How Goldman drives AI conversation

Goldman Sachs, as most any Goldman partner will tell you, stands apart. That applies to its approach to AI as well.

It speaks often and by any qualitative measure smartly about the broader impacts of AI. The data backs it up: Goldman crops up in over 1,200 AI-related articles in the past 24 months in media tracked by us – 25 times more often than the average bank in the Evident AI Index (see chart below). It brings heavy hitters to TV screen, conference stage and editorial board rooms: CIO Marco Argenti (who wrote a viral piece in the Harvard Business Review on “Why Engineers Should Study Philosophy”), Office of Applied Innovation Co-Head George Lee (who spoke to Business Insider about the firm’s GenAI plans this morning) and his co-head and prolific writer on tech and geopolitics Jared Cohen. Goldman made the most eye-catching calls on the impact of GenAI’s on jobs and on Nvidia’s coming share price surge. That’s what driving the conversation looks like.

Goldman stands apart in another way too. It has – until recently – revealed relatively little about how it is using AI to transform its own business. CEO David Solomon mentioned AI for the first time ever in an earnings call and shareholder letter this year, and did so with a skeptical note. (See “Market Rewards Bank AI Leaders,” The Brief, April 19). Compare that to JPMorgan Chase’s Jamie Dimon, who has regularly hyped it up (‘as transformational as electricity’) for years. Unlike Goldman, JPMC and other top banks also talk in greater detail about their internal use cases.

In part, that kind of high-profile thought leadership burnishes Goldman’s reputation for smarts – and is often more attention-grabbing than talking about the plumbing. But the coyness may also have been deliberate. In an interview with the Wall Street Journal published this morning, Argenti said that the bank has prioritized GenAI infrastructure over applications. That decision “might have slowed us down initially,” he admitted, but said that it’s picking up speed.

Goldman places 4th and 8th in the Innovation and Talent pillars of our index respectively – an indication of both its formidable know-how and talented tech team. With its new platform in place, expect to hear more from them on the tech itself in coming months.

In Citi’s newly-published poll of 90 banks, insurers, and asset managers, 93% said AI would boost profits, with the report suggesting that it could boost global banking industry returns to $2 trillion by 2028. Citi also said AI would displace more jobs in banking than any other sector.

The Bank of International Settlements said the world's central banks needed to “raise their game,” calling AI a “game changer for many activities” that will have “a profound impact" on the economy and financial system.

Five questions for Morgan Stanley’s new firmwide AI chief Jeff McMillan. “Ultimately,” he said, “you’ll be able to use just your voice.”

AI can now outperform human financial analysts and will only get better, according to fintech investor and writer Lex Sokolin, who noted that speech and vision features should see AI agents pick up things that humans miss in live events like earnings calls.

AI can be a game changer for corporate treasurers, reported Deutsche Bank, as long as C-suites understand the potential return on investment from choosing certain tools.

"There is one final layer in the AI stack, which is humans. I think it's important for us to all embrace that AI will go nowhere without people being creative about what they can do with it."

- Manuela Veloso, head of AI research at JPMC, speaking at the Evident AI Symposium in London June 19

As banks move from concepts to honing in on the most promising use cases, the top priority for financial institutions now that emerged from conversations at last week’s Evident AI Symposium is how to reduce the time to production. Here are three practical ways that banks are using to do that:

#1 Generative AI streamlines production process

The team at NatWest uses generative AI to shorten the time it takes to develop a prototype that it can put in front of customers, their CDAO Zachery Anderson told us.

An example: If you want to know whether image classification works for a specific use case, you would previously spend weeks to build a model for that. Now the team is using LLMs to model it out much faster, Anderson said. Although a traditional ML model is cheaper and more accurate than generative AI at classifying images, he said, you find out the suitability of a model much quicker by testing with an LLM.

Why it matters? The goal is to get a proof of concept out as quickly as possible to see if there’s value, Anderson said: “You can test things and then say, ‘all right, this will work, the customers like it, now let's break it down and go build a traditional AI model in that space’.”

#2 A store of features makes work on AI use cases reusable

CommBank is creating a “feature store,” stocking up variables (or features) that have worked well in one AI setting for teams to redeploy in another. Features are key elements within an AI model that influence how useful that model is for a specific use case. While many banks have a central repository for tracking their AI use cases, a feature store provides access to a level deeper in the use case production process.

For example: Dan Jermyn, chief decision scientist at CommBank, asked us to imagine a credit risk team building a model for a lending use case with a feature that supports that use case – say the number of times you spent X on Y over the last three years. With a feature store, another team in a different part of the business may find that variable works with a model they are building, but may never have thought about using.

