AI Implementation in Treasury – Attention Points

Introduction: AI in Treasury

AI is becoming mainstream in Treasury. We see more and more applications in Treasury in areas such as:

• Generative AI enables to gather documentation and to drafting documents including loan agreements, treasury policies, etc. 
• Predictive analytics, where AI analyses historical transaction patterns and external factors – such as interest rates or currency fluctuations, to provide more accurate forecasts, helping organizations manage liquidity more effectively and make better-informed decisions.
• Treasury Helpdesks and Chatbots: AI-powered chatbots can handle routine internal queries related to treasury policies, payment statuses, or bank account information, freeing up treasury staff for more complex tasks
• Automated Cash Application: AI, particularly ML (Machine Learning) and NLP (Natural Language Processing), can significantly improve the automated matching of incoming payments to outstanding invoices, even when remittance advice is missing or in unstructured formats. 
• Trade Finance Document Processing: NLP can be used to extract key information from complex trade finance documents (like Letters of Credit or Bills of Lading), reducing manual data entry and speeding up processing times.
• Fraud detection and anomaly identification: AI continuously monitors transactions and flags suspicious activity and anomalous patterns in real time. AI can generate synthetic data to simulate fraud scenarios, helping improving the defences against new and evolving threats.

However setting up AI in Treasury is complex and there are several attention points.

Regulatory issues

In 2024, the EU enacted the The EU AI Act (The Regulation (EU) 2024/1689), the world’s first comprehensive, legally binding law regulating artificial intelligence. Its main goal is to make sure AI is safe, trustworthy, and respects human rights.

It applies to any company handling EU citizen data, regardless of where the company is based. It affects developers, deployers, and providers worldwide whose AI outputs are used. 

It operates on a risk-based framework, meaning the stricter the rules, the higher the risk the AI poses to safety and fundamental rights. The EU AI Act classifies all AI systems into four risk categories: 

1. Prohibited / Unacceptable Riks: certain AI practices are banned. Examples include prohibited manipulation/exploitation techniques and certain social-scoring uses. Applies since 2 Feb 2025.
2. High Risk: Strict requirements for defined use cases, including risk management, data governance, documentation, credit scoring, profiling, AI in critical infrastructure.
3. Limited Risk: Systems requiring clear disclosure, such as chatbots, personalized product recommendations must inform users they are talking to AI.
4. Minimal Risk: Unregulated everyday applications, such as video game AI or email spam filters

Down below, you will find some examples of the impact thereof on Corporate Treasurers:

Treasury Use Case	Risk Category	Core Obligation	Impact on Treasurer
Credit Scoring & Risk Assessment	High Risk	Full conformity assessments, data bias auditing, continuous human oversight.	High: Treasurers using AI to assess supplier risk or counterparty credit must ensure systems comply by Dec 2027.
Cash Forecasting & Liquidity Planning	Medium Risk	Basic transparency, internal documentation of the algorithmic logic.	Medium: No heavy legal burden, but treasurers must document model assumptions to prevent inaccurate liquidity decisions.
Fraud Detection & AML Bots	Limited Risk (if internal only)	Ensure data quality and implement “human-in-the-loop” guardrails.	Medium: Safeguard against AI blindly blocking legitimate operational transactions without human review.
Third-Party TMS & ERP Plug-ins	Vendor Risk	Strict vendor due diligence and contract reviews.	Critical: Treasurers are legally considered “deployers” and hold liability if vendor tools violate EU standards.

The EU AI Act has a phased implementation from August 1,2024 until August 2, 2028.

The impact is important, because: 

• The EU AI Act obliges corporate treasury to move  from uncontrolled AI automation to strict, risk-based governance. 
• Treasurers deploying AI must ensure transparency, implement robust risk controls, and guarantee human oversight over automated decisions. 
• Furthermore, all treasury staff using AI must meet mandatory AI literacy training requirements
• Treasury teams must understand how their AI tools work, not just what they deliver. That means training staff to interpret model outputs, question anomalies, and maintain the critical thinking.

