Financial services companies have one of the toughest jobs in AI: they need systems that can act instantly on market moves, customer actions, and regulatory changes while keeping every byte locked down. That’s why agentic AI—AI that doesn’t just chat but plans and executes tasks—has become the holy grail in banking, trading, and insurance. The catch? Most firms don’t have the data infrastructure to make it work. “It all starts with the data,” says Steve Mayzak, global managing director of Search AI at Elastic. “No matter how smart your AI model is, if the data feeding it is stale, siloed, or scattered, the system collapses under its own weight.” In a recent survey, Gartner found that 54% of financial services firms admit they can’t deploy agentic AI because their data pipelines can’t keep up with real-time demands. The numbers don’t lie: 62% of banks say their biggest AI hurdle isn’t the model’s intelligence—it’s the quality and accessibility of the data it pulls from. Trading floors, for example, generate terabytes of market data every hour. A single delayed quote or misplaced decimal can trigger a cascade of bad trades. Agentic AI promises to cut through the noise by instantly analyzing this flood of information, but only if the underlying data is reliable and unified. Yet most banks still operate with fragmented systems where customer data lives in one vault, transaction records in another, and market feeds in a third. Even worse, many firms still rely on legacy databases that weren’t built for real-time processing. The result? AI systems that either choke on outdated information or flat-out refuse to act because they can’t verify the data’s accuracy. Security and compliance make this problem even harder. Financial services is the most regulated industry on Earth, and agentic AI doesn’t get a free pass. Every decision an AI makes—whether approving a loan or flagging a fraudulent transaction—must be explainable, auditable, and tamper-proof. That means data can’t just be clean; it has to be traceable back to its source. Firms that try to bolt AI onto existing systems often hit a wall when regulators demand proof that the AI’s conclusions aren’t biased or manipulated. “You can’t just dump data into an AI and hope it works,” says Mayzak. “Regulators will ask for the receipts—where did this data come from? How was it cleaned? Who approved it?” The firms winning this race aren’t the ones with the fanciest AI models. They’re the ones that have rebuilt their data infrastructure from the ground up, prioritizing real-time pipelines, strict version control, and automated compliance checks. Take JPMorgan Chase, for example. The bank spent three years overhauling its data architecture to support agentic AI across its trading, risk management, and customer service teams. The project cost hundreds of millions, but the payoff was immediate: the AI now processes 1.2 million transactions per second with zero downtime during market hours. Or consider HSBC, which built a single, cloud-based data lake that unifies customer profiles, transaction histories, and market feeds. By eliminating silos, HSBC cut its fraud detection time from hours to milliseconds—and reduced false positives by 40%. The lesson is clear: agentic AI isn’t about replacing humans with machines. It’s about giving humans the right tools to act faster and smarter. The catch is that the tools only work if the data they rely on is just as smart. For financial firms, that means investing in data infrastructure isn’t optional—it’s existential. The banks that get this right won’t just gain a competitive edge. They’ll redefine what’s possible in finance. The ones that don’t will keep throwing money at AI that can’t deliver.

What You Need to Know

  • Source: MIT Technology Review
  • Published: May 14, 2026 at 13:00 UTC
  • Category: Ai
  • Topics: #mit · #research · #science · #biology · #genetics

Read the Full Story

This is a curated summary. For the complete article, original data, quotes and full analysis:

Read the full story on MIT Technology Review →

All reporting rights belong to the respective author(s) at MIT Technology Review. GlobalBR News summarizes publicly available content to help readers discover the most relevant global news.

Curated by GlobalBR News · May 14, 2026

🇧🇷 Resumo em Português

O sonho das empresas financeiras de ter agentes de IA autônomos esbarra em um problema mais simples — e mais caro — do que se imaginava: a lama dos dados.

Mais da metade dos bancos e instituições financeiras no mundo já tentaram implementar soluções de agentic AI, aquele tipo de inteligência artificial que age por conta própria, tomando decisões em tempo real. No entanto, o que as pesquisas recentes revelam é que o maior entrave não está na sofisticação dos modelos, mas na capacidade — ou falta dela — de suas estruturas de dados acompanharem o ritmo acelerado dessas demandas. No Brasil, onde o setor financeiro é um dos mais dinâmicos e regulados, a adoção dessas tecnologias esbarra em sistemas legados, dados fragmentados e uma cultura de gestão ainda presa ao passado, o que pode custar caro em produtividade e competitividade.

Se as instituições não modernizarem suas bases de dados com urgência, o abismo entre o que a IA promete e o que o Brasil consegue entregar só vai aumentar — deixando bancos e clientes reféns de processos lentos e decisões defasadas.