Erum Manzoor is Senior Vice President—Global Wealth Technology of Citigroup and thought leader, Linkedin Top Voice, author and speaker.
AI is massively reshaping the financial world, but fully integrating this technology is burdensome and expensive.
As Citi explained in a recent report, AI will profoundly change the future of finance and money, and “it could potentially drive global banking industry profits to $2 trillion by 2028, a 9% increase over the next five years.” Heavyweights such as Citi, JPMorgan Chase and Capitol One, among others, have committed to AI investments for gains in efficiencies, enhanced customer experience and enhanced risk assessment.
However, behind the headlines, companies face tough regulations, an entangled web of legacy systems and the creation of customer-ready solutions that continue to pose major challenges.
Solving The Regulatory Maze: Data Privacy And Compliance Challenges
Financial institutions are among the most regulated environments on the planet, and the deployment of AI introduces troublesome compliance issues that are becoming increasingly strict by the year.
Within the European Union, for example, heavy fines under the General Data Protection Regulation have been levied against organizations for mishandling consumer data. Each serious breach could amount to a fine of up to 4% of a company’s global revenue.
Similarly, with the AI Act, the EU classifies AI applications according to risk, placing financial applications like credit scoring or fraud detection under the “high-risk” category. That means more disclosures, auditing and explainability—each of which adds more compliance and administration. In the U.S., state-by-state data privacy laws are either proposed or already in place—such as the CCPA and the NYDFS Part 500 regulation.
This is a fragmented regulatory landscape that forces banks into region-specific AI protocols, driving up implementation costs and elongating the timescale required to fully implement AI-driven services.
Integration With Legacy Systems: Costly And Complicated
Most banks, as well as insurance and asset management companies, have big investments in legacy infrastructure—in many cases, going back decades. Although still operational, older systems are often not structured for the heavy computation loads or complex data analytics that are required by AI.
Deloitte points out that many financial institutions still rely on core operations managed through mainframe-based infrastructure. With this infrastructure, in general, AI requires extensive and expensive workarounds for integration. As McKinsey pointed out in a 2021 report, legacy systems “often lack the capacity and flexibility required to support the variable computing requirements, data-processing needs, and real-time analysis that closed-loop AI applications require.”
Most smaller institutions cannot afford bespoke repairs to update their systems, meaning access to scaleable AI remains beyond them and is deepening a technological divide between global players and regional banks.
Legacy systems also use incompatible data formats, which create difficulties for AI algorithms to operate on existing information. One challenge, for instance, is if AI tools often cannot access siloed customer data from operations around the organization, which can significantly limit a model’s effectiveness.
This lack of interoperability generally results in fragmentations of AI systems that cannot reach the complete dataset required for their accuracy. In such cases, AI fails to live up to its pledge of allowing advanced customer personalization and risk analytics.
Creating Customer-Ready Solutions: Usability, Trust And The Talent Gap
Creating customer-ready solutions using AI presents a host of different challenges. For financial firms, usability, scalability and building customer trust are a few of the major concerns if the technology is to see widespread acceptance.
For example, AI-powered customer-service bots are multiplying but have variously suffered from everything from minimal capability to resolve problems to subpar hand-offs with human agents. Meanwhile, perfecting those tools to deliver seamless customer experiences reliably remains an uphill task: According to a SurveyMonkey report, “61% of consumers say humans understand their needs better than AI.”
But banks are aware of this issue. According to a report from EY-Parthenon, 60% of banks are implementing GenAI to make customer experience enhancements—one of the greatest motivators to implement the technology, second only to productivity enhancements.
The lack of transparency is also an issue. This is especially true of applications related to credit assessment and investment advice, among others, where a lack of transparency could cloud any decision and affect bottom-line financial outcomes. The drive toward explainable AI (XAI), an expensive undertaking wherein most systems do a lot more than carry out complex computations, also explains those decisions in clear, simple language.
And add to this the talent shortage that plagues the industry. A report by Rivel Banking found that “80% of community banks and credit unions list staffing as their biggest concern.” Even for institutions with moderately sufficient budgets, the recruitment and retention of talent to construct AI solutions is hard to find.
The Way Forward: Structural And Strategic Shifts Required
These challenges underpin the truism that while AI can transform finance, the potential will be realized only with more than just technological investment. This involves several structural and strategic changes, ranging from regulatory harmonization to the gradual replacement of legacy systems.
Until then, AI in finance remains an elite tool available to a few institutions—which are capable and willing to meet the financial, regulatory and operational demands. For the rest, the transformative promise of AI will continue to remain out of grasp.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
