AI costs spiral

One clear theme has emerged from the Q2 bank earnings season: AI costs are rising fast. And cost savings or revenue upside from the investment are still elusive, at least from the results the banks are putting out.

Analysts pulled on that thread in their questions, forcing bank leaders to justify the strategic pivot to this technology. Here are five highlights:

• Bank of America expects to spend $4 billion on tech initiatives like AI in 2024, a 5% increase on last year. AI had “moved from cost savings ideas to enhancing the quality of our customer interactions,” CEO Brian Moynihan said.
  
• Asked how Goldman Sachs would boost productivity with AI, CEO David Solomon listed AI assistants for analysts and coding tools. (See “How Goldman Drives AI Conversation”, The Brief, June 27 and “Solomonic conversion”, The Brief, April 18).
  
• Citi’s tech spend will rise to $3 billion, said CFO Mark Mason. Last year, the bank’s overall expenses rose 6% year on year, in large part driven by technology costs.
  
• Nordea CEO Frank Vang-Jensen, who hadn’t talked about AI in earnings calls before, said that tech and AI spend had “increased significantly to a high level.” The bank is staying on course and expects the investment to deliver long-term, he said.
  
• On JPMorgan’s call Wells Fargo analyst Mike Mayo offered a probing question: “if you're spending $17 billion a year to improve the company, if you're gaining share with digital banking, if you're automating the back office, if you're moving ahead with AI… why wouldn't those returns go higher over time? Or do you just assume you'll be competing those benefits away?” CFO Jeremy Barnum responded “it is a very, very, very competitive market. And we are very happy with our performance.”

The upshot: Twenty-two of the Evident AI Index 50 banks have held Q2 calls, and AI was mentioned in 15 of them. That’s a higher share than we’ve seen before. Clearly expectations for returns from the market are growing in line with costs.

AI’s biggest brains are gathering this week in Vienna for the International Conference on Machine Learning (ICML). The financial services industry is there too. HSBC, Morgan Stanley and Capital One are among the sponsors.

As AI goes mainstream, industry is getting more involved on the research side of the emerging technology. Nine of the 50 Evident AI Index banks had papers accepted at the top 20 AI conferences last year. And six banks have gone one step further, sponsoring those conferences this year, up from four in 2023.

Why lean in on research? As we saw from the quarterly calls above, leading banks have made a strategic and expensive commitment to AI adoption. They have to be part of the innovation ecosystem to win. The top conferences are venues to present and test out ideas – and as importantly, attract and retain the talent you need to come up with them.

Leaderboard: Here’s our most recent ranking of the top banks on research.

The full Dispatch: AI Research is available to Evident members. Non-members can download an excerpt. To learn more about membership, please contact us.

"I actually think all the companies that are investing are making a rational decision because the downside of being behind is that you’re out of position for, like, the most important technology for the next 10 to 15 years."

- Meta founder Mark Zuckerberg, interviewed for Bloomberg Originals, 23 July

At the Evident AI Roundtable last week – in which our newly published 2024 AI Talent Report was discussed – EY’s Katherine Savage said that the leading banks have moved away from focusing on just filling priority job roles to building an organization that develops talent for the longer haul.

How have successful organizations pulled off this shift? By assessing their current AI capabilities regularly and measuring the impact of upskilling and reskilling workers, she said. “Many of my clients are very good at resource planning, looking at the 12-month cycle… But when you try to enable line managers to articulate demand over the next three to five years, they really struggle.”

The new focus will pay off in attracting and retaining the best people, Société Générale’s Noémie Ellezam told listeners. Banks need to create the “environment to empower” talent, she said. Her advice: You need to show that roles will have an impact, provide a good platform to advance research with strong resources and partnerships, and commit to continuous learning.

Watch the full roundtable here.

OPENINGS… AND A MOVE

HSBC’s Edward Achtner, who had led the bank’s office of applied AI for the past year, has taken on the newly established role of group head of generative AI capabilities.

Morgan Stanley is hiring an executive director in generative AI solutions & strategy.

JPMC is looking for a VP for to be lead engineer for large language models.

Capital One has an opening for a senior manager to cover generative AI risk management.

Banks have a lot of information about how they’ve served clients and customers. This week’s Use Case Corner looks at how they’re tapping into those storehouses with generative AI to help answer customer queries.

#1 Generative AI virtual assistant

Use Case: Customer Service Officer (CSO) Assistant
Vendor: n/a
Bank: DBS

What it is: A generative AI-powered virtual assistant that transcribes customer queries live, providing answers to customer support staff based on the bank’s sets of data and information.

Why it’s interesting: It is one of over 20 generative AI use cases the bank is implementing, works in local languages and was developed in-house.

Potential ROI → increased customer satisfaction, increased employee efficiency, increased accuracy
Reported ROI → reduced call handling time by up to 20%, transcription and solutioning accuracy of “nearly 100%”

#2 Generative AI knowledge management

Use Case: Generative AI Knowledge Support
Vendor: n/a
Bank: CIBC

What it is: A question and answer tool, now moving from pilot to production, that helps employees find the information they need in the company’s repository of information to help them address client concerns.

