10 AI Deployments Financial Services Leaders Actually Cite in Board Meetings

Key Takeaways

• The AI deployments financial services leaders actually cite in board meetings aren’t usually the newest ones. They’re the ones with a continuous, multi-year trail of published, dated figures, Bank of America’s Erica has been reporting usage numbers since 2018, DBS has published economic-value figures for years running, and that continuity is what makes a number citable rather than just impressive-sounding once.
• Scale varies by orders of magnitude depending on what’s actually being measured. Bank of America’s Erica has handled more than 3 billion client interactions since 2018; American Express’s fraud model makes a decision on every one of roughly $1.2 trillion in transactions it monitors annually, in milliseconds. Neither number is more “real” than the other, they’re just measuring completely different things, and citing one as though it proves the other is a common, avoidable mistake.
• Every major payments network now discloses that AI is being used against it as much as for it. Visa’s own 2026 threats report found AI-assisted fraud attempts rose 1,210% in 2025, with nearly half of all fraud now believed to be machine-driven, a genuinely useful counterweight to any board discussion that treats AI fraud detection as a solved problem.
• UK-headquartered banks are not behind their US and Asian counterparts in disclosure discipline. Lloyds Banking Group and Santander both published dated, specific value figures for 2025 and 2026, in Santander’s case broken down to a UK-specific self-service call target, matching or exceeding the granularity of Wall Street’s own disclosures.
• The most commercially significant AI deployments on this list aren’t customer-facing chatbots at all. DBS’s S$1 billion in 2025 economic value and Santander’s €1 billion 2026-2028 target both come primarily from internal productivity and risk tooling, not the assistants customers actually interact with, a distinction that’s easy to lose in coverage that defaults to writing about the chatbot because it’s the part a journalist can actually try.

Ten AI deployments across UK and global financial services, each with a dated, checkable figure behind it rather than a recycled statistic. None of these are the household names, Klarna, Kraken, Rolls-Royce, that this publication has already covered in detail. These are the ones that come up specifically in banking and payments board discussions, because their evidence is current enough to still be worth citing.

1. Bank of America’s Erica: the longest continuous disclosure record in the industry

Erica launched in 2018 and, according to Bank of America’s own press release marking its tenth anniversary on 20 August 2025, has now handled more than 3 billion client interactions from nearly 50 million users, averaging over 58 million interactions a month. More than 90% of Bank of America’s own staff use it internally, and it now handles 40%-plus of interactions on the bank’s CashPro corporate platform. “Erica has been learning from our clients for years, enabling us to leverage AI at scale, globally,” said Hari Gopalkrishnan, the bank’s Chief Technology Officer, at the anniversary. What makes Erica specifically citable, rather than merely large, is that Bank of America has published an updated figure roughly every year since launch, not just once at the start.

2. Morgan Stanley’s AI @ Morgan Stanley Debrief: 98% adoption inside months

Built on OpenAI’s GPT-4 and announced in March 2023, Morgan Stanley’s original AI @ Morgan Stanley Assistant gave financial advisors instant access to roughly 100,000 internal research documents. Morgan Stanley extended that in 2024 with Debrief, which drafts meeting notes and follow-up emails with client consent, rolling it out to the firm’s roughly 15,000 wealth advisors by early July 2024. According to Morgan Stanley’s own reporting, Debrief reached a 98% adoption rate among advisor teams within months of launch, a genuinely unusual figure for any enterprise software rollout, AI or otherwise. The firm extended the same approach into investment banking, sales and trading with AskResearchGPT in October 2024.

3. Wells Fargo’s Fargo: from 21 million to 245 million interactions in a year

Wells Fargo’s Fargo assistant, built on Google Cloud and launched in early 2023, handled 21.3 million interactions in 2023 and 245.4 million in 2024, an 11.5-times increase, according to reporting by VentureBeat. The system uses a privacy-first architecture that transcribes speech locally and, Wells Fargo says, has exposed no sensitive customer data to the underlying language models. “We want to bring the same smart, intuitive experience to banking. We want to be with our customers at all times,” said Michelle Moore, Wells Fargo’s Head of Digital for Consumer and Wealth & Investment Management, at launch. Notably, Wells Fargo’s own chief executive, Charlie Scharf, offered a rare dose of public caution in August 2025, saying “it’s very early to see any impact of any significance from AI,” a useful corrective from inside the same bank citing 245 million interactions.

4. Goldman Sachs’ GS AI Assistant: from 10,000 users to firmwide in a year

Goldman Sachs rolled its internal GS AI Assistant out to roughly 10,000 employees before expanding it firmwide in January 2025, according to CNBC’s reporting. Chief Information Officer Marco Argenti, who authored the internal memo announcing the rollout, said the firm’s ambition was “agentic behavior,” where the model would eventually act like a Goldman employee rather than simply describe what one would do. Different functions, developers, bankers, research analysts, wealth managers, received versions of the tool tailored to their specific work, including translation features for research documents delivered to international clients.

