Banks have long used traditional AI and machine learning techniques for various functions, such as customer service bots and decision algorithms that provide a faster-than-human response to market swings. But modern generative AI is different from prior AI/ML methods, and it has its own strengths and weaknesses. Hari Gopalkrishnan, Bank of America’s chief information officer and head of retail, preferred, small business, and wealth technology, said generative AI is a new tool that offers new capabilities, rather than a replacement for prior AI efforts. “We have a four-layer framework that we think about with regards to AI,” Gopalkrishnan told. The first layer is rules-based automation that takes actions based on specific conditions, like collecting and preserving data about a declined credit card transaction when one occurs. The second is analytical models, such as those used for fraud detection. The third layer is language classification, which Bank of America used to build Erica, a virtual financial assistant, in 2016. “Our journey of Erica started off with understanding language for the purposes of classification,” Gopalkrishnan said. But the company isn’t generating anything with Erica, he added: “We’re classifying customer questions into buckets of intents and using those intents to take customers to the right part of the app or website to help them serve themselves.” The fourth layer, of course, is generative AI. Given the history, it’d be reasonable to think banks would turn generative-AI tools into new chatbots that more or less serve as better versions of Bank of America’s Erica, or as autonomous financial advisors. But the most immediate changes instead came to internal processes and tools. Bank of America is pursuing similar applications, including a call center tool that saves customer associates’ time by transcribing customer conversations in real time, classifying the customer’s needs, and generating a summary for the agent. The decision to deploy generative AI internally first, rather than externally, was in part due to generative AI’s most notable weakness: hallucinations. Banks are wary of consumer-facing AI chatbots that could make similar errors about bank products and policies. Deploying generative AI internally lessens the concern. It’s not used to autonomously serve a bank’s customers and clients but to assist bank employees, who have the option to accept or reject its advice or assistance. Bank of America provides AI tools that can help relationship bankers prep.
Walmart’s second freestanding, 3D-printed store in Huntsville with 16-foot concrete walls to serve as the online grocery pick-up and delivery location
Walmart has partnered with 3D concrete printing company Alquist 3D and general contractor FMGI to complete construction of its second freestanding, 3D-printed store addition. Working with the two firms, Walmart printed the 16-foot concrete walls of the structure, which will serve as an extension of the grocery pickup area in Walmart’s supercenter in Huntsville, Ala. Set to open the week of May 5, the completed Huntsville addition will serve as the retailer’s online grocery pick-up and delivery location as part of an overall store remodel. Other companies working on Walmart’s Huntsville commercial 3D printing project included Sika USA, which supplied customized concrete mixes formulated to address varying environmental conditions. In addition, Alquist’s robotics partner RJC Technology, which furnished robotic systems designed to achieve high-precision printing with reduced labor requirements. “In a commercial construction world that pays so much attention to project timelines and costs, our work with Walmart shows that 3D printing isn’t just a novelty – it’s an innovation ready to scale for retail and other industries,” said Patrick Callahan, CEO of Alquist 3D. “This second project clearly demonstrates how retail expansions can be faster, more cost-effective and less wasteful, paving the way for broader adoption for large-scale commercial builds.”
Morgan Stanley is concentrating on making its AI tools easy to understand, thinking through the associated UX to make them intuitive to use
Koren Picariello, a Morgan Stanley managing director and its head of wealth management generative AI, said Morgan Stanley took a similar path. Throughout the 2010s, the company used machine learning for several purposes, like seeking investment opportunities that meet the needs and preferences of specific clients. Many of these techniques are still used. Morgan Stanley’s first major generative-AI tool, Morgan Stanley Assistant, was launched in September 2023 for employees such as financial advisors and support staff who help clients manage their money. Powered by OpenAI’s GPT-4, it was designed to give responses grounded in the company’s library of over 100,000 research reports and documents. The second tool, Morgan Stanley Debrief, was launched in June. It helps financial advisors create, review, and summarize notes from meetings with clients. “It’s kind of like having the most informed person at Morgan Stanley sitting next to you,” Picariello said. “Because any question you have, whether it was operational in nature or research in nature, what we’ve asked the model to do is source an answer to the user based on our internal content.” Picariello said Morgan Stanley takes a similar approach to using generative AI while maintaining accuracy. The company’s AI-generated meeting summaries could be automatically shared with clients, but they’re not. Instead, financial advisors review them before they’re sent. Meanwhile, Morgan Stanley is concentrating on making the company’s AI tools easy to understand. “We’ve spent a lot of time thinking through the UX associated with these tools, to make them intuitive to use, and taking users through the process and cycle of working with generative AI,” Picariello said. “Much of the training is built into the workflow and the user experience.” For example, Morgan Stanley’s tools can advise employees on how to reframe or change a prompt to yield a better response.
