Generative AI in banking: promising use cases and potential hurdles

Fraud detection

Customer service automation

Credit score calculations and risk assessment

Algorithmic trading

Market trend and customer behavior prediction

Compared to conventional AI, Gen AI’s large language models (LLMs) can learn patterns and structure from even larger volumes of data, including information from unstructured inputs. In banking, this represents a shift towards more sophisticated and creative AI models (such as GPT, Generative Pre-trained Transformer) that can generate highly personalized content for communicating with customers.

While traditional AI simply analyzes data and makes predictions by following pre-programmed rules, generative AI, built upon deep neural networks, autonomously creates coherent and contextually relevant outputs.

Gen AI models, such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and large language models like GPT can create synthetic data that replicates the statistical properties of real-world datasets. This is particularly important when real data is scarce, expensive, or sensitive. A practical example could be generating transactional data for anti-fraud models.

Due to their capability to produce meaningful text that resembles human-written content, AI-driven chatbots allow banks to provide 24/7 support, removing customers’ need to wait in long lines or navigate difficult phone menus. These smart virtual assistants act autonomously and can provide customers with on-the-spot guidance and prompt support by responding to basic customer requests, such as:

• Recommending financial services and banking products

• Showing deposit options

• Checking account balances and transaction history

• Initiating and completing transactions

Along with direct use by customers, generative AI-based chatbots can also significantly assist the front line by suggesting client-specific actions. By examining real-time customer interactions and transactions, Gen AI can offer insightful data on specific customer behaviors and preferences, which can then be used by financial advisors to offer more personalized user experiences.

Generative AI algorithms analyze vast customer data, including transaction history, account balances, spending patterns, investment portfolios, and financial goals, to build a comprehensive customer profile. This allows banks to enhance their service operations by offering hyper-personalized recommendations based on their specific circumstances, creating customized financial plans, and providing tailored financial advice and product suggestions.

For example, Morgan Stanley has launched an AI assistant based on OpenAI’s GPT-4 that allows its 16,000 financial advisors instant access to a database of about 100,000 research reports and documents. The AI model aims to help financial advisors quickly find and synthesize answers to investing and finance queries and offer highly personalized instant insights.

There is a rich potential for generative AI tools to considerably assist in strategic decision-making. For one, generative AI can analyze market trends, financial market data, economic indicators, and investment opportunities to generate personalized investment recommendations. Furthermore, it can synthesize and test different market scenarios to propose and evaluate the effectiveness of new trading strategies, thereby helping banks identify profitable opportunities and minimize losses.

While there’s been increasing interest in applying generative AI across these functions, banks are still exploring how generative AI could be used for generating market and investment strategies. According to Jason Napier, Head of European Banks Research at UBS, “While later there will be other, probably more important, deployments, a lot of the potential of AI appears really nascent at this stage.”

With generative AI on board, banks are well-equipped to enhance their fraud detection capabilities and improve risk assessment. Here is how Gen AI can help.

• With little help from human users, generative AI can promptly identify transaction anomalies indicating fraudulent activity, such as odd locations and devices or unusual spending patterns, and automatically flag potential hazards.

• What’s more, Gen AI techniques, such as GANs, can create synthetic fraudulent transactions to provide a more diverse set of scenarios for training fraud detection models. This can prove critical in improving the robustness and accuracy of fraud detection algorithms.

• Gen AI algorithms can provide insights into the underlying patterns contributing to fraud alerts, which enables more effective decision-making.

• Generative AI models can help banks identify possible risk areas and preserve profitability by analyzing historical data patterns and market trends. By simulating different economic scenarios, GANs can help banks assess and mitigate risks, such as credit risk, market risk, and operational risk.

Just as banks and financial institutions need to verify the identity of their clients to avoid commercial relations with synthetic businesses or people related, for example, to fraud, corruption, or money laundering, it is equally critical for them to achieve requirements set by multiple KYC compliance regulations, such as AML, GDPR, and eIDAS.

Banks can use generative AI technology to automate the time-consuming process of customer due diligence by analyzing large amounts of customer personal data. This may help cut down on customer onboarding time, reduce false alarms, and enhance the accuracy of risk assessment while also ensuring compliance with stringent AML and KYC regulations.

For example, Airwallex, a global payments company, has introduced a generative AI copilot that utilizes large language models to accelerate the company’s KYC assessment and onboarding processes. The implementation of the Gen AI tool has reportedly reduced false-positive alerts by 50% during the KYC due diligence phase and sped up the KYC onboarding process by 20%. Previously, the company used rules-based analytics and NLP to scrutinize new customers’ websites, which often resulted in numerous false-positive alerts.

Regulatory compliance is another area to benefit from generative AI in the banking industry. Banking is a highly regulated industry—regulatory requirements are subject to frequent changes and updates. Keeping up with increasingly frequent regulatory changes puts a strain on banking and finance staff, as it requires a vast amount of manual and repetitive work to interpret new requirements and ensure alignment with regulatory standards.

Generative AI can help banks maintain ongoing compliance with ever-evolving regulatory requirements by continuously monitoring regulations in real-time, identifying compliance risks, and generating accurate and timely reports.