5. Visa: $40 billion in fraud blocked, and an unusually honest 2026 warning

Visa’s AI-driven fraud systems, dating back to the first neural-network application on a payment network in 1993, helped block an estimated $40 billion in fraud in the twelve months to mid-2024, CNBC reported. But Visa’s own Spring 2026 Biannual Threats Report is worth reading alongside that figure, not instead of it: it found AI-assisted fraud attempts rose 1,210% during 2025, with nearly half of all fraud now believed to involve some machine-driven element, and flagged close to $1 billion in AI-driven scam activity in the second half of 2025 alone. Visa disclosing both figures in the same period is itself a genuinely useful data point for a board weighing its own AI fraud investment.

6. Mastercard’s Decision Intelligence Pro: a trillion data points per decision

Mastercard’s generative AI fraud model, Decision Intelligence Pro, announced in February 2024, scans roughly one trillion data points to assess each of the 143 billion transactions Mastercard’s existing Decision Intelligence system already scores annually. According to Mastercard’s own figures reported at launch, the enhancement improved fraud detection rates by an average of 20%, and as much as 300% in specific cases, while cutting false positives, transactions wrongly flagged as fraudulent, by more than 85%. That false-positive figure matters commercially as much as the fraud-catching one: every wrongly declined legitimate transaction is a customer experience cost most fraud case studies don’t count.

7. DBS Bank: S$1 billion in economic value, and the bank’s own AI chatbot doubling its numbers

Singapore’s DBS Bank generated approximately S$1 billion in economic value from its data analytics and AI/ML initiatives in 2025, according to a Global Finance Magazine profile published 8 May 2026, powered by more than 430 deployed AI use cases, four times 2021 levels, running on more than 2,000 models. “We aim to sustain our leadership as an AI-enabled bank with a heart, using technology to deliver a competitive advantage,” said Tan Su Shan, DBS Group’s Deputy CEO. DBS’s generative AI chatbot, DBS Joy, handled more than 235,000 interactions in 2025 and lifted customer satisfaction scores by 23%, on top of a bank that crossed $100 billion in market capitalisation in June 2025, closing the year at $124 billion.

8. Lloyds Banking Group: £50 million in 2025, £100 million-plus expected in 2026

Lloyds Banking Group said generative AI delivered around £50 million of value in 2025 through more than 50 deployed GenAI solutions, and expects to generate over £100 million more in 2026, according to reporting dated 2 April 2026. Individual tools disclosed alongside that figure include Athena, an internal knowledge-search tool used by around 20,000 colleagues that cut average search times by 66%, GitHub Copilot adoption among roughly 5,000 engineers that improved legacy code conversion by 50%, and an AI-based HR assistant resolving around 90% of queries correctly on first contact. Lloyds is training an AI Academy for its full 67,000-strong workforce as it moves toward agentic AI use cases in 2026.

9. Santander: all 185,000 employees, and a UK-specific self-service target

Santander extended AI tool access to its entire global workforce of 185,000 employees by June 2026, according to the bank’s own announcement dated 22 June 2026, generating €35 million in business value in the first quarter of 2026 alone and targeting more than €1 billion in AI-driven business value across 2026 to 2028. “Being AI-first means applying AI where it can have tangible impact,” said Ricardo Martín Manjón, Santander’s Chief Data & AI Officer. Among the figures disclosed specifically for the UK business: a target of resolving 240,000 calls a year, around 40% of annual voice-channel volume, through AI-powered self-service, alongside a group-wide figure that AI now writes around 40% of the bank’s newly developed code.

10. American Express: the fraud model deciding on $1.2 trillion a year

American Express has applied machine learning to fraud detection since 2010, longer than most of the deployments on this list, and its current model, internally called Gen X, generates a fraud decision in milliseconds on every one of the roughly $1.2 trillion in transaction value the network processes annually, according to Amex’s own published account of the system. The model draws on more than a decade of transaction history, spending velocity, merchant category and device signals to make each call. Amex hasn’t published newer headline figures for Gen X specifically in the way Visa or Mastercard have for their own systems, which is itself worth noting: a fraud model that quietly keeps running at that scale, without needing a fresh statistic every year to justify itself, is its own kind of evidence.

What a UK C-suite should actually take from this list

None of these ten deployments are directly comparable to each other, a customer chatbot’s interaction count and a fraud model’s transaction-decision volume measure entirely different things, and citing either as a stand-in for “AI success” in general is exactly the kind of imprecision a board should push back on. What the ten do share is disclosure discipline: every figure here comes with a specific date, and in most cases, a named executive willing to attach their title to it. That’s a lower bar than it sounds like, and it’s the bar that separates a genuinely citable AI case study from one of the many that gets repeated in slide decks years after anyone last checked whether the number was still true.