Artificial Intelligence in Treasury

A critical look at the state of technology, opportunities, jobs and sensible next steps for companies

Few topics are currently debated as controversially as artificial intelligence (AI). In Treasury, too, vendors promise efficiency gains, better decisions and automated processes. But what's really behind this, and where does AI already deliver genuine value today?

In our last article (Artificial Intelligence in Corporate Treasury - KPMG in Germany), we examined the drivers of transformation and success factors. This article focuses primarily on the current state of technology and market maturity of various use cases, as well as the implications for the Treasury department.

AI in Treasury has reached a significantly higher level of maturity in recent years. We're currently in a transition phase: from rule-based automation (e.g., macros, if-then logic) through machine learning to generative and agentic AI systems.

While machine learning is already being used productively in pilot projects today, agentic AI–meaning autonomous, decision-capable systems–is still in the development phase. Long term, it promises a nearly autonomous Treasury where AI agents independently run scenarios, assess risks and propose measures.

This is still fairly experimental, but the direction is clear: AI will change Treasury – gradually, not disruptively.

In our previous article, we introduced three core use cases with immediate added value, which we now revisit in terms of their market maturity.

1. Predictive Liquidity Planning
AI models analyze historical payment flows, seasonal effects and external factors to forecast future cash flows. This makes planning more realistic, fluctuations are detected earlier and the options for risk-reducing countermeasures become greater. AI can also quickly analyze multiple drivers in parallel and identify dependencies.

However, one thing is clear: AI can only forecast as well as the quality of the underlying data. Data collection, sensible aggregation and comparability over the years are the current challenges here. This use case therefore sits somewhere between machine learning and generative AI. 

2. Fraud Prevention in Payment Transactions – Protection Through Pattern Recognition
In payment transactions, AI can identify suspicious patterns that indicate fraud attempts. Through continuous learning from historical transactions, these systems become increasingly accurate. This minimizes risks without slowing down the payment process. 

Nevertheless, humans remain critical – AI detects anomalies, but the Treasury team must ultimately assess them. Additionally, available data history is decisive here as well. The stage is therefore still at machine learning.

Supplementing our last article, we'd like to present two additional use cases: 

3. Bank Statement Processing – Automation with High Precision
AI-powered systems automatically recognize and categorize transactions, reconcile them with booking data and capture them in the system within seconds. 

AI improves what a TMS system has previously solved through manually creating rules. TMS providers are already working on bringing corresponding functionalities to market. However, this isn't exactly a major innovation.

4. Treasury Assistants – Support in Daily Operations
Virtual assistants, comparable to chatbots, can help with information gathering or decision-making: "What's our net interest expense this month?" Questions like these can be answered in seconds in the future. 

Market assessment: The assistants are already live with some TMS providers but still in limited scope and in an advisory and supporting capacity.

From the use cases presented, it's clear where we're heading: toward interactive, data-driven Treasury workplaces. But what exactly does this mean for employees and jobs. 

Front Office
The effort for information gathering will be drastically reduced. AI takes over both research and interpretation of market news and prediction of upcoming fluctuations. Additionally, FX rollovers, money market facilities or intraday transfers will be semi-automatically generated. In the future, concrete hedging or funding strategies will also be proposed with cost and risk effects. Important: The role remains but shifts so that humans are only kept "in the loop" and decide and confirm strategies.

Middle Office
The fraud detection capabilities mentioned above will continue to develop and there will be a limit and policy monitoring system that detects violations before they become operationally relevant. Tasks shift toward validation and governance authority for the AI, meaning verification of the accuracy of underlying algorithms for report generation. In sum, the Middle Office becomes smaller but more qualified, particularly technically.

Back Office
This is where the biggest change is expected. For example, through AI-powered processing of documents (SWIFT messages, KYC documentation, contracts) but also through automation of controls, error correction for incorrect postings, counterparty and settlement information. In theory, the Back Office function would disappear entirely. But a company will probably never be completely free of special cases. For example, system anomalies and errors must be resolved and special regulatory topics must be incorporated accordingly. Over time, the department will shrink significantly and, depending on automation success, may disappear entirely.

Successful deployment of AI doesn't begin with purchasing software but with proper preparation. Companies that lay the groundwork today can scale faster and more securely later. In our last article, we outlined the strategic roadmap for your company in phases. Below we discuss the specific technical steps for getting started:

1. Data preparation:
   Building a central modern data lake where all relevant Treasury data is collected, structured and quality-assured (!).
2. Preselection of criteria:
   Definition of relevant indicators and company-specific drivers (e.g., cash receipts, volatilities, FX rates) that serve as the basis for analyses.
3. Application of machine learning:
   Use of statistical models to identify patterns, relationships and anomalies for the specific company-specific use case.
4. Machine-assisted evaluation:
   Interpretation of results – still with human judgment – to derive action options.
5. Automation:
   Integration into existing systems (e.g., TMS) to automate the corresponding task.
6. Reporting:
   Development of clear, comprehensible reports that make AI results transparent and understandable.

As much potential as AI offers – it's no panacea. Four limitations are particularly relevant at present:

• Data quality and security:
  AI is only as good as the data it processes. Incomplete or erroneous datasets lead to incorrect results. Additionally, data protection and compliance must be ensured at all times.
• Implementation Complexity:
  Introducing AI requires expertise, resources and adjustments to the existing system landscape. "Plug and play" rarely works.
• Adaptation of Employee Profiles and Change Management
  Integration of AI competencies, e.g., parameterization of AI models, data steward, prompt/workflow design and integration of these skills in the team
• Compliance:
  Treasury decisions have financial consequences. It must remain traceable how an algorithm arrives at a recommendation. Only this way can trust be built–internally and externally.

No panic – AI won't turn the Treasury function upside down overnight. Rather, it's a gradual evolutionary process.​

Companies should begin today to structure their data landscape, build competencies and work closely with their Treasury Management System (TMS) providers. Together, pilot projects can be implemented from which practical and tailored solutions emerge.​

The true added value of AI doesn't lie in the technology itself but in its intelligent application – combined with human experience, control and responsibility. Those who consistently pursue this path not only remain competitive but actively shape the Treasury of the future.

Source: KPMG Corporate Treasury News, Edition 160, November 2025
Authors:
Nils Bothe, Partner, Finance and Treasury Management, Corporate Treasury Advisory, KPMG AG
Hansjörg Behrens-Ramberg, Senior Manager, Finance and Treasury Management, Corporate Treasury Advisory, KPMG AG