Agentic AI in treasury risk management: Why data foundations matter more than algorithms

This content was originally published in The Treasurer

Until now, AI use cases have been largely assistive. AI informs decisions, but people still make them. That model is now beginning to shift. ION sees treasury teams looking beyond automative capabilities, to analyze how agentic AI can begin to manage risk in real time.

In this capacity, agentic AI capabilities will become increasingly important tools in the corporate treasury toolbox as the function continues to evolve into a strategic partner. So how can treasury teams ensure they are primed for success as the use of agentic AI in risk management proliferates?

Where are the opportunities for agentic AI in risk management?

Agentic AI has come of age at a time marked by political and economic volatility. With increasingly dynamic risk environments, treasurers are now expected to have continuous and real-time awareness of exposures and faster responses to emerging threats.

The transformative potential of agentic systems for risk management lies in their ability to reason, decide, and act within clearly defined boundaries. As such, corporate treasurers have begun exploring the use of agentic AI to rebalance portfolios as liquidity conditions change, and in flagging emerging risks that require immediate attention.

In a macroeconomic environment that necessitates greater liquidity, agentic AI can augment risk modeling and capital optimization by running Risk-Weighted Asset assessments and stress tests at greater speed and frequency than traditional approaches allow, creating the best possible environment to support human decision-making.

Preparing for the transition

However, as treasuries embark on the transition from assistive analytics to agentic decision making, responsible and careful roll out must be a priority. Rather than simply jumping on the AI bandwagon, organizations must understand the implications of using AI and assess the cybersecurity aspect before inducting new tools.

In practice, many organizations will find themselves constrained by limitations in the underlying data foundations these systems depend on. Risk, liquidity, and exposure information is often spread across multiple enterprise resource planning systems, treasury platforms, and business units, each with its own definitions, update cycles, and controls. While these environments have historically been tolerable for traditional, human-led decision making, they place clear limits on what agentic AI can responsibly achieve.

Agentic systems rely on a complete and accurate view of an organization’s financial position, drawing on a diverse range of sources such as TMS and banking portals. When exposure data is incomplete or unreadable, AI models are forced to fill in the gaps. In a risk management  context, those assumptions can quickly compound, leading to decisions that appear rational from a narrow data set but are misaligned with the organization’s true risk profile or policy constraints.

This danger is particularly acute when it comes to hedging. As a crucial process primed for agentic AI, we see corporate treasurers increasingly using agents to understand and normalize hedging policies, compare exposures and hedges, and suggest trades to improve hedge ratios. Organizations must ensure that adequate data foundations are in place to maintain an accurate risk profile, or else leave themselves exposed to the potential impact of compounded AI errors.

Those treasuries seeing early value from advanced AI capabilities share a common trait: they have invested in connecting systems to normalize and consolidate data from diverse sources, standardizing data definitions, and establishing clear governance over critical risk and exposure data. As a result, AI tools operate on trusted and logically linked information, allowing insights and actions to be interpreted and audited with confidence.

Organizations that have not yet trialed agentic AI have the advantage of beginning their pilot schemes with these learnings already addressed. However, implementing the necessary data systems sooner rather than later is critical. Laying strong foundations for AI within treasury functions will enable teams to move with the technology, rather than remaining one step behind.