Why it’s interesting: The pilot demonstrated it’s easier and faster for team members to access information.

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

Have feedback on or ideas about use cases? Let us know at [email protected]. The next Brief will be entirely devoted to use cases.

TD Bank tapped Canadian AI startup Cohere for access to their mid-size LLMs. Maksims Volkovs, TD’s chief AI scientist, highlighted the cost benefits of using smaller models (see “Bigger AI’nt Better”, The Brief, July 11). Relatedly, OpenAI’s new GPT-4o mini model is 30x cheaper than their current flagship.

For Intesa Sanpaolo AI integration has led to a 80x reduction in fraud since 2020, and the total savings from AI could total 500 million euros. But Intesa CEO Charles Messina reassuringly said that “no-one will lose their job” as “new paths will be created within the bank.”

“Will AI eliminate jobs? Yeah,” said JPMC CEO Jamie Dimon, taking a different tack from Messina in a LinkedIn interview. At the same time, he sees his AI-specific employee numbers rising from 2,000 to 5,000 in the next few years. Net outcome? Dimon said he “doesn’t know,” but stressed that AI has fundamentally reshaped “everything we do.”

AI banking risk is back on the agenda in Washington. A bipartisan report last week from the House of Representatives’ Committee on Financial Services looked at the impact of AI on the financial industry, from job loss to the potential for discrimination, and called for regulations. Meanwhile Eric Gerding, director of the SEC’s Division of Corporation Finance, warned last month in his “State of Disclosure” that AI implementation carries “additional operational and regulatory risks.”

Groupe BPCE launched a new strategy featuring clear timelines on AI ROI. The French lender expects to see priority AI investments recoup their cost within three years. They also expect half of staff to use AI by 2026.

The EU AI Act is going into force August 2nd. Here are answers to six questions that banks should be asking (with an assist from Minesh Tanna, partner and global AI lead at Simmons & Simmons).

What is it and what does it do?

It’s focused on AI product safety, similar to how medical devices or machinery are regulated by the EU. The AI Act breaks down products into prohibited, high-risk, and low-risk categories. If a bank creates a “high-risk product,” it has to abide by a set of requirements on things like technical documentation, data and human rights impact assessments, and transparency to end users. Developers are the main target of the rules, but they also apply to others like deployers and distributors.

OK, so which bits are most relevant to the financial industry?

Banks mainly need to focus on one thing: what to do about products that Annex III deems “high risk.” Three key ones for the financial sector are those that assess creditworthiness, assist in recruitment, and provide insurance quotes. For any bank that uses AI to handle any of these, they’ll have to give technical documentation, demonstrate human oversight and be transparent with customers. Businesses will have to make their workforces more AI literate to comply – “a potentially challenging exercise for, say, a large international bank,” Tanna said.

How can banks best prepare?

First, keep a good inventory of all your AI products and decide which fall into the high-risk category. Then for each product you will want to assess where there are differences between existing compliance measures (e.g. for model governance) and the new obligations under the Act – a task that could perhaps be one of the biggest challenges for banks.

Second, because the Act is applicable across the entire AI lifecycle – through design, development and deployment – compliance is everyone’s job. Banks have to “collaborate across tech folks, data privacy staff and legal,” Tanna said.

And third, you’re not just responsible for what you create but what you procure from outside vendors. Even when implementing a low risk use case using ChatGPT or Gemini for example, you should think about due diligence. “It’s not just how you’re doing your bit, but how you are ensuring that everyone in your supply chain and ecosystem is doing their bit,” Tanna said.

How costly or timely is complying with the EU AI Act going to be?

As with the GDPR, it’s likely that banks will bring in legal and compliance people dedicated to AI regulation.

But banks get off relatively lightly compared to other sectors because existing financial rules for model risk management mean they already have many of the requirements of the Act (e.g. data governance, model documentation and human oversight) in place.

I don’t work in the EU. How does it apply to me?

The EU says that any bank anywhere that deploys an AI system within its jurisdiction has to comply.

Will the EU AI Act become the global standard?

The EU hopes so – to do with AI what it did with antitrust and privacy. But there’s a danger. By taking the most hands-on approach to regulation of AI with this legislation, the bloc risks stifling AI innovation in Europe. Some Big Tech companies have decided to restrict access to their models in the region, citing regulatory burdens. (See “Big Tech to Europe: Up Yours,” The Brief, June 27).

No banks headquartered in the EU made it into the top 10 for Transparency, which focuses on responsible AI adoption, in the last update of the Evident AI Index. As we launch the next Index in October, we’ll be looking to see whether the AI Act prompts them to do better.

Do you know or run an event that you think should be featured? Let us know at [email protected].

We misspelled Kristen Bennie’s first name in “This is AI London,” The Brief, July 11. Our apologies.