5 Questions with ... Zack Anderson, JPMorgan’s chief data and analytics officer for payments and global banking

Financial institutions are racing to develop and deploy agentic AI treasuries to provide more autonomous tools for chief financial officers. The agentic treasury market is expected to swell to $43.5 billion by 2031 in the financial services industry, up from $7.7 billion in 2026, according to data analytics company Mordor Intelligence . JPMorgan is approaching agentic treasury by shifting from three decades of incremental automation toward AI agents that can forecast liquidity needs, optimize cash positioning, and eventually execute transactions with minimal human intervention, Zack Anderson , chief data and analytics officer for payments and global banking, told FinAi News. [caption id="attachment_57389426" align="alignright" width="281"] Zack Anderson, JPMorgan’s chief data and analytics officer for payments and global banking[/caption] The opportunity isn't efficiency as most treasuries are already running like well-oiled machines, Anderson said. It's effectiveness. But getting there requires building trust incrementally. Anderson sat down with FinAi News to discuss how JPMorgan, which processes nearly $12 trillion a day, is building agentic AI tools for clients. What follows is an edited version of that conversation. FinAi News: JPMorgan has been automating treasury functions for years. How is agentic treasury different from prior treasury iterations. What level of autonomy are you comfortable giving AI agents in areas like liquidity management, payments routing and cash positioning? Zach Anderson: Automation in treasury isn't new. It's been happening for decades, and it's how small teams already manage massive global operations that need to run like clockwork. However, the major value creation from agentic AI isn't going to come from efficiency. Most treasuries are already efficient. The real opportunity is effectiveness: making better decisions that reduce working capital, shrink cash buffers and prevent operational issues before they occur. Agentic AI unlocks this through three layers. First, visibility: creating a real-time, global picture of operations, payments and positions. Second, prediction: forecasting operational issues and liquidity needs before they materialize. Third, action: agents that can execute on behalf of the treasury function. FinAi News: Are you orchestrating multiple specialized agents (liquidity, payments, FX) or using a single generalized model with tool-calling? ZA : Most agentic treasuries will use a mix of domain-specific agents and generalized agents, and the trade-offs between them are real. A specialized agent handling a narrow task, like routing a specific payment type, is easier to evaluate, monitor and trust because you can measure its outcomes clearly. A generalist agent is necessary for complex, cross-functional work like synthesizing data into a report for senior leadership. The architecture needs to support both. What guides the design are control and visibility. Some treasury payments are so critical that people verify them manually every single day, not because the automation is broken, but because failure is simply not acceptable. The agentic treasury has to earn that same level of trust. That means model governance and evaluation frameworks are not negotiable. Every agent, whether specialized or general, must operate within control requirements, and its decisions need to be transparent enough that teams and leadership can confidently depend on it. The architecture will need to start with trust. FinAi News: Are your treasury agents powered by proprietary models trained on JPM’s transaction data, or are you leveraging external LLMs with fine-tuning? How do you handle model drift in volatile markets? ZA : We are undertaking a layered strategy: foundational models for general reasoning and orchestration, domain-specific models for deep business context, and architectural patterns that are practical to maintain and adapt as markets shift. We work with a variety of AI models to leverage foundational large language models, and we primarily build agents on top of those models. The industry is shifting toward more practical approaches like retrieval-augmented generation, a move that is driven by rapid improvements in foundation models; larger context windows; and more sophisticated prompting. These approaches are also easier to update, which helps address concerns around model drift. FinAi News: What guardrails are built into the system? Can humans override agent decisions in real time? ZA : The starting point is accountability. Treasury teams are accountable for cash management and for ensuring the treasury machine continues to run. That accountability doesn't transfer to the agents. The system must be designed to give those teams full visibility and the ability to act when something isn't right. Governance begins well before an agent is deployed. We develop robust evaluation criteria to ensure model performance meets the bar for each specific use case. Once in production, model performance and potential drift are monitored continuously. If performance degrades, people need to know quickly and have the tools to intervene. Humans are still very much in the loop. The ability to override agent decisions in real time is a core design requirement, not an afterthought. The goal is to make sure people are focused on judgment and oversight rather than repetitive execution, while always retaining the authority to step in. FinAi News: Can you share metrics on the number of agents deployed or the volume of transactions they handle? ZA : We approach generative AI and agents in three layers, each at a different level of maturity. The first is AI for our teams. This is now quite mature. Everyone has access, whether through our firmwide LLM Suite, coding tools, or domain-specific agents. The second is delivering our core business. We process nearly $12 trillion a day, and there are many AI use cases already in production. JPMorganChase has been using AI and machine learning for many years, focused on increasing the efficiency of payments and fraud detection. For example, we deployed our Account Confidence Score tool last year, which is an AI and ML-powered proprietary fraud indicator risk score within J.P. Morgan Access, our global cash management platform. The involvement of truly agentic AI in actually processing transactions autonomously is nascent. We are focusing on removing friction for clients by building domain-specific AI that targets specific functions and workflows. The third is the agentic treasury, where we are enabling clients directly. This is still in early stages. We're starting with visibility, giving clients a clearer picture of their operations. There is a long runway ahead. FinAi Lending Summit, set for Oct. 7-8 in Las Vegas, will include speakers from Fifth Third and Capital One as well as a fireside chat with U.S. Bank Senior Vice President – Lending Operations Group Manager Suzanne Rathbun . To learn more about the 2026 event and register, visit here .

5 Questions with … Zack Anderson, JPMorgan’s chief data and analytics officer for payments and global banking

Agentic AI in treasury about better decisions, not just faster ones, Anderson says

Podcast: Fintech Sagent taps former Mr. Cooper exec as CEO

Related Posts

Ripple’s agentic treasury tool can reduce reporting cycle times by 90%

AI agents need machine-speed money; enter stablecoins

6 banks co-author principles for creating unified agentic commerce infrastructure

EMERGING FINTECH DIRECTORY

FinAi Podcast

SPONSORED

The Next Generation of Cyberthreats: Powered by AI

Build an Antifragile Strategy to Outperform the Market

How AI and Product Experts Turn Fuzzy Requirements Into Focused Dev-ready Roadmaps

Connect

Welcome Back!

Retrieve your password

Continue Reading with FinAi News Premium - Less than $2/Day