Feature: Unlocking value through autonomous pattern recognition in financial services

FOR CHARTERED ACCOUNTANTS advising financial institutions or working within them, understanding real-world applications is no longer optional. AI/ML is reshaping how institutions acquire customers, manage risk, detect fraud, and optimise operations. This three-part series explores the concrete use cases delivering results today, beginning with the revenue-critical domains of customer acquisition and retention.

THE FUNDAMENTAL ADVANTAGE: AUTONOMOUS PATTERN RECOGNITION

Traditional analytics requires humans to hypothesise relationships, then test them. AI/ML inverts this paradigm entirely. Modern machine-learning algorithms process millions of data points across hundreds of variables, identifying complex, non-linear relationships that would be impossible to detect through conventional statistical methods.

Consider a simple example: traditional analysis might reveal that customers aged 35−45 with home loans are good prospects for vehicle finance. Machine learning

might discover that customers aged 37−42, with home loans originated in the past 18 months, who make regular grocery purchases on Tuesdays between 6 pm and 8 pm, and who have recently searched for child-related products online, have a 40% higher propensity to purchase vehicle finance − but only if the offer is presented via mobile app notification rather than email.

This capability to autonomously identify intricate patterns across multiple dimensions simultaneously is what makes AI/ML transformative. The machine finds relationships no analyst or business owner would think to test.

INTELLIGENT REVENUE GROWTH: TRANSFORMING ACQUISITION ECONOMICS

Customer acquisition in financial services is expensive. South African banks and insurers face acquisition costs ranging from R500 to R5 000 per customer, depending on the product. AI/ML is fundamentally changing the economics of growth by replacing spray-and-pray marketing with surgical precision.

Propensity modelling at scale

Machine learning models analyse behavioural, transactional, and demographic data to predict which customers are most likely to purchase specific products. Unlike traditional segmentation − which might divide customers into a dozen static groups − these models create individual propensity scores for every customer, continuously learning and adapting as new data arrives.

The impact is dramatic. Rather than broad campaigns targeting thousands of customers with 2−3% conversion rates, AI/ML enables hyper-targeted interventions achieving conversion rates of 10−15%. The mathematics is compelling: even with higher per-contact costs, the superior conversion rates deliver acquisition costs substantially lower than traditional approaches.

Next-best-action engines

Advanced implementations move beyond simple propensity scoring to holistic optimisation. Next-best-action engines consider multiple factors simultaneously: customer lifetime value, product profitability, regulatory constraints, current customer sentiment, channel preference, and competitive threats. The system autonomously determines not just what to offer, but when, how, and through which channel.

These systems operate in real-time across all customer touchpoints. When a customer logs into mobile banking, the engine can determine the optimal interaction in milliseconds. When a customer calls the contact centre, agents can receive real-time recommendations based on the conversation context. The result is dramatically higher relevance and conversion whilst improving customer experience − customers receive offers they actually want, when they want them (and sometimes before they realise they want them).

Dynamic pricing and personalisation

In insurance particularly, AI/ML enables sophisticated risk-based pricing that transcends traditional actuarial tables. Telematics data from vehicles, wearable health devices, and behavioural patterns allow insurers to price risk more accurately whilst offering better rates to lower-risk customers. This creates a virtuous cycle: better customers receive better pricing, improving retention and profitability simultaneously whilst enabling financial inclusion for customers previously priced out of markets.

RETENTION: FROM REACTIVE TO PREDICTIVE RELATIONSHIP MANAGEMENT

Acquiring a new customer costs five to seven times more than retaining an existing one, and long-standing customers are substantially more profitable than new ones. Yet many institutions still treat retention as reactive damage control. AI/ML transforms retention into proactive relationship management, identifying problems before customers know they are leaving.

Churn-prediction models

Machine-learning models predict, with remarkable accuracy, which customers will close their accounts or cancel policies in the next three months. These models analyse hundreds of signals: transaction frequency, balance trends, customer-service interactions, competitive activity, life events, and subtle behavioural changes invisible to human observers.

The models identify early warning signals: declining transaction frequency, reduced digital engagement, increased customer service calls about fees, competitive product searches online, or changes in payment-timing patterns. Critically, models identify which combination of signals predicts churn for specific customer segments − the patterns differ dramatically across demographics, product holdings, and tenure.

Early warning creates intervention opportunities. Instead of discovering a customer has left only when they submit closure instructions, institutions can intervene months earlier when retention is still possible.

Prescriptive intervention strategies

Knowing who will churn is valuable; knowing what intervention will retain them is transformative. Advanced ML systems move beyond prediction to prescription, recommending specific retention strategies for each at-risk customer based on predicted effectiveness.

These models recognise that churn drivers vary dramatically. One customer might be leaving due to fees, another due to poor digital experience, another because a competitor offered better rates, and another because of a single negative service interaction. The same intervention won’t work for all – indeed, the wrong intervention might accelerate departure.

Leading institutions now deploy models that predict intervention effectiveness for each customer.

For example, waiving fees might retain many fee-sensitive customers but only a small proportion of those leaving for better digital services. The model prescribes the intervention most likely to succeed – whether that’s a fee waiver, a digital experience upgrade, a personal relationship manager, or a product bundle.

This eliminates wasteful retention spending.

Institutions no longer offer expensive incentives to customers who would have stayed anyway, whilst ensuring high-risk customers receive interventions that actually work.

The competitive imperative

The institutions gaining market share in South Africa’s competitive financial landscape are those mastering these AI/ML capabilities. The advantages compound over time: lower acquisition costs enable more aggressive growth; higher retention rates build larger, more profitable customer bases; and the continuous learning nature of these systems means they improve with every interaction.

For chartered accountants advising financial institutions, understanding these applications is crucial. The question is no longer whether AI/ML delivers value in customer acquisition and retention – the evidence is overwhelming – but rather how to measure that value accurately, ensure appropriate governance, and deploy these capabilities whilst managing associated risks.

TO COME

In the next part of this series, we’ll examine how AI/ML is revolutionising credit risk assessment and collections optimisation, transforming institutions’ ability to manage risk whilst maintaining customer relationships.

Author
Danny Saksenberg is a UK- and South Africa-qualified actuary and founder of Emerge, a global AI/ML specialist firm. Active in AI since 2002, he is particularly passionate about its application in financial services, asset management and medical research.