Other Aspects to Consider 

The main attention point for the use of AI is the data collection and the data quality; : “garbage in, garbage out”. According to the survey “CFO in the Age of AI” published by of Bearing Point in 2026, 74% of the CFO’s cite poor data quality and 63% cite data silos and fragmentation as barriers to AI adoption. 

Treasurers deploying AI must ensure transparency, implement robust risk controls, and guarantee human oversight over automated decisions. They need to focus also on the transparency and explainability of the AI technology. Regulators, the CFO and auditors want traceability, logs, and reason codes. 

Furthermore, all treasury staff using AI must meet mandatory AI literacy training requirements. Treasury teams must understand how their AI tools work, not just what they deliver. That means training staff to interpret model outputs, question anomalies, and maintain the critical thinking.

An important aspect is the data security and data protection, including the governance on data and ensuring data integrity, while mitigating cybersecurity risks. Treasury also needs to install guard rails specifying which persons have access to which applications. 

Treasurers need also consider the risk of deepfake: modern AI technologies have the potential to push disinformation and inaccuracies to a new level.

AI also create new fraud risks using person impersonation (voice- or image-based fraud),which are  difficult to detect. Hence this risk must be managed carefully. 

 AI projects also face the challenge of the AI integration with the IT legacy systems.

One need also to take into account Regulation and Compliance, including the EU AI  Act. Treasury must Make sure that by end 2027 all its AI tools and the one of your providers are compliant with the EU AI Act.

Measures to overcome the risks of AI in Treasury

Beside the technical robustness and safety, and the privacy and data governance which usually are the responsibility of IT team, please find down below some recommendations:

• Ensure the data quality and the model. AI can only scale where data is consistent, reliable, and interoperable. Implementing AI in treasury requires a data-first approach and Treasury teams must transition from legacy data silos to unified, secure data pipelines.
• Human judgment and oversight remain very important. Keep in mind  that AI does not consider relationship management nor human judgment. AI is a powerful tool, but human judgment remains critical for decision-making.
• Make sure that your systems and the one of your providers are comply with the EU AI Act EU AI Act: implement its guidelines and focus on the transparency and explainability of the model. This requires 3rd party risk assessment and a  vendor due diligence. Assess each vendor-supplied AI capability for documentation, data handling, audit rights, concentration and exit. 
• Treasury needs also to ensure that “human internal knowledge” is mapped. Hence,  when you build your model, make sure that internal processes, which are in the minds of people,  are correctly mapped and documented. 
• Human oversight: make sure that necessary rules and controls to are in place, including also the accountability. AI does nit replace human judgment. It is recommended to define clear ownership across Treasury, Finance, IT, Data, Risk, Compliance and Internal Audit. Treasury should not not allow autonomous AI agents to bypass existing treasury controls.
• Always be able to explain the model. A treasury, which can clearly explain how its AI systems work, will build trust amongst its stakeholders, including the lenders and the suppliers. Stakeholders will know that forecasts are backed by reliable data and methodology.
• Ensure a controlled automation. Only pre-approved, reversible actions with kill switches and circuit breakers should be put in place. In addition, ensure also audit trails 
• Make sure that the relationship dimension is considered when using AI. Take also human judgment in the decision-making process.

Conclusion

AI is a powerful tool which provides large opportunities for Treasury. However,  when putting it in place,  one cannot blindly rely on it, and several limitations and requirements need to be addressed. 

Data quality is very important, and hence it is important to map your data and data sources correctly. 

The human factor remains very important. Keep the control. AI can be your virtual treasury assistant, but you need to keep the ownership and control. The successful implementation of AI in treasury management requires a shift in mindset and a willingness to blend human expertise with machine intelligence.

Keep also in mind that December 2027, the treasury department needs to be compliant with the EU AI Act, requiring, a quality management system, technical documentation and a vendor due diligence. This takes time. Hence do not delay this.