Who's poaching whom

Two weeks into Q2 bank earnings season, it’s never been more apparent that banks are struggling to calculate returns on AI.

It hasn’t stopped them from talking it up. Every bank’s call so far has had at least some discussion of AI. The total number of mentions across those 21 banks has nearly tripled compared to Q2 last year.

But instead of pointing to specific AI returns, bank leaders are steering investors hungry for investment justifications to the most tangible ROI metric they can: a smaller payroll.

At Bank of America, CEO Brian Moynihan said halving the consumer business’ headcount and doubling revenue in the last 15 years has been “enabled by application of technology on scale.” Wells Fargo’s CEO Charles W. Scharf said headcount at the bank has come down and “technology will be able to help that trajectory continue even more.” PNC CEO Bill Demchak responded to a question about AI by saying the bank would be able to “keep expenses on the low end as it relates to people, even as we invest in the technology.”

If last year was about leaders pledging themselves to AI spend (see: “Hey big (AI) spender,” The Brief, July 2024), this year sounds like a renewal of vows. Nordea CEO Frank Vang-Jensen justified rising costs by pointing to “significant investments” in AI. Citizens Financial’s president Brendan Coughlin remained “bullish on long-term value creation from AI.” And BNY CEO Robin Vince said the bank views AI as “an expense story” while it uses the tech for “unlocking capacity.”

Bottom Line: Banks are buying time on investor ROI questions until AI’s value is more fully realized. In the meantime, it looks like they’ll take a page from Big Tech’s playbook and show they can do more with less.

"The big banks may be spending a lot of money and doing some kind of pioneer work, though I think there'll be ready-made turnkey solutions available that, hopefully, we can take advantage of."

- Bruce Van Saun, CEO of Citizens Financial

Banks are most likely to find new people to build their AI models in university lecture halls, new Evident analysis shows. That doesn’t stop them from wanting practical banking experience.

RBC this week announced a finance-specific research initiative with MIT’s Computer and AI Laboratory (CSAIL) that will offer the bank “increased recruitment opportunities.” Capital One last month expanded a generative AI partnership with the University of Illinois Urbana-Champaign. CommBank sponsored an AI program focused on finance with the University of Adelaide late last year with the aim of being “a magnet for top-tier global AI talent.”

Good old-fashioned poaching happens, too. Banks were three times more likely to hire this top AI talent from other banks than from consultancies, tech firms or other industries. Goldman Sachs, JPMorganChase and Wells Fargo are chief among the Wall Street raiders, driving more than 40% of the hires in this revolving door.

Bottom line: Banks are filling their most advanced AI roles with academics – but they want more who can also think like bankers.

Banks have debuted new agentic tools in recent weeks. In our Use Case Corner today, one third-party solution has become a favorite and an in-house tool recently got an agentic tune-up.

#1 CODER COLLEAGUES

Use Case: Agentic AI for software engineering
Line of Business: IT & Security
Vendor: Cognition
Banks: Citi, Goldman Sachs

Why it’s interesting: Citi and Goldman Sachs both “hired” the same software engineer in recent weeks. No, not Soham Parekh; it was Devin, an agentic system developed by Cognition that reads, writes, tests and documents code on its own. Where a copilot might autocomplete code alongside an engineer, Devin can autonomously complete software development tasks without needing step-by-step prompting from a human employee.

How it works: Developers can assign Devin a high-level task – such as rewriting a software component – and the agentic tool will fetch relevant code, draft documentation and run tests and rewrite where necessary before presenting a finished package for human review.

How it scales: At Citi, the focus is training Devin on its house style to ensure consistency. “I don’t want creativity,” Citi CTO David Griffiths said last week. “I want it done the same every time.” At Goldman, the focus is on getting engineers to be a part of a “hybrid workforce” that sees them overseeing agents. “Engineers are going to be expected to describe problems in a coherent way…and then be able to supervise the work of those agents,” Goldman Sachs CIO Marco Argenti said.

By the numbers: The firms report productivity gains differently. Citi estimates that using Devin for specific tasks can increase productivity between twofold and 20-fold. Goldman Sachs says it will use hundreds if not thousands of Devins and can boost productivity by a factor of three or four. At both firms, humans remain in charge: Devin can’t execute code, only prepare it for human staff.

#2 SPECIAL DELIVERY

Use Case: Autonomous triage of client email requests
Line of Business: Commercial Banking
Vendor: In-house (AIDox)
Bank: Scotiabank

Why it’s interesting: Scotiabank’s Commercial Banking team was getting over 1,500 emails a day – more than 60,000 per month – ranging from routine requests to complex legal questions. Each message had to be manually opened, interpreted, routed to the right team, and logged by an employee. The whole process could take up to two hours.

How it works: The bank first built AIDox in 2018 to verify documents in bank processes, like mortgage applications. It began training the system on commercial banking requests in 2023 and this year gave the tool the autonomy to read, interpret and route emails to the right person within the bank and open a case file as well. When the system can’t recognize where to route a request, it kicks it up to a human “SWAT team” for review.

