Financial institutions stand at the edge of a major shift. Agentic AI systems promise to plan, decide and act across complex tasks from fraud detection to portfolio adjustments without constant human direction. Yet fresh analysis shows many banks and asset managers risk watching this wave pass them by. Their data simply isn’t prepared.
Published today, a MIT Technology Review article lays out the core issue. Success hinges less on model sophistication and more on an authoritative context data store. One that remains accessible, reliable and governed at scale. “It all starts with the data,” says Steve Mayzak, global managing director of Search AI at Elastic. Short sentence. Long one that follows: Agentic AI, unlike earlier generative tools that produce text or images, independently maps steps toward goals, pulls real-time information and executes actions. This autonomy multiplies whatever flaws sit in the underlying information architecture.
Numbers tell a story of enthusiasm colliding with reality. More than half of financial services teams have implemented or plan to implement these systems, according to Gartner findings cited in the MIT Technology Review piece. Adoption in finance teams has jumped over 600 percent in a year, reaching 44 percent by early 2026 per Wolters Kluwer data highlighted in a Neurons Lab research roundup. Yet only 11 percent of companies that aim to put agents into production have actually done so. Implementation barriers tied to data, governance and security explain most of the gap.
Consider the specifics. A Forrester study referenced in the MIT Technology Review reports that 57 percent of financial organizations still build the internal capabilities needed to take full advantage. Data challenges lead the list of obstacles. In the Neurons Lab analysis drawing from Forrester, AWS, EY and other sources, 48 percent of organizations cite governance concerns. Thirty percent flag privacy issues. Twenty percent admit their own data isn’t ready. Seventy percent of senior leaders, per an EY study in the same roundup, say organizations don’t grasp how vital data readiness has become. The consequences appear in real incidents. Ninety-five percent of respondents in one survey had faced at least one AI-related event. Seventy-seven percent of those produced financial losses. Fifty-five percent caused reputational harm.
But. The problems run deeper than statistics. Banks carry decades of accumulated information. Structured records sit alongside messy natural language in emails, contracts, PDFs and chat logs. A bank operating for 50 years might hold 60 different PDF formats describing the exact same process, Mayzak notes in the MIT Technology Review. Natural language proves far messier than spreadsheet columns. Organizing and cleaning it demands extra effort. Silos compound the difficulty. Data locked in separate systems leads agents to lag, return inconsistent answers or generate decisions hard to trace. Regulators, customers and internal teams lose confidence fast.
Accountability requirements raise the stakes higher. Financial rules demand clear explanations not just of inputs and outputs but of why specific data was chosen and how logic flowed to the next action. “You can’t just stop at explaining where the data came from and what it was transformed into,” Mayzak explains. “You need an auditable and governable way to explain what information the model found and the logic of why that data was right for the next step.” Hallucinations that early AI tolerated find zero room here. Markets move quickly. Risks and opportunities shift by the minute. Agents must deliver speed and precision together.
Recent coverage reinforces the point. A Grid Dynamics analysis posted this week reports 76 percent of firms plan agentic AI deployment within the next year. The hard part lies not in models but in readying data, controls and integrations with core banking, trading and compliance platforms. Active metadata and observability become essential. In wealth management, an agent assessing investment suitability pulls from risk questionnaires, liquidity needs, income records, KYC files and historical performance. Outdated metadata risks recommending unsuitable products. “If metadata isn’t accurate or current, the agent could recommend a product that no longer matches a client’s profile,” the Grid Dynamics post states. Observability tracks every input, update and audit trail.
So what does readiness actually require? Search emerges as foundational technology. Platforms that index both structured and unstructured data, apply security at every layer and deliver context turn fragmented repositories into usable memory for agents. The MIT Technology Review positions these search systems as the authoritative context and memory stores powering the shift. “Search is the foundational technology that makes AI accurate and grounded in real data,” Mayzak says. “Search platforms have become the authoritative context and memory stores that will power this AI revolution.”
Applications already show promise when data aligns. Agents monitor client exposure by scanning transactions, market signals and external feeds to flag risks in real time. They review trade workflows, spot discrepancies across formats and resolve exceptions with limited human input. In regulatory reporting they gather information from multiple systems, produce documents and document exactly how each output was derived. These tasks once manual, fragmented and difficult to scale now move toward automation while preserving the traceability regulators demand.
Yet progress demands discipline. Mayzak advises starting small. Pick a manageable use case. Let success compound. “Success can build on success,” he says. “While companies may aim to automate a 70-step business process, they are discovering that you have to start somewhere. What is working in the market is tackling the problem one step at a time. Once you get the first step working, then you can take the next step, and the next.” Incremental pilots feed an AI feedback loop. Executives gain signals to judge investment returns and refine approaches. Strong security controls, data governance and performance management surround the whole effort.
Other recent reporting adds color. The Neurons Lab roundup from earlier this month compiles projections that agentic AI could deliver $3 trillion in corporate productivity gains and improve average EBITDA by 5.4 percent. Companies using these systems report 55 percent higher operational efficiency and 35 percent average cost reductions. Specific wins include a Dutch financial institution that cut onboarding time 90 percent and staff workload 30 percent through KYC automation. An American bank saw calls to its IT help desk drop more than 50 percent thanks to an employee-facing agent. AML investigations shrank 50 percent in duration at one firm.
Still the readiness gap persists. Only 14 percent of leaders have fully implemented agents despite 34 percent having started. Security and risk concerns top the list at 63 percent. Lack of interoperability across technology stacks hits 55 percent. Tech debt and poor data governance each register at 55 and 48 percent respectively. Cybersecurity and data privacy dominate executive worries. Eighty-six percent recognize that agentic systems introduce fresh compliance headaches. Underinvestment in responsible AI frameworks runs about 30 percent below what many experts recommend.
Financial services cannot treat data preparation as an afterthought. Fragmented histories, regulatory demands and the autonomous nature of these agents combine to create a narrow path. Institutions that consolidate information, index it thoroughly, govern it tightly and secure it at every layer will move fastest. Those that don’t may watch competitors pull ahead. The models stand ready. The question is whether the data will catch up in time. Early movers already demonstrate measurable gains in productivity, cost savings and customer responsiveness. Laggards risk finding themselves with systems that amplify yesterday’s weaknesses instead of tomorrow’s opportunities.
Why Banks’ Data Foundations May Derail the Agentic AI Surge first appeared on Web and IT News.