Payment processors looking at platformization to offer an end-to-end product stack adjacent to payments such as advanced fraud prevention, network tokens, real-time account updates, and acceptance rate enhancement tools
“We’re seeing a shift where businesses are now looking for a payment processor that is more inclusive of a product stack, so a one-stop shop for everything,” Justin Downey, vice president of product at Maverick Payments, said. “Payment processors are looking for services that are adjacent to payments. That could be advanced fraud prevention, network tokens, real-time account updater, other tools that can increase the acceptance rate while reducing fraud,” Downey said. He highlighted the quest for a “frictionless checkout experience” — the new gold standard for merchants and consumers alike, as “something that truly makes it easy for customers to submit payments,” he added. The future Downey envisions, and the picture of the present he has painted, is neither purely competitive nor fully collaborative. It’s both. Processors will need to be architects — building unique, defensible intellectual property at their core — as well as curators, integrating complementary services to offer breadth and agility. The platformization trend means processors are stretching beyond payments into tangentially related domains — sometimes encroaching on territory once exclusive to FinTechs or even banks. “Payment processors are expanding into areas that are close to payments, but not exactly payments, like financial services, alternative payment methods, embedded finance,” Downey said. “Processors are in this unique position where, generally, they have a very strong distribution network, and they’re expanding into new product offerings that they can offer to their businesses, all as a one-stop shop. That’s a win-win for everybody,” he added.
Inflows into Cash App ecosystem have slowed to 8% growth to $77 billion in the first quarter, down from the 17% in the year ago first quarter; Cash App Card in 1Q slowed, to 7% growth in monthly transacting active users (at 25 million)
Cash App saw a marked slowdown in the first quarter, monthly transacting members using the digital wallet showed 0% year-over-year growth, remaining stagnant at 57 million users. Drill down a bit and the use of Cash App Card slowed, to 7% growth in monthly transacting active users slowed to 7% (at 25 million), where in previous quarters that growth rate had been in the mid-teens percentage points. Inflows have slowed to 8% growth to $76.8 billion in the first quarter, down from the 17% year on year growth that had been logged in the year ago first quarter. On an individual basis, the inflows come out to $1,355 per transacting active in the latest quarter, at 8% growth, also down from double-digit growth rates. The read across here is that at least for now, users are arguably being conservative about how much money they want to — or can — put to work with the digital wallet as entry point into the Block financial ecosystem. This performance is attributed to changing consumer behavior, including a shift in spending away from discretionary items like travel and media toward non-discretionary areas like groceries and gas. CEO Jack Dorsey said. “Tax refunds are an important seasonal driver of Cash App inflows. This year, we saw a pronounced shift in consumer behavior during the time period that we typically see the largest disbursements, late February and into March,” said Dorsey. “This coincided with inflows coming in below our expectations. During the quarter, non-discretionary Cash App Card spend in areas like grocery and gas was more resilient, while we saw a more pronounced impact to discretionary spending in areas like travel and media. We believe this consumer softness was a key driver of our forecast miss.”
Community banks and credit unions can enable extensibility through an internally built, custom middleware system, or by using external vendors with capabilities that stand on top of existing core systems
Through extensible capabilities, community banks and credit unions can punch above their weight by connecting to modern third-party apps and features — without swapping out their core systems. Extensible systems allow FIs to integrate with third-party apps with minimal friction, enabling easier access to account data and quicker pivots, says Christian Ruppe, SVP and chief innovation officer at Fitzgerald, Georgia-based Colony Bank. It lets banks and credit unions add modern apps and features — including quicker onboarding and transaction capabilities — without changing their core systems. Yet despite these benefits, many are falling behind on adoption. According to Ryan Siebecker, a forward deployed engineer at Narmi, a banking software firm, the route to extensibility can be enabled through an internally built, custom middleware system, or institutions can work with outside vendors whose systems operate in parallel with core systems, including Narmi. Other FIs that work with vendors — including Colony Bank and Grasshopper Bank — say using outside partners with capabilities that stand on top of existing core systems allow them to maintain lean internal operations without sacrificing the quality of the integrations. Luther Liang, SVP of product at Grasshopper Bank, told that by working with a vendor, the bank didn’t have to hire additional staff to manage software integrations enabled by extensibility. Colony Bank is starting to see results two years since it began its extensibility rollout. It’s enabled three major use cases: a modern account opening solution; an app that improves call center efficiency by allowing call center reps to co-browse with customers; and a client data visualization tool. Colony’s core provider “charges us per integration, and so now we’re not having to pay per integration — we have one integration, and we pay for that,” says Ruppe. “If you do it right, you can make it make sense immediately, but the long term is where you really win.” While Colony Bank might not be looking to compete with the top megabanks, “we know our communities better than they do…we can then provide technology that is specific to those customers,” he says.