Why it matters? A central repository, or store, makes it easy to reuse features, said Jermyn, who added that teams working with models can get much better outcomes because they can apply features in ways they wouldn’t have previously thought of. Mastercard recently filed a patent along the same lines.

#3 Platform for easy testing and scaling of LLMs

Nvidia’s generative AI microservices platform, running on millions of Nvidia GPUs in different cloud settings, allows banks to create generative AI powered applications on their own hosting service (like Azure or Sagemaker) while retaining full control of their IP.

Why it matters? Users of microservices can access and test models, such as Meta’s Llama LLM, until they are happy with it for the use case they are working on, said Jochen Papenbrock, Nvidia’s head of financial technology in Europe, Middle East and Africa. This makes experimentation easier, letting you test models against each other to work out which is better, he said at the Symposium.

Join us for our 2024 Talent Roundtable: How can banks close the AI talent gap?

We'll dive into the latest data from Evident’s upcoming 2024 Talent Report to consider: What sorts of skills do banks need in order to prepare for – and thrive in – an AI-first future? How are they going about attracting, developing and nurturing this talent? And what role can external partners play in building capability in the short- and long-term?

We track AI job posts in over 140 cities, and in looking over the data from the past six months five jumped out at us. New York and London still lead but here is our list of the hottest new AI talent clusters, ranked by the growth in openings for AI developers.

Dallas/Fort Worth. It’s no secret the area is seeing an influx of Big Tech talent, but banks are getting involved too. Capital One is driving a lot of AI hiring in Texas’s third-largest city. Looking forwards, the recently announced Texas Stock Exchange will generate more Dallas tech-finance buzz.

Mexico City. It’s become a hub for fintech entrepreneurs trying to bank the country's unbanked population. And BBVA, the biggest financial player in Mexico, has been adding quickly to its AI developer ranks, now over 200, in the last six months – a sign that it wants to keep that market share.

Paris. The French capital is seeing faster jobs growth compared to its larger peers New York and London, where AI development recruiting is cooling off. Call it the Mistral effect: Parisian banks are looking to capitalize on the excitement around the French AI startup as we see total AI development jobs approach 2,000 in the last six months.

São Paulo. Like Mexico City, Brazil’s business capital is growing because the Evident AI Index banks with a big presence in the region, in particular Santander, are adding bulk here –  perhaps in response to the rise of the Brazilian neobanking pioneer Nubank, who just hit 100 million customers.

Utrecht. A university town with strong AI programs, Utrecht is an attractive place to hire from. ING’s AI research partnership with Utrecht University suggests it is looking to the city as an alternative AI talent hub to Amsterdam.

More on the latest hiring trends in Evident’s latest issue of The Dispatch. To learn more about membership, please get in touch with us.

BNY named Leigh Ann Russell CIO and global head of engineering. Russell comes from BP, which was one of the first companies to roll out a Microsoft AI Copilot. The hire reinforces the strategic importance of AI for BNY and CEO Robin Vince. (See “Vincene Conversion,” The Brief, May 30.)

CaixaBank’s technology subsidiary plans to fill 200 new roles focused on AI, cloud and data analysis, as part of push to bring 1000 technologists into the Spanish bank.

Prudential named Grace Park chief data & analytics officer. Park previously headed up analytics monetization and business planning at Standard Chartered.

Another American tech giant is AI-snubbing Europe – this time Apple, which announced last week that its new generative AI offering, Apple Intelligence, won’t be available in the EU before 2025, at the earliest.

Apple’s gripe is familiar: EU regulations. In this case, the gang from Cupertino says the “regulatory uncertainties” around the new Digital Markets Acts forced it to hold back new AI-enabled features in the iPhone and computers that will be rolled out elsewhere. The EU law forces Big Tech companies to share data with competitors and forbids them from favoring their own services.

Earlier this month, Meta paused the rollout of its AI models in Europe in response to concerns about privacy raised by the Irish data watchdog. Citing regulatory concerns, Google and Anthropic have limited access to their AI models in Europe. (See “World AIn’t Flat,” The Brief, April 18.)

Big Tech could be flexing some political muscle here. Apple, Meta and Google are in open war with various EU regulators. Days after Apple said its AI products won’t be rolled out in Europe, the EU unveiled the first-ever charges under the Digital Markets Act, accusing Apple of abusing its dominant position through its App Store. So if you’re a Valley giant, why not a shot across the bow to remind Europeans there’s a price to pay for the choices made by their politicians.

What’s the practical importance of Apple’s snub for business, including banks? No big ones directly. But all these recent decisions by the American companies help create an environment in the EU where people don’t have the same access to the most advanced AI technology as those in other places. And that’s probably not a competitive advantage for Europe, no matter how much EU politicians like to boast about being out front on regulating tech.  

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