When will AI pay off?

As CEOs focused on the benefits of AI this earnings season, we asked you in the last Brief: “When do you expect significant return on AI investment in banks?”

The largest number of you (42%) said by 2026. A quarter picked next year. Almost 20% said not before 2050 or never. So our readers are ever so skeptically optimistic. That tracks with what some CEO’s are saying out loud. AI’s “a ’26 and out benefit,” per Robin Vince in BNY Mellon’s earnings call last month. Two-thirds of C-suite execs across industries believe it’ll take at least two years to “move beyond the hype” of AI, according to a BCG survey.  

This early in a technological wave, the biggest known unknown is what “significant return” will precisely look like and when it’ll come. No wonder most people are hedging. But we do have enough known knowns in hand to venture some clear conclusions about where we are in the journey.

The first known AI known: AI is paying dividends for banks. They’re just not saying so loudly.

This is new. A year ago, most executives we spoke with didn’t have good oversight on their use cases—much less how AI might be helping the bottom line. Many didn't have a common definition of what counted as an "AI use case" or a way to calculate return on investment.

Fast forward to this earnings season. One executive told us that their bank came up with more than two dozen instances of a “return” on their investment in AI, only for the board to take it out of the results release. Individual companies are reluctant to be the first to put numbers on productivity gains or cost savings. Why give analysts a target to hold them to account for in future quarters? That’s happening at many banks.

There’s a downside to the coyness. As long as there’s little evidence of an AI productivity surge or other financial gains to back up promises made in various places (see these recent studies from Accenture, PWC and EY), the public and not least regulators will remain dubious.

A senior financial regulator in the U.S. griped to us this week that banks are touting “big bets on AI” without providing details on returns to please Wall Street analysts. For many outside the sector this is still mostly “window dressing”. Substance would make it feel real.

Second known known: AI use cases are developed and get rolled out to staff faster than to bank customers.

This approach will initially bring cost savings more than new revenue. A reasonable trade off. A bank might live with a not wholly perfect, as in 97% reliable, tool to help staff do their jobs better. They have no room for error from when it comes to dealing with clients’ money.

That’s also why banks won’t and can’t be truly in the vanguard of AI adoption. As another official in Washington noted, "scammers and fraudsters are the ones doing best on AI"—which is one of the reasons that banks moved to upgrade their tech stacks in the first place.

Asterisk here: The number of use cases isn’t as important as choosing the few that will pay off for them. Beware the use case spiral.

The last known known: Banks leading on AI are pulling away from the pack.

Early AI adopters changed operating models, built out strong research teams and moved fast to implement use cases. The gap is widening, as the Evident AI Index shows. The concentration of power roughly mirrors what you see in the tech sector, where the old giants (Microsoft, Google, Meta, Amazon) can invest heavily in AI and stay preeminent in a new era.

The problem: This undermines the often heard promise that AI is a democratizing technology. And it invites political and hence regulatory backlash to the concentration of power in industry. Banks are feeling it already, if not as acutely as Big Tech.

So will we see significant return on AI investment? Based on what’s happening in the banking industry, this is a matter of when not if. There’s an old saying from previous technological transformations: Everyone overestimates the impact of it a year from now and underestimates it over 10 years. We expect incremental progress, then a sudden spike up. Just don’t ask us to bet on when.

"To borrow from the electronic music duo Daft Punk, AI/ML models may be harder, faster and stronger than existing ones, but are they necessarily better for financial stability?"

- Pablo Hernández de Cos, Chair of the Basel Committee on Banking Supervision and Governor of the Bank of Spain

Next week, we’re launching the latest monthly issue of The Dispatch where we’ll unpack the latest findings from the Talent pillar of the Evident AI Index.

In an operating environment where a quarter of banks are seeing flat or reduced levels of overall headcount, the AI talent pool continues to expand. AI staffing levels are up +8% across the 50 Index banks—exceeding changes in overall headcount by nearly 2x.

Now look under the hood: Banks are prioritizing data engineers and implementers.

What does that tell you? AI is (slowly) moving from the lab to the front office. The focus is now clearly on figuring out how to implement AI and make it work for the bottom line.

We also found something else: More data engineers make for a happier workplace. When you look at how a greater “density” of data engineers (relative to a bank's overall headcount) translated to a bank’s current ratings on Glassdoor, provided by those employees working in AI-specific roles, the correlation is clear, as this chart shows.

