Tune into the news nearly any day, and you’ll hear pundits giving their best-informed guesses about the prospects for and timing of the next global slowdown. Whether the economic tide turns in the next year or sometime in the next five years, change is inevitable.
What do these prospects mean for the banking sector, and where can institutions double down now to be sure they’re in the best possible position when the next slowdown makes its mark?
There is little debate that changes to the U.S. financial system and its regulation – including more rigorous and frequent stress testing – have made the industry and its institutions healthier and more secure. As important, however, is the vital role technology – including advanced analytics and, increasingly, artificial intelligence (AI) – has and will continue to play in helping banks to improve overall efficiency, attract more profitable customers, and reduce risk.
The industry has made great strides in recent years, and there are still significant gains to be realized. Organizations that actively pursue the creation of a robust and flexible analytical foundation today best position themselves for strong performance even in an unpredictable future.
Capturing New Cost Efficiencies
Interest rates are at near record lows while the price of talent escalates, and compliance costs are stable, at best, or in many cases continue to rise. As such, margin pressures remain a core concern for financial institutions of all sizes and scope. Elevating operational efficiencies is a top priority with institutions looking well beyond their branch strategies for opportunities to drive efficiencies and reduce costs.
Process automation, boosted to new levels by nascent AI initiatives, yields unprecedented opportunities to capture greater efficiencies. One area in which these technologies are reducing time-consuming and labor-intensive manual processes is in regulatory compliance, specifically financial crime and compliance management (FCCM).
Within the last decade, institutions have made great strides in terms of deploying automation to screen for potential nefarious transactions and activities. A byproduct of these first- and second-generation rules-based solutions has been a high false-positive rate, which drives new levels of manual investigation. Today, firms seek solutions that enable them to adopt a risk-based approach to FCCM that optimizes the quality of alerts.
To achieve this goal, banks are increasingly turning towards advanced FCCM approaches (layered on their rules-based environments) that leverage graph analytics, machine learning (ML) and other AI techniques to improve detection, drive down the incidence of false positives, and thereby drive down associated costs. As automated FCCM tools become more mainstream and available out-of-the-box, banks will realize a more sustainable and cost-effective framework for addressing financial crime compliance.
Cultivating More Profitable Customers
For many years, the financial services industry subscribed to the 80/20 rule—that is, roughly, 80 percent of customers fail to turn a profit and the 20 percent that were profitable could yield significant revenue. Today, this is no longer the case, and firms must pursue a more strategic approach when it comes to customer profitability.
Customer profitability varies widely – and the ability to accurately determine current and future profitability remains elusive in many organizations. At the same time, capturing wallet share of the most profitable customers is more pressing than ever as options expand, and customers increasingly purchase new, more lucrative products, such as credit cards, insurance and investments from challenger banks or fintechs.
It started with the goal of achieving a 360-degree view of customers and their relationships with the bank. To determine true profitability, however, the bank also must consider the current financial value of the customer across the enterprise as well as the costs associated with obtaining and maintaining that relationship – including on-boarding, customer service, marketing, and more. It is important to understand how a customer interacts with the firm, as there are dramatic cost differences between serving a customer in a branch versus via an online portal or mobile channel. One also cannot neglect factoring in the potential risk associated with each customer.
It’s also important to consider that calculating current profitability is just the start. Firms must then work to stratify customers, identifying not only those that are currently most profitable, but also those that have the potential to be. AI and ML increasingly factor into this process.
However, in most financial institutions, the data they need to gain the comprehensive insight required today remains locked in disparate systems with little integration.
Today’s plan to optimize profitability must encompass customer insight systems and a data infrastructure that includes:
A highly secure, single platform for enterprise-wide customer insight;
A unified data model that integrates varied structured and unstructured data from internal and third-party systems;
Flexibility to configure the solution and scalability as the number and types of data sources multiply rapidly;
Deep, native, and flexible integration between enterprise resource planning, risk management, governance, risk, and compliance, and enterprise performance management systems; and
Advanced, pre-seeded analytical models that incorporate ML to support more meaningful engagement and highly personalized offers that expand relationships and reduce attrition for the most profitable customers.
Technology plays a vital role in helping financial institutions reduce risks, especially when it comes to less profitable or risky customers. Today, ML and graph analytics increasingly factor into the mix. Credit risk is a perfect example of this evolution.
One of ML’s most prevalent financial use cases to date has been in calculating credit default risk. And, as ML techniques advance and the industry’s appetite for it grows, its role in credit risk management will evolve. The universe of available data sources and data layers that can apply to credit risk projections are expanding as well, offering the prospect for even greater insight and improved accuracy.
Historically, credit risk estimation has been a linear calculation – a review of income, assets, and debt combined with a look at the applicant’s history and a dash of relationship. This approach has proved imperfect, as market events have illustrated. The attraction of ML for credit risk management is that it considers numerous factors beyond the reach of traditional methods, and it appears to do better at incorporating non-linear factors. Looking ahead, expanding the use of semi-supervised and unsupervised ML learning has great potential in credit risk management, but widespread adoption remains elusive, in part due to business and regulator caution over its “black box” calculations.
Layering of data creates a richer image and may provider much greater insight into assessing an individual’s credit risk. For instance, in addition to financial statements and loan payment behavioral data, firms in some jurisdictions are increasingly leveraging data as diverse as transaction information, geography and whether the user has a mobile phone account or pays their utility bills on time. The ability to gather and use more varied, non-traditional forms of data is key to improving our ability to accurately assess risk – at the customer, institution and even system level.
Firms should also consider incorporating macrofactors, such as the broad institutional data from banks to improve prediction accuracy. For example, the housing price index in a locality could help predict the levels of credit card default there. It follows logically that, as the ability to assess credit risk at the individual portfolio level improves, ML will get better at predicting the overall systemic risk, as well.
If an organization does not already have an ML model to support credit risk, or a system that enables ML models to be built, it should start looking at ML at least for a detailed portfolio because that is an area in which there is enough internal data for a supervised algorithm to properly calibrate itself.
Bank Technology and Its Benefits Are Not an Overnight Effort
While banks may be eager to reap the benefits of technology, including AI and ML, they must remember going digital is a journey. It does not need to happen all at once. Banks should take a systematic approach to adopting technology and should look for solutions that enable them to continue to evolve and grow. Banks should pick and choose in terms of where to make changes – understanding that rip and replace is rarely the key to immediate or long-term success. A steady and strategic plan wins the race.