Data-driven insights and news
on how banks are adopting AI
Source: Capitol AI
6 March 2025
Welcome back to The Brief!
Responsible AI conjures up thoughts of policy seminars, transatlantic disagreements over regulations and executive-level compliance workshops. But as we looked into it for our latest report on Responsible AI that came out yesterday, we found lots of upside for business. It turns out that banks that lean in most to embedding RAI practices are able to grease the wheels on AI deployment.
We spoke with senior leaders at six banks to get an on-the-ground look at how they’re deploying AI models responsibly, working through roadblocks and developing best practices in real time. Read on below for the toplines from the study, interesting RAI use cases and talent trends. Next week, on March 11, we’re hosting a virtual roundtable to talk it all through. Join us by registering here.
Also today, the long-promised AI wave that will change how banks hire (and fire) is now hitting. Plus, what the latest earnings from Europe’s big players tell us about their AI journeys.
Lastly, Alexandra was on the Searching Smarter podcast to give a lay of the land on where banks are on AI and what’s coming next. Take a listen!
The Brief is 2,090 words, a 7 minute read. You can read this Brief online and subscribe here.
– Alexandra Mousavizadeh & Annabel Ayles
The AI workforce shakeup is here.
DBS announced last week it would cut 4,000 jobs over the next few years thanks to AI efficiency gains. Italy’s Intesa Sanpaolo last October said it would both cut 9,000 jobs by 2027 and hire 3,500 “young people” by the first half of 2028 – a “generational change” the bank believes will give it a better chance to capitalize on AI.
Banks are quantifying the financial impact of their AI initiatives – and staffing for a Gen AI-enabled world as a result. They may not all make bold, public statements tying AI use to layoffs, but the way they hire (and fire) and talk about efficiency are all early markers that AI is delivering value.
Take Wells Fargo and JPMorganChase: On a January earnings call, JPMC’s CFO Jeremy Barnum mentioned the company’s headcount trajectory and the need for efficiency in the same breath; Wells Fargo touted headcount reductions as a result of efficiency initiatives to investors. Neither mentioned AI explicitly in those statements, but their AI investments clearly play a role.
Citi boss Jane Fraser has been a bit more specific: On the latest earnings call, she said “simplification work” and “using AI tools” were key to bringing about the bank’s transformation. The bank, meanwhile, continues to cut jobs with descriptions that seem ripe for AI disruption – like a team that compiles data and analysis on clients.
Moving forward, not every bank will mirror DBS and speak so candidly about AI-related layoffs, but they’ll all be taking stock of their AI returns and giving hints of how much of a workforce shakeup is in store.
📅 Tuesday, 11th March 2025 🕒 15:00-16:00 GMT | 11:00-12:00 EDT 📍 Virtual/ Zoom
For our newly-released Responsible AI report, we sat down with bank leaders – including NatWest’s Paul Dongha, CommBank’s Dan Jermyn, BNY’s Kirsten Mycroft, CIBC’s Aditya Anne and UBS’s Luke Vilain – to see how they’re building Responsible AI practices into their org charts, policies and processes to make managing risk a boon for innovation instead of a blocker.
Below are three takeaways from the Responsible AI report. Read the full report to learn about how leading banks are adapting their people, policies and processes to deploy AI models responsibly.
SAFER AND FASTER
When banks design processes specifically with Responsible AI in mind, they cut down the amount of time it takes to get new use cases into production.
Gen AI has created a backlog of AI use cases at banks, a lot of them stuck in different stages of the risk management process. The “heightened and new risks” of the technology requires “greater investment” in governance, risk and compliance functions, NatWest’s Head of Responsible AI and AI Strategy Paul Dongha told us.
Banks leading the way are using those investments to bake Responsible AI into the whole AI development process. Instead of having use cases stall at the final stages of production, these pre-flight safety checks give teams a better chance to address problems as they arise.
RAI AT THE TOP
Two-thirds of the banks we track now have a senior Responsible AI leader – someone who can “look over all these risks and consider them in a holistic way and put the end impact to the customer to our clients, at the real heart of what we’re doing,” UBS’s AI Governance Lead Luke Vilain said.
A key part of their job is to establish the principles of RAI within the bank (the easy part) and embed them into the existing controls within the AI development process so that governance isn’t a roadblock (the hard part).
Of the 50 banks in the Evident AI Index, two-thirds have a designated Responsible AI leader, more than double the number in November 2023
Source: Evident AI Index
Training is the key to getting a business on board with RAI. Consider these examples from banking. CommBank’s AI for All learning video series aims to teach 43,000 bank employees Gen AI skills. RBC has a one-day crash course for senior executives and board members focused on Gen AI and its risks. And CIBC doesn’t let anyone use Gen AI tools without training.
In turn, RAI leaders have been instrumental in designing processes so employees can develop Gen AI tools properly and avoid bottlenecks down the road. BNY trimmed a self-service questionnaire from 60+ questions to just 14 “triage” questions which it uses to decide how risky any particular AI project is and how much attention it needs. UBS has developers flag key risks to an independent risk assessment team that will forgo a more time-consuming impact assessment if it’s below a certain risk threshold.
RESPONSIBLE USE CASES
Banks have been busy laying the groundwork to build one-stop shops for Responsible AI – “assurance platforms” that will allow leaders to see how compliant every AI use case is in real time.
As leaders build them, they’re automating one of the most manual processes within the bank. To date, RAI teams have needed to review each AI use case every time an internal control or government regulation changed. And manual reviews were required before any new use case made it to production.