For example, Gen AI has been recently employed by Citigroup to evaluate the effects of new US capital regulations. The bank’s risk and compliance team used generative AI to sift through and summarize 1,089 pages of new capital rules released by the federal regulators. Moreover, they are looking to use large language models to parse legislation and regulations in the countries they operate in to ensure they abide by regulations in each jurisdiction.

One of the highest-value generative AI use cases in banking revolves around automating tedious activities that previously required human input. As an MIT Technology Review Insights report says, “Banks and insurance are among the industries with the greatest proportion of their workforces exposed to potential automation.” Since personnel expenses account for a sizable amount of total costs, the introduction of Gen AI automation into banking operations has the potential to substantially reduce operational costs. These cost savings mostly result from using Gen AI technology to take away the need for analyzing vast volumes of frequently unstructured data.

Owing to its enhanced ability to understand context and generate natural language texts, summarization capabilities, and predictive intelligence, generative AI holds promise to automate and streamline most of the back-office processes for greater operational efficiency. This will enable operations staff to focus on customers rather than crunching numbers. Accenture forecasts that by 2028, the banking industry will witness a 30% increase in employee productivity. Here is how generative AI can augment the back-office workforce:

• Accelerate report generation using gen AI tools to search and summarize volumes of data in unstructured documents

• Compress lengthy documents into summaries

• Speed up complicated post-trade processes in post-trade operations

• Process loan applications by analyzing various data points, including the applicant’s credit score, financial history, and current data

• Create summaries following business interactions or phone conversations

The potential benefits and opportunities generative AI offers to the financial sector are undeniable. However, the adoption of generative AI also raises data privacy and security concerns, which are major for the banking sector.

• First, there is always a risk of unintentional violation of customers’ privacy rights when collecting publicly available client data for profiling and forecasting. Gen AI can inadvertently reveal sensitive or personally identifiable information, such as personal identification details, transaction history, and account balances.

• Just as generative AI in banking continues to evolve, so do fraudsters, seeking new ways to exploit new technology for scaling their scams. For example, scammers might use Gen AI to create phishing and SMiShing attacks, fake browser extensions, or impersonation scams.

• Finally, the nature of generative AI is still largely unregulated. This poses a significant barrier to the large-scale adoption of generative AI in the banking industry. As the chief executive of the UK’s Financial Conduct Authority (FCA) said, “While the FCA does not regulate technology, we do regulate the effect on— and use of—tech in financial services…. With these developments [the growing use of generative AI], it is critical we do not lose sight of our duty to protect the most vulnerable and to safeguard financial inclusion and access.” While full regulation of AI by the government is under consideration so far, the potential value of an extensive application of generative AI should be balanced against regulatory risks.

Legacy technology is another factor slowing down Gen AI’s commercial use. Such systems impede the integration of innovative capabilities that novel technologies deliver. First, they often use outdated data formats, structures, and protocols that may be incompatible with modern Gen AI technologies. Secondly, they may store data in siloed or proprietary formats, making it difficult to access and retrieve data for generative AI model training and analysis.

Interestingly, generative AI itself can serve as a solution to the legacy infrastructure problem by propelling the transition from legacy software and data storage, which previously seemed unreasonable or cost-prohibitive. Gen AI’s ability to generate code can further assist with the transformation.

Legacy modernization is a daunting challenge—it involves a lot of time, financial resources, and effort. A trusted financial software development company that knows the ropes can help smoothly transform the existing infrastructure while also providing end-to-end support in building a powerful Gen AI solution.

Among the biggest concerns for the banking sector is Gen AI’s propensity for biases and unfairness.

Key points to consider:

• The resulting outputs can be biased if the data used to train a Gen AI model is incomplete or insufficient. Algorithmic bias may lead to unfair and discriminating lending decisions for certain population groups.

• Since Gen AI models are complex and sophisticated, bank employees can have a hard time interpreting the output of AI algorithms, which leads to the inability to explain the reason behind a model’s decision to customers or regulators.

• Generative AI models tend to produce confident wrong answers, referred to as “hallucinations.” While looking hyper-realistic, these outcomes are entirely fictitious in fact, which is catastrophic if applied in banking.

The talent shortage is another barrier standing in the way of generative AI adoption in the banking sector. According to John Mileham, CTO at Betterment, “Currently, generative AI is so new that you can’t really hire a whole lot of experience.”

Integrating Gen AI into banking operations will certainly reshape many roles in the banking workforce in that workers will have to learn new skills or change occupations.

To bridge the skills gap, financial services firms will have to figure out what new competencies and skills the workforce will have to acquire and whether they need to reskill and upskill existing employees or hire new ones. This will require extensive investments in retraining and hiring initiatives to meet changing talent needs. Providing internal training programs for employees is key to generating excitement and equipping your existing teams with the resources, skills, and capabilities required for the new roles, such as prompt engineering or model fine-tuning skills.

Specify priority areas (functions or units) to experience the biggest impact from generative AI technology and plan for specific use cases (frontline copilot, customer operations, or discovery of regulatory changes, to name a few)

Clearly define the objective and outcomes

Examine the interoperability of your current data infrastructure with generative AI tools, assess skills, and evaluate data and technology