Artificial intelligence was a key theme at Swift’s Sibos 2026 event in Miami. In its guise as a Transaction Foundation Model (TFM), “it’s incredible” said Zack Anderson, Chief Data & Analytics Officer at J.P. Morgan, in a conference session entitled ‘Building enterprise intelligence with payments & banking data’.
“I’ve never seen an AI model perform as well as our new TFM. It’s incredible and has outperformed all our previous models,” said Anderson at the Miami Beach Convention Center (MBCC) on 28th September.
“For example, I’ve never seen a fraud application find a weak signal so well. It also helps us with straight through processing (STP), efficiency in payments and in many other business aspects.”
A Transaction Foundation Model is an advanced AI architecture trained on sequences of financial events – such as a where a payment emanated from and whether this aligns with prior patterns – to learn the language of money movement and finance. It transforms billions of banking transactions into a reusable AI asset that can improve efficiency and consistency across hundreds of use cases, such as anti-fraud applications, smart routing and personalised recommendation engines.
“It’s not easy to do, but it is so worth it,” said Anderson, as he advised putting some prep work into the underlying data to make sure it’s discoverable and useable to feed the model.
The signal and the noise
TFMs are beneficial in cutting through the ‘noise’ in a world where data is everywhere and finding a good use for it isn’t always evident.
As Paul Chang, Head of Payment Networks, Amazon Web Services (AWS), said during the Sibos 2026 panel, “58% of web activity is now driven by bots.” This is why cutting through the noise to find the signal you’re looking for is increasingly important.
“Previously financial services (FS) was rife with AI machine learning (ML) models in line-of-business (LOB) applications on loans, collections and so on – all rolled out as standalone tools. But generative AI has transformed this,” said Pahal Patangia, Global Head of Payments Industry at Nvidia, which makes the chips that power AI.
Pantagi explained that TFMs can “tell you more”, using transaction data to shed light on the behaviour of users. “This helps downstream uses cases on fraud and many other applications, as well as potentially increasingly the lifetime value of customers via better personalisation.”
He explained to the audience that the link between payments and data can power AI-enabled agentic digital commerce, which will impact corporate treasuries in the near future. Automated agentic AI-enabled commerce will increase transaction volumes as well as automation capabilities and interoperability advances.
“This is a new frontier,” said Patangia. He outlined how TFMs are a necessary component of this development, and the three crucial pillars needed for a successful rollout:
-
Technology (the model itself) and a clean, good dataset.
-
A scalable platform to deploy it enterprise-wide.
-
Human talent.
Competition for AI talent
The point about AI talent is pertinent as there is currently a lack of a decent pool of human talent deeply trained in AI design and implementation. This was addressed in a separate Sibos 2026 session focused on Building the AI-ready enterprise.
Dr. Diana Wolfe, VP & Head of AI Research and Strategy at infrastructure provider and consultancy Kyndryl, noted that “technological feasibility doesn’t equate to organisational capability”. Unless they are rolled out well, with an over-arching human mind and with appropriate controls and governance, AI implementations could fail.
Wolfe previously led emerging technology research, innovation strategy and global R&D initiatives at Avanade and Microsoft. She recommended that any organisation planning an AI rollout needs to understand the end goal first, and then pass three ‘gates’ before progressing:
-
Gate 1: articulate the value of your project and intention.
-
Gate 2: can your organisation absorb the huge amount of procedural and other changes needed to deploy AI well? Maybe it’s better and more affordable to let a human do it, especially if the cost of AI tokens is a factor.
-
Gate 3: should augmentation tools, which assist a human, be deployed or should fully automated AI used? Can your organisation handle the latter?
“Only if your organisation can say yes to all of the above should you do a full redesign and transformation project,” concluded Wolfe.
Other panellists in the AI talent debate included Fed Cohen Freue, EVP of AI & Data Operations at Mastercard; Bratin Saha, CEO of NTT Data AI Vista, a native AI firm; and Anjali Shah, Head of AI Strategy & Org Readiness, Global Payments & Trade (GP&T) at BNY.
During the session, moderators Stefanie Coleman, EY, and Karalee Close, the Global Lead for Talent at Accenture, ran an audience poll with those in attendance and online via Slido. When asked ‘What are the largest capability gaps in your current workforce?’ 33% cited the broad AI literacy of employees.
Close concluded that the findings show “workforce angst about AI literacy is real. It is up to leaders to train and reassure staff.”