How it scales: The bank says routing capabilities are among the hardest agentic tasks for the bank to start with. “Because of that, we have had a lot of success now and we’re well-positioned to add other use cases across the Bank,” said Beth Einwechter, director of strategy & operations at Scotiabank.

By the numbers: AIDox now handles about 90% of all incoming Commercial Banking emails on its own, and the time it takes to get an email into the right inbox dropped from hours to minutes, the bank said. Automation also let Scotiabank redeploy 70% of the email triage team to handle other tasks.

The White House on Wednesday shared “America’s AI Action Plan,” a 24-page homage to deregulation which criticized “limited and slow adoption of AI, particularly within large, established organizations.” Among the recommended policy actions made by Michael Kratsios, assistant to the president for science and technology, David Sacks, AI and crypto special advisor, and Secretary of State Marco Rubio was having the government foot the bill for AI training via a tax-free reimbursement program – surely a welcome idea for the 37 banks in the Evident AI Index that already offer some kind of AI training. The plan also recommends a regulatory sandbox for the SEC that would allow financial firms to test the technology more broadly, not unlike the Singaporean approach that’s sped up AI advancement.

OpenAI’s new ChatGPT Agent might be as good as your new summer analyst. The AI lab assessed its new agentic system on how well it could complete investment banking tasks like creating a three-statement model or modeling an LBO from start to finish and found it could do it right 41% of the time – an improvement over its o3 model and Deep Research tool.

Anthropic launched a “Claude for Financial Services,” a financial analysis solution the lab says can “modernize trading systems, develop proprietary models, automate compliance, and run complex analyses including Monte Carlo simulations and risk modeling.” The system can complete “complex Excel tasks” with 83% accuracy, Anthropic said. Early adopters include CommBank and AIG.

DBS wants bankers to use generative AI to plan their next career moves. The Singaporean bank’s new tool called iCoach was built in partnership with executive leadership coach and author Marshall Goldsmith and is available to the whole staff, the bank said.

Lloyds unveiled Athena, its new “generative AI-powered knowledge hub” for customer-facing employees. The tool scans 13,000 documents after being prompted and has cut the time it takes for employees to find information by 66%. So far it’s handled 2.1 million questions, but the bank says that could rise to 40 million by year’s end.

JPMorganChase’s quantum heavyweights Marco Pistoia and Charles Lim – who led the bank to breakthroughs like using the technology to generate truly random numbers – have left the bank. Under their leadership, the bank had published more than half of all the quantum-related papers produced by the 50 banks we track and accounted for two-thirds of all quantum job postings by that same group (see: “Banks’ Quantum Solace,” The Brief, March 20). Pistoia didn’t clarify what was next in a LinkedIn post announcing his departure Wednesday. Lim couldn't be reached for comment, but the bank confirmed his departure. Replacing the duo at the bank is State Street’s global head of digital technology and quantum computing Rob Otter, who will head JPMC’s global technology group.

Freddy Lecue joined Wells Fargo as head of frontier AI models and methodology. Lecue was an AI research director at JPMorganChase.

BNY hired Nastassja Hagan as COO for Global Engineering, where she’ll focus on AI and transformation. Previously, she worked in senior roles focused on transformation and sustainability in the energy industry, including at BP and ExxonMobil.

German data process mining and AI company Celonis brought onboard former Deutsche Bank India CEO Dilipkumar Khandelwal as the group’s chief customer officer and chairman of its India Advisory Board. At Deutsche Bank, Khandelwal oversaw the bank’s global technology centers from Bengaluru and Pune to Bucharest, Berlin, and Cary.

Sam Altman used his audience this week with many of the U.S.’s most important banking regulators and executives to issue a stern warning: A “significant impending fraud crisis” is coming. Speaking at the Fed’s banking conference Tuesday with Michelle Bowman, the “terrified” Altman described how AI had “fully defeated” standard authentication methods banks use, like selfies, waves and voiceprints.

If it was a play for their business, it may be too late.

As we pointed out earlier this month, banks aren’t sitting on their hands when it comes to fraud (see: “Caught red handed,” The Brief, July 10). Fraud use cases from the 50 banks we track were up 80% in the last year alone and banks are investing a lot here.

OpenAI is one of many vendors looking to get in on the action. Chief rival Anthropic last week launched Claude for Financial Services, its foray into providing AI solutions for banks. A lot of the tools are getting developed by the banks themselves, who aren’t as consumed by the noise around the big model makers in Silicon Valley as the tech media is.

Altman didn’t sound deterred. “I promise I didn’t come here for the TV commercial,” he said. “But we would love to work with any of you.”

Mon 11 - Weds 13 August
AI4, Las Vegas

Mon 8 - Weds 10 Sept
Finovate Fall, New York

Tues 9 - Weds 10 Sept
Artificial Intelligence in Financial Services Conference, London