Why CFOs Are Letting AI Agents Touch Their Cash — Carefully

Artificial intelligence has already improved the forecasting side of that equation. Now, a more consequential shift is underway as AI systems that don’t just predict cash needs but act on them enter the marketplace through core finance tech providers like Oracle, SAP and others.

This is the emergence of agentic treasury, systems designed to observe, decide and execute cash movements within predefined rules. For chief financial officers (CFOs), the promise is straightforward: higher yield on operating cash without sacrificing liquidity or control. The challenge is knowing where automation adds value, and where governance still needs a human hand.

Traditional treasury platforms are passive. They show balances, highlight trends and flag exceptions. Even when they use machine learning, the output is typically advisory and comprised of a forecast, a recommendation and a scenario.

Agentic systems change this posture. Instead of asking, “What should we do with this cash?” the system asks, “What am I allowed to do right now?”

Then it goes ahead and does it.

See also: Agentic AI Puts a Face on Corporate Treasury’s Next Leap 

Treasury Moves From Intelligence to Automation

The shift heralded by AI across the finance department has been gradual. Early artificial intelligence applications focused on pattern recognition such as predicting demand, identifying late payments and improving forecast accuracy. Over time, these systems gained confidence, fed by larger datasets and reinforced by measurable wins.

“Folks are just starting to understand that AI isn’t just automation with kind of sexier marketing,” Finexio CEO and founder Ernest Rolfson told PYMNTS in December.

Research in a PYMNTS Intelligence report, “How Agentic AI Went From Zero to CFO Test Runs in 90 Days,” has shown that close to 7% of enterprise CFOs in the U.S. have already deployed agentic AI in live finance workflows, while an additional 5% are running pilots.

The word “agentic” can sound abstract. In treasury, it boils down to autonomy with limits.

An agentic treasury system typically has clear authority over defined actions it is permitted to take, such as sweeping balances above a set threshold. It maintains policy awareness through embedded rules around liquidity buffers, counterparty exposure and instrument eligibility. The system provides continuous feedback by learning from forecast errors, late payments or volatility, while maintaining full auditability through records of why each action was taken, based on which inputs and policies.

In practice, that means monitoring balances across operating, concentration and subsidiary accounts in near real time, then comparing projected cash needs against policy thresholds. The system can automatically sweep excess cash into approved vehicles such as money market funds, overnight deposits or internal pools, and reverse those moves when payables, payroll or unexpected draws require liquidity. The system is not improvising. It is executing within guardrails set by finance leadership.

Large banks and treasury vendors have been laying the groundwork for this shift. Firms like JPMorgan Chase have already embedded predictive analytics into cash management. The next step is letting those predictions trigger action.

Bottomline’s AI agent, named Bea and set to roll out in the coming months, is designed to act as a digital team member in the office of the CFO, enabling treasurers, cash managers, and compliance professionals to interact daily with financial data using natural language.

See also: Vibe Coding Comes to Finance as CFOs Embrace Conversational AI 

Changing the Calculus Around the CFO’s Risk Equation

Three forces are converging to make this agentic treasury moment different from past automation cycles within finance. First, with rates no longer near zero, idle balances represent a real opportunity cost. CFOs are under pressure to treat operating cash as a managed asset, not a byproduct of transactions. Second, complexity has outpaced manual control, as global entities manage dozens or hundreds of accounts across currencies, banks and jurisdictions, and manual sweeps and end-of-day decisions simply don’t scale. Third, AI-driven cash forecasting has reached a level of reliability that makes automation defensible, not reckless.

CFOs are rightly wary of “black box” automation touching cash. That caution has shaped how agentic treasury systems are being deployed. Actions are typically limited to short-duration, low-risk instruments, while approval thresholds can escalate larger moves to human review. Systems are tested in parallel with manual processes before being trusted, and audit and compliance teams are involved early, not after rollout.

According to the PYMNTS Intelligence report “Time to Cash: A New Measure of Business Resilience,” 70% of firms surveyed already use at least one AI tool to manage cash flow. The most advanced, those using agentic AI, capable of autonomous decision-making, have automated up to 95% their accounts receivable processes, compared to just 38% among firms without AI integration.

The question for finance leaders is no longer whether artificial intelligence will move corporate cash. It is how much autonomy they are comfortable granting, and how quickly they want to capture the upside of a treasury function that doesn’t just watch money, but works it.