That makes intuitive sense. Data engineers support every other technical role in a bank. Well-supported employees tend to be happy, productive employees. So a bank’s ability to hoover up data engineers could help them recruit for other jobs in the wider AI ecosystem.

We’ll be diving into this finding (and more) throughout upcoming reports on Talent Capability and Talent Development, exclusively available to Evident members. Join us. Find out more about membership here.

Join us at the Evident AI Symposium, where we will bring senior AI leaders from across the banking sector together to cut through the hype and drive forward a global conversation around the realities of AI adoption.

At our first European gathering, we’re zeroing in on the theme of: Accelerating Outcomes: How are banks delivering value from AI now?

Speakers include: David Schwimmer, CEO, LSEG; Clare Barclay, CEO, Microsoft UK; Manuela Veloso, Head of AI Research, JPMorgan Chase; and Sameer Gupta, Chief Analytics Officer, DBS.

NatWest’s CEO Paul Thwaite recently said the bank has identified more than 100 priority areas where AI can be used to “boost efficiency” and improve productivity. Send us details! We scour and reality check bank announcements about AI use cases. These two stood out in the past two weeks.

#1: IT: Workflow Automation

Bank: JPMorgan Chase
Use Case: FlowMind
Vendor: n/a

Why it’s interesting: JPMorgan Chase’s AI Research team addressed an ongoing concern about LLM adoption—reliability. FlowMind is an LLM-driven application that’s accurate 90-99% of the time depending on the complexity of the task. The paper was submitted to the 4th ACM International Conference on AI in Finance (ICAIF ’26). This is a step towards widespread adoption of LLMs, particularly in data sensitive industries.

Potential ROI → increased efficiency
Reported ROI → 90-99% accuracy, dependent on task difficulty

#2: Decision Support: Early Debt Recovery

Bank: BBVA
Use Case: Machine Learning Pipeline for Debt Recovery
Vendor: n/a

Why it’s interesting: BBVA was recognized last month for its work developing a machine learning pipeline for early debt recovery.

Potential ROI → increased decision efficiency, increased customer satisfaction
Reported ROI → n/a

HSBC CEO Noel Quinn, whose resignation on Tuesday surprised colleagues and markets, offers his successor and peers some parting lessons about banking and technology.

For much of his five-year run as CEO, Quinn had to focus on shifting the bank’s strategic direction, its operations in China and the Americas. More recently, Quinn had started to ramp up the bank’s focus on tech, and AI specifically. The result: HSBC was one of the most improved banks in the Evident AI Index in 2023, leading the way for the U.K. banks. AI had clearly become a CEO-level priority. Will the next CEO take on the mantle? Time will tell.

OTHER NOTABLE MOVES AND OPENINGS

Cristian Adamo, JP Morgan’s global head of engineering, architecture and data for applied AI, left this week to co-found Patagon AI.

JP Morgan is hiring for an Executive Director of Generative AI in London, as part of a recent push to beef out their Applied AI team in the city.

BNP Paribas hired Su Yang in Paris as head of AI for transaction banking. With Yang’s experience in scaling big data lakes, the hire positions BNP to double down on transaction use cases such as fraud detection, liquidity management, and forecasting.

Music made by AI has been around for a few years—you might want to avert ears sometimes—but a couple new services, Suno and Udio, have achieved significant breakthroughs. They let you write and “perform” lyrics from scratch and create an entire song, with proper arrangements and choruses.

Our own creation “Banking on the Future” was written and generated in about a minute from a 10 word prompt.

Suno and Udio want to get everyone to be a music creator. Why should only a few artists dictate what we listen to? One of Udio’s investors is will.i.am of the Black Eyed Peas, who’s heralding a new musical Renaissance. Don’t hold your breath. So far, users are creating novelties like a spaghetti showtune and "Dune the Musical". As some other text-prompted generative AI creative tools, these are fun to play around with but aren’t anytime soon going to replace professional musicians any more than ChatGPT will journalists or consultants

Yet the business applications aren’t farfetched—even now. Marketing departments take note: This will make it easier to create ads with music tailored for different demographics. Now you can just enter a prompt to put something together—or dozens of different tunes, depending on who it’s for—and not waste time hunting down and licensing music. (Red Lobster tried something like it). You’ll probably be hearing more AI tunes as store background or the proverbial elevator music. This cheaper, easier-to-source music won’t set the world on fire, but it’s another generative AI technology that looks bound to be adopted by business.