With assurance platforms, banks will get an end-to-end view of the whole AI development process, meaning they can automate each of those processes and move quicker as a result. Banks are rolling out the building blocks of these platforms. Here are four use cases that illustrate the industry’s approach to RAI:
#1. The Stress Test
Application: Automated model assessment (with human oversight)
Bank: CommBank
Why it’s interesting: The bank assesses how reliable its models are using a set of automated tools – a way to make it “quicker and more efficient” to deploy AI in addition to having it be safer, Dan Jermyn, the bank’s Chief Decision Scientist said.
Impact → Greater efficiency in checking models and better auditability
#2. The Wellness Check
Application: Risk assessment tool for identifying and mitigating risks across use cases
Bank: CIBC
Why it’s interesting: The bank’s platform identifies risks across the bank’s full AI use case portfolio, provides recommendations for how to mitigate risks and ensures accountability is clearly defined. It ensures that “AI-related risks are considered at every stage of the AI lifecycle,” said Aditya Anne, the bank’s Senior Director of Enterprise AI Governance.
Impact → Standardized treatment of AI-related risks for every use case launched
#3. The Warning System
Application: Gen AI monitoring platform with built-in technical guardrails and alerting capabilities
Bank: BNY
Why it’s interesting: The tool alerts executives when LLMs function unusually. It “combines the facets of AI innovation and governance into a single platform,” said Kirsten Mycroft, the bank’s Chief Privacy and Data Ethics Officer.
Impact → Improved ability to quickly adapt to newly published standards or guidelines.
#4. The Bird’s-Eye View
Application: Company-wide single source of truth for end-to-end risk management
Bank: NatWest
Why it’s interesting: NatWest is piloting a platform that aligns the bank’s policies, compliance criteria and technical measures in an “efficient single source where model validation teams can provide assurance across all of the controls stemming from your internal and external principles and policies, including the EU AI Act, ISO, OECD,” said Dongha. Once implemented, assurance platforms such as this would prevent noncompliant models from progressing through the development cycle.
Impact → Improved efficiency of AI lifecycle risk management.
As in North America earlier this quarter, these are good times for European banks. But…
Efficiency has become every CEO’s new favorite word.
Revenues are going up, but so are costs. And investors are demanding proof that CEOs aren’t too far out over their skis as their tech investments add up. In response, as you saw with the AI-driven job cuts announcements above, banks want to show how efficient they’re becoming with AI – and how much they’re saving in the process.
Lloyds says generative AI is allowing for faster product launches and reducing how much it costs to retain customers in the process. ABN AMRO increased its IT spend to cut costs in other areas, singling Gen AI out as its way to achieve this “cost discipline.” UniCredit earmarked about €2.5 billion over the next three years to "deliver IT projects at a lower unit cost and faster time to market” after finding the revenue and cost savings its existing AI projects delivered were ten times the amount it spent.
Commerzbank illustrated this spend to save mindset most explicitly. The bank says it will spend €140 million on AI initiatives over the next four years to bring €300 million in cost savings and revenue loss prevention in areas like fraud and cyber.
What’s Next: As investors continue prioritizing cost cutting, banks will introduce more AI use cases in higher cost areas like investment banking and wealth management.
HSBC CEO Georges Elhedery told investors he plans to “seize the opportunity of AI and generative AI” this year – using the technology to improve customer service in its mobile app and call centers and introduce coding assistants and other productivity-boosting tools for client onboarding, KYC and credit applications, among other practices.
Lloyds Banking Group will be focusing on “extending Gen AI use cases across the group,” CEO Charles Nunn said in the bank’s earnings announcement. That includes “providing personalised financial goal planning and money management journeys for customers as we continue to identify ways to close the advice gap.”
On ABN AMRO’s earnings call, CEO Robert Swaak noted that the bank had introduced its first client-facing Gen AI tools: A chatbot for payments app Tikkie and a voicebot for credit cards – both designed to “build on our digital product experience and client contact”
After seeing “reductions in call hold times and escalations to second level support,” with the Gen AI virtual assistant it uses in personal banking, TD Bank is rolling the tool out to contact centers in its wealth and insurance businesses, it said during its earnings presentation.
BNY is the latest bank to strike a multi-year deal with OpenAI, giving the bank access to new reasoning models it can embed into Eliza – the bank’s AI tool. AI Hub head Sarthak Pattanaik also recently described the firm’s agentic strategy, detailing a lead recommendation agent which gives the bank’s sales teams recommendations about how to approach clients.
Citi CTO David Griffiths discussed the bank’s approach to fine tuning and customizing LLMs on a Bloomberg podcast. Griffiths noted that Citi doesn’t like to fine tune models for general tasks “unless you absolutely have to” because it hinders speed and complexity. It’s a different approach than than the pro-customization approach RBC’s Foteini Agrafioti described in our Q&A🔒.
Chinese banks have embraced DeepSeek, a Global Finance report said. Postal Savings Bank and Haain Bank both use DeepSeek for its chatbots and the Industrial and Commercial Bank of China uses DeepSeek for wealth management tasks, the report said.
Sun 9 - Thurs 13 March:
HumanX, Las Vegas
Mon 10 - Tues 11 March:
MoneyLIVE summit, London
25 March:
AI for CFOs, London
Tues 15 - Weds 16 April:
AI in Finance Summit New York, New York
Mon 28 April:
Financing the AI Revolution, New York
Tues 29 April:
LlamaCon, TBD
Alexandra Mousavizadeh | Co-founder & CEO | [email protected]
Annabel Ayles | Co-founder & co-CEO | [email protected]
Colin Gilbert | VP, Intelligence | [email protected]
Andrew Haynes | VP, Innovation | [email protected]
Alex Inch | Data Scientist | [email protected]
Gabriel Perez Jaen | Research Manager | [email protected]
Sam Meeson | AI Research Analyst | [email protected]
Matthew Kaminski | Senior Advisor | [email protected]
Kevin McAllister | Senior Editor | [email protected]