AI is rewriting the rules of customer lifetime value

Imagine a bank's favourite credit card customer: their transaction volume is consistently high, they are rapidly utilising credit limits, and their engagement metrics are off the charts. Traditional marketing dashboards would label them a "whale"—a highly valuable, intensely loyal customer. But suddenly, the account defaults, taking all that projected future revenue down with it. This is the "Risky Whale" phenomenon: a scenario where high spending is not a sign of financial health, but a desperate scramble right before a crash. Treating a distressed borrower like a wealthy VIP is an incredibly expensive blind spot that costs banks and fintechs every day.

The root of this problem lies in how traditional financial analytics calculate Customer Lifetime Value (CLV). Historically, institutions have treated revenue generation and credit risk as two entirely separate, independent dimensions. They estimate how much money a customer will bring in, calculate the probability of default separately, and simply subtract the risk from the revenue. This "linear subtraction" paradigm completely ignores human behavioural realities. It assumes a customer's spending habits remain the exact same regardless of their financial health. In reality, a customer on the brink of a default often alters their spending behaviour drastically, intertwining risk and value in complex, hidden ways that traditional formulas miss.

To solve this critical blind spot, our research completely dismantles the linear assumption. We developed a Multi-Layered Stacked Generalisation Architecture—a sophisticated artificial intelligence framework that fundamentally rethinks Risk-Adjusted CLV. At its core, our model uses 'stacking'—an advanced machine learning technique where multiple base algorithms first analyse a customer's behavioural and transaction data independently. A higher-level 'meta-learner' then synthesises all these diverse predictions together to uncover hidden, non-linear patterns that a traditional equation would completely miss. Instead of treating risk and revenue as isolated silos, our architecture processes them dynamically through three interconnected stages:

Multi-layered stacked generalisation architecture

◆ Layer 1 — The Risk Layer: First, the system accurately estimates the precise probability and severity of a customer defaulting based on historical indicators. This foundational step anchors every subsequent calculation in grounded financial reality.

◆ Layer 2 — The Revenue Layer: Rather than calculating revenue potential in a vacuum, this layer estimates future value conditionally, stacking it directly on top of the risk insights discovered in the first step. Revenue projections are no longer blind to the borrower's underlying financial trajectory.

◆ Layer 3 — The Meta-Synthesis Layer: Finally, an intelligent meta-learner bridges the gap. It empirically discovers the complex, non-linear "exchange rate" between risk and value, pinpointing exactly how underlying financial distress is altering purchasing behaviour. This is where the architecture's true power emerges.

By shifting away from simplistic subtraction, the results of this stacked architecture are highly revealing. Our framework successfully identifies the intricate behavioural patterns that conventional models completely overlook. It isolates the Risky Whales, correctly recognising that their sudden spikes in transaction velocity are actually leading indicators of impending default, rather than genuine engagement. By catching these subtle, intertwined shifts, the meta-learning model delivers a dramatically more accurate assessment of a customer's true lifetime value, preventing systems from automatically rewarding high-risk behaviour.

The implications for the broader business sector are immense. For commercial banks, credit issuers, and digital subscription services, adopting a multi-layered, risk-adjusted approach to CLV is no longer just a technical upgrade—it is a necessity for survival in a volatile market. This framework empowers financial institutions to optimise their capital allocation, proactively manage high-risk portfolios, and design targeted interventions long before a default actually occurs. By seeing beyond the illusion of linear subtraction, businesses can finally separate true, profitable loyalty from the deceptive metrics of financial distress, securing their bottom line and fostering a much healthier digital economy.

Dr Shuvashish Roy, Senior Researcher, Research and Innovation Division, Prime Bank PLC. Md Tuhin Rana, Student, Department of Statistics, University of Dhaka

Disclaimer: The views and opinions expressed in this article are those of the authors and do not necessarily reflect the opinions and views of The Business Standard.