AI AND THE FUTURE OF TREASURY
Navigating roadblocks to AI adoption
AI evolution is creating new opportunities and new challenges for treasury teams.
Key Takeaways
AI adoption in Treasury is scaling faster than most organizations can keep up with. Here’s what leaders need to understand before they jump in.
- The biggest roadblocks to scaling AI in Treasury aren’t technical. They’re organizational.
- Fragmented data and legacy systems remain the most common technical barriers to scaling AI in Treasury.
- Capabilities that once required a dedicated team or budget are now within reach through narrow, well-tested AI use cases.
- A successful pilot tests more than the model. It also tests whether the organization trusts and can act on what AI produces.
- Moving past the pilot stage means answering hard governance questions before adoption expands, not after.
- Waiting for perfect data or certainty costs more than starting with narrow, well-governed experiments.
Scaling AI effectively depends on core organizational capabilities
A renewed commitment to technology and infrastructure is essential.
Despite growing interest and investment in AI’s potential to transform treasury operations, many organizations remain stuck in a familiar cycle of pilots, proofs of concept and limited deployments.
Successful adoption will require resolving fragmented data, legacy infrastructure and ever-changing workflows. Organizations must also make a cultural pivot to build trust in new ways of working amidst market volatility, rapid innovation and regulatory scrutiny.
The concept of AI-powered treasury is, of course, compelling. AI's potential begins with a deliberate, disciplined approach grounded in strong governance, high-quality data and human accountability.
AI outputs are only as reliable as the data behind them
High-quality data needs to move efficiently across systems.
There are a wide range of AI use cases, each varying in complexity, business value and implementation readiness. The quality of underlying data, and how it flows through the organization, will have a major impact on implementation and results.
Treasury organizations generate and consume enormous amounts of information every day. Cash positions, forecasts, payment activity, market data, ERP information and bank reporting are all integral parts of the decision-making process.
The challenge lies in ensuring data that is accurate, well-managed and aligned with regulatory, privacy and security expectations. Your ability to deliver meaningful AI-powered impact depends on it.
“If teams don’t understand the process or how to input the proper data behind it, AI will only accelerate the confusion,” says Heather Manuel, SVP, Strategy Director, Treasury Management at PNC.
Organizations will quickly discover that before they can automate a process successfully, they need to address how data flows throughout the business. That starts with mapping where key data lives today, who owns it and where it breaks down between systems. Teams that get this right typically begin with one high-value data set, such as daily cash positions, and trace it end to end before expanding to others.
Treasury Barrier | AI and Operational Impact |
Fragmented data sources | Most treasury functions draw data from multiple systems: ERPs, banking platforms, treasury management systems and external market feeds. |
Legacy system constraints | Older platforms weren’t designed for modern integration. |
Data quality and standardization gaps | Incomplete records and data stored across siloed spreadsheets create accuracy risks. |
Unstructured data | A significant portion of treasury-relevant information lives in contracts, invoices and documents that aren’t yet machine-readable. |
API connectivity limitations | Connecting treasury platforms to real-time external data sources requires robust integration infrastructure. |
Cybersecurity and data privacy controls | Sensitive financial data flowing across systems and borders requires strict controls around how it is sourced, shared and protected. |
Costly capabilities become more accessible with AI
Starting narrow makes it easier to test use cases and establish AI value.
Many treasury teams are running the tools that make sense for their size and budget, even if they know a more advanced capability exists. Some teams that can’t justify a full market-data terminal might still want to model how currency moves affect its cash position. Other teams that can’t staff a dedicated scenario-planning function might still want to run more what-ifs before a board meeting.
AI creates an opportunity to go after problems that were previously too expensive or too complex to solve.
What this Means: The technology is usually ready before the organization is.
Moving beyond pilot hinges on organizational readiness
AI recommendations need a clear path into decision-making.
Cash forecasting, reporting, liquidity analysis and exception management are often among the first use cases teams pilot, because the benefits are easy to see. What's harder to see, until a pilot tries to scale, is whether the organization itself is set up to trust and act on what AI produces.
A forecasting model can perform well in a controlled environment and still stall once it reaches the rest of the organization. A well-designed pilot doesn’t just test the AI model alone, but how treasury leaders will review, challenge and act upon its recommendations. Working in tandem, teams should be able to clarify how AI outputs fit into existing decision-making, who’s accountable for reviewing them, and how the team responds when a recommendation looks wrong.
The answers to these questions often determine whether AI remains just a promising pilot or becomes part of how a competitive treasury organization operates.
One practical starting point: don’t build a new governance process from scratch. Fold AI review into a meeting or workflow the team already trusts, and let comfort with the tool build from there rather than asking employees to adopt a new process and a new technology at the same time.
Human-in-the-Loop Workflows
On a 2026 benchmark test of skills like information retrieval, market research and financial projections, the top AI model scored just 63% in accuracy. The other leading models scored as low as 51%.[2]
What this Means: Trust is one of the biggest roadblocks to scaling AI, and this data explains why. Keeping a human in the loop provides governance without giving up efficiency gains.
Waiting for perfect conditions slows progress
The cost of delay can outweigh the potential of controlled testing.
Treasury leaders don’t need perfect data, complete certainty or a fully polished AI strategy to take the first step. Instead, they should focus on removing barriers around data quality, operational readiness, workforce capability and trust. Those barriers are real, but waiting for perfect conditions carries its own risk: the cost of delay.
Meaningful progress requires starting with use cases that are narrow enough to measure and well-defined enough to govern. Teams should reimagine workflows before automating them rather than applying AI to processes built for an outdated operating environment. Finally, it’s important to quickly identify the employees who are adept at engaging with AI and let them experiment.
Encourage controlled experimentation within defined guardrails, then scale what proves effective.
Mike Gallagher, COO of Treasury Management at PNC, puts it this way, “Don’t stand in the way of employees finding their own ways to apply the tools, but watch it closely.”
Once an application starts to gain traction — measured by how widely it’s used or how much it touches — bring in the formal governance and continuity planning that any other business-critical tool would require.
The gap between experimentation and implementation remains wide. Most organizations are earlier in this process than they'd like to admit, which means there's still time to take these steps to get the foundational work right.