Marketplaces’ third-party sellers efforts to stock up to avoid the cost of tariffs is inadequate because shoppers are also buying ahead
The efforts of third-party sellers on platforms like Amazon to stock up on goods to avoid the cost of tariffs will reportedly work for only a limited time. Because shoppers are also buying ahead to avoid the impact of tariffs, merchants will eventually sell down their inventory, place new orders, and be faced with the challenge of trying to avoid price increases. It is unlikely sellers can stock up on enough inventory to meet their needs for more than six months and that they will feel the full impact of tariffs in the third or fourth quarter. Amazon CEO Andy Jassy said that demand had not yet softened because of tariffs and that if anything, the company had seen “heightened buying in certain categories that may indicate stocking up in advance of any potential tariff impact.” Amazon pulled forward inventory in the first quarter, while many marketplace merchants accelerated shipments to U.S. warehouses to insulate customers from price spikes. Jassy added that Amazon’s risk is muted relative to rivals because many traditional retailers buy from middlemen who themselves import from China, “so the total tariff will be higher for these retailers than for China-direct sellers” on Amazon’s platform.
Success of Pix and UPI is paving way for a three-stage framework for state-led fast payment systems that involves weighting pre-requisites, implementation and scaling and establishing engagement mechanisms and regulatory adjustments
Pix and Unified Payments Interface (UPI), Brazil and India’s respective instant payment systems, provide two key lessons for governments interested in implementing new fast or immediate payment systems. First, the significant effect that government-led instant payment systems can have on citizens and the financial market transforms financial inclusion and market structures. Second, decisions made during the early stages of the process, such as system pricing and ownership structure, shape the power dynamics between local and international players, as well as incumbent and new entrants. These lessons are shaping an emerging framework governments can use to evaluate their need for central bank-led immediate payment systems, their potential structure, organizational features, and trade-offs involved in implementing a similar approach. The framework is composed of a three-step approach, including prerequisite weighting (i.e., “do we need this system”), the preparations needed to hit the ground running, and the process of setting up new immediate payment systems.
Researchers from MIT, McGill University, ETH Zurich, Johns Hopkins University, Yale and the Mila-Quebec Artificial Intelligence Institute have developed a new method for ensuring that AI-generated codes are more accurate and useful. In the paper, the researchers used Sequential Monte Carlo (SMC) to “tackle a number of challenging semantic parsing problems, guiding generation with incremental static and dynamic analysis.” Sequential Monte Carlo refers to a family of algorithms that help figure out solutions to filtering problems. This method spans various programming languages and instructs the LLM to adhere to the rules of each language. The group found that by adapting new sampling methods, AI models can be guided to follow programming language rules and even enhance the performance of small language models (SLMs), which are typically used for code generation, surpassing that of large language models. João Loula, co-lead writer of the paper, said that the method “could improve programming assistants, AI-powered data analysis and scientific discovery tools.” It can also cut compute costs and be more efficient than reranking methods. Key features of adapting SMC sampling to model generation include proposal distribution where the token-by-token sampling is guided by cheap constraints, important weights that correct for biases and resampling which reallocates compute effort towards partial generations.
Crowdsourced AI benchmarks should be dynamic rather than static datasets, and tailored specifically to distinct use casesAI benchmarks
Over the past few years, labs including OpenAI, Google, and Meta have turned to platforms that recruit users to help evaluate upcoming models’ capabilities. When a model scores favorably, the lab behind it will often tout that score as evidence of a meaningful improvement. It’s a flawed approach, however, according to Emily Bender, a University of Washington linguistics professor and co-author of the book “The AI Con.” Bender takes particular issue with Chatbot Arena, which tasks volunteers with prompting two anonymous models and selecting the response they prefer. To be valid, a benchmark needs to measure something specific, and it needs to have construct validity — that is, there has to be evidence that the construct of interest is well-defined and that the measurements actually relate to the construct,” Bender said. “Chatbot Arena hasn’t shown that voting for one output over another actually correlates with preferences, however they may be defined.” Asmelash Teka Hadgu, the co-founder of AI firm Lesan and a fellow at the Distributed AI Research Institute, said that he thinks benchmarks like Chatbot Arena are being “co-opted” by AI labs to “promote exaggerated claims.” Benchmarks should be dynamic rather than static datasets,” Hadgu said, “distributed across multiple independent entities, such as organizations or universities, and tailored specifically to distinct use cases, like education, healthcare, and other fields done by practicing professionals who use these [models] for work.” Wei-Lin Chiang, an AI doctoral student at UC Berkeley and one of the founders of LMArena, which maintains Chatbot Arena said that incidents such as the Maverick benchmark discrepancy aren’t the result of a flaw in Chatbot Arena’s design, but rather labs misinterpreting its policy.