Banks cautiously embrace AI for internal applications

Some of North America’s and Europe’s biggest banks are creating revenue and whittling down costs using AI tools that can advise sales teams, automate credit decisions, and generate code. Yet they haven’t widely put the technology in front of clients, and most generative AI deployments are still at least a year away.

JP Morgan Chase (JPMC), HSBC, Deutsche Bank, and Royal Bank of Canada (RBC) are among those training pattern-spotting, process-automating AI software to help manage back-office functions, including rooting out credit card fraud, green-lighting lending, guiding client teams, and writing computer code, executives said at the Evident AI Symposium in New York in November 2023.

JPMC, which topped London-based researcher Evident’s latest ranking of global banks’ AI maturity, is using AI to personalize offers to credit card customers, make recommendations for sales teams assigned to corporate clients, and curtail fraud. JPMC isn’t realizing revenue yet from generative AI, the field which uses large language models trained on vast amounts of internet and private data to compose text, summarize documents, and plan investing strategies. Deutsche Bank deploys AI to fight money laundering, but it’s holding back on deploying the technology in software used by clients.

Generative AI apps that could answer clients’ investment research queries, summarize companies’ earnings, or prep detailed customer meeting briefings aren’t in production and won’t be until about 2025, bank and stock exchange executives said at the conference.

“Low-hanging fruit is not as ripe as I think people would want,” said Teresa Heitsenrether, chief data and analytics officer at JPMC. The bank will realize more than $1.5 billion in value this year from running AI tools to streamline operations, detect fraud, screen transactions for potential sanctions violations, and render credit decisions, Heitsenrether said. Revenue is coming from better card offers or suggested next steps for sales teams. “We have vast amounts of data and we have the ability to invest,” she said.

Heitsenrether, a JPMC veteran tapped this year to head AI adoption companywide, said “all of that value to date for the firm has been delivered through business intelligence tools and more traditional AI methods like machine learning” that can find patterns and make predictions, not yet from generative AI. “We definitely see tremendous potential there,” she said, citing JPMC’s hundreds of use cases in progress.

Canada’s RBC, third in the Evident ranking, is also taking a cautious approach to putting generative technology in front of corporate customers. “We don’t expect to see a lot of banking clients interacting with a chatbot to get financial advice in 2024,” Foteini Agrafioti, RBC’s chief science officer and head of its Borealis AI incubator, said at the conference.

RBC doesn’t think generative AI is “ready for prime time,” client-facing apps, Agrafioti said. Instead, it’s building test beds to determine whether large language models can let research analysts and associates compile reports quickly, or if the models can cut call center costs.

Huge market seen

Banks are piloting generative AI systems to analyze reams of documents and other unstructured information that doesn’t reside in databases for applications such as summarizing capital markets research, proofing clients’ investment portfolios for risk and rebalancing, and researching customers.

The technology could create an additional $200 billion to $340 billion in annual value across the sector, consultancy McKinsey & Company estimates, if banks maximize its use for regulatory compliance, customer service, coding, and risk management. Yet many banks have been hesitant to roll out generative AI in production, and commercial projects will probably debut in 2024, with an impact on profit in 2025, according to Evident CEO Alexandra Mousavizadeh.

Among the generative AI projects banks and exchanges have on tap: improving their enterprise search capabilities, compiling briefing books for meetings between senior banking execs and their clients, replacing manually compiled reports in spreadsheets and business intelligence dashboards, and letting clients query bodies of capital markets research. “We’re going to have to be able to use machines in that manner to compete,” said George Lee, co-head of Goldman Sachs’ applied innovation office.

One gating factor: Banks still need to see which generative AI use cases clear regulatory hurdles first, said Stefan Simon, a member of the management board and head of Americas at Deutsche Bank. “A lot of banks are not very hungry to be the first mover,” he said. “The regulatory landscape adds a unique angle to that competition.”

Competing for AI talent

Evident publishes a twice-yearly index that ranks 50 of the largest North American, European, and Asian banks by their AI capabilities using four criteria: top-down leadership, talent, innovation, and transparency. JP Morgan Chase topped the November index—which measures more than 100 criteria, including research, patents, talent retention, ventures, and partnerships—followed by Capital One, RBC, Wells Fargo, and UBS.

Banks are hiring droves of data scientists, engineers, software developers, and other AI experts, even as they’re cutting back in other areas. The institutions ranked in the Evident study increased their number of AI positions by 10% between May and September 2023, while they cut overall headcount by 2.5%.

Among executives in data, analytics, and AI, pay is rising. Median compensation (PDF) for these execs, including equity grants, was $901,000 in the United States and $676,000 in Europe in 2021, according to recruiting firm Heidrick & Struggles. European AI and data analysis execs in financial services took home a median $961,000, topping all fields.

Eye on costs

Still, banks are keeping an eye on costs. Generative AI models are extremely expensive to train and calibrate. Most banks are turning to commercial large language models running in public clouds rather than building and training models themselves.

“First of all, you’re not going to build these things yourself, at least not this year,” said Jeff McMillan, chief analytics and data officer at Morgan Stanley. “You can work with any of the major providers, and as long as you get your legal, compliance, and risk stuff together, you can change the world with what’s in place right now.”

Oracle, Meta, IBM, and other tech companies, as well as universities and research organizations, launched the AI Alliance in December 2023 to create software tools, model explanations, benchmarks, and standards that organizations can share, giving businesses an alternative to buying AI models whose workings may be more closed to organizations outside of the tech industry.

Banks understand the risk of concentrating too much business in the hands of a single generative AI vendor, McMillan and JPMC’s Heitsenrether said. “The name of the game is diversification,” she said. “We’re a multi-cloud shop and we’ll be a multi-model shop.”

View more Oracle Connect articles