Artificial intelligence continues to dominate discussion across corporate finance, with promises of rapid efficiency gains and more informed decision-making. Yet as treasurers assess what the technology can realistically deliver, a more grounded picture is emerging. According to new analysis from KPMG, AI is advancing quickly, but the near-term value lies in targeted use cases rather than sweeping automation.

The report ‘AI in Treasury - Between Vision and Reality’, authored by Nils Bothe and Hansjörg Behrens-Ramberg, examines where the technology has reached maturity, where it remains experimental and how treasury teams should prepare for the next phase of adoption. It follows an earlier article from the same authors that explored the strategic drivers behind AI transformation.

The paper argues that treasury is now in a transition from rule-based workflows to learning systems and, eventually, autonomous agents. Machine learning is already used in production pilots across several areas, while generative and agentic models are progressing in early development. The long-term direction is clear: more automation, more interaction between humans and systems, and a greater reliance on data-driven decision-making.

A maturing technology landscape

KPMG’s assessment shows that AI in treasury has moved well beyond proof-of-concept experiments. Traditional automation based on macros or if-then logic is giving way to machine learning models that evaluate large datasets, detect patterns and improve over time. These capabilities underpin a range of emerging solutions, from predictive cash flow analysis to anomaly detection.

Agentic AI, which could one day run scenarios independently and propose hedging or funding actions, remains at an early stage. The report stresses that no such system is making autonomous decisions in a live treasury environment today. Instead, the current phase is one of augmentation, where AI assists but does not replace human oversight.

This distinction matters for treasurers evaluating vendor claims. The report presents AI as an enabler of incremental rather than disruptive change, noting that adoption will build step by step as data quality, governance frameworks and human skills evolve.

Four use cases showing real impact

The authors revisit three previously identified use cases that already demonstrate tangible benefits, and add a fourth that is gaining traction.

1. Predictive liquidity planning. Machine learning models can analyse historical payment behaviour, seasonality and external factors to forecast cash positions more accurately. This provides earlier visibility of fluctuations and strengthens risk mitigation. The constraint is data quality. Forecasts are only as robust as the inputs, and many corporates still lack standardised and comparable datasets. Market maturity therefore sits at the boundary between machine learning and early generative capabilities.

2. Fraud prevention through pattern recognition. AI-based systems can learn from historic transactions to highlight unusual behaviour, improving payment security without slowing operational processes. While effective at identifying anomalies, the final judgement still rests with treasury teams. The maturity level remains firmly within machine learning, with model performance dependent on the breadth and depth of available data.

3. Bank statement processing. AI can classify transactions, match them against booking data and capture them in treasury systems within seconds. This improves upon traditional rules-based mapping by reducing manual intervention and increasing precision. Although not a radical innovation, it represents a step-change in efficiency and is already being integrated into treasury management systems.

4. Treasury assistants. Generative and conversational AI models are beginning to support everyday operational queries, such as retrieving key metrics or summarising positions. These assistants are live with several TMS providers but currently operate in an advisory capacity rather than executing actions. Adoption is expected to accelerate as models become more reliable and better integrated into workflows.

Shifting roles across the treasury organisation

As AI becomes embedded in systems, its impact on treasury teams will grow. The report maps out how front, middle and back-office roles are likely to evolve.

In the front office, time spent on data gathering will fall significantly. AI will support interpretation of market developments and help anticipate funding or hedging needs. Routine transactions such as FX rollovers or intraday transfers may become semi-automated. Human decision-making, however, remains central. The role shifts from executor to overseer, ensuring proposed strategies align with policies and risk appetite.

The middle office is set for a more technical evolution. As anomaly detection and policy monitoring tools advance, staff will focus on validating model outputs, ensuring controls function as intended and overseeing governance frameworks. The function may become smaller but more specialised, with rising demand for quantitative and data expertise.

In the back office, automation is expected to have the greatest effect. AI can process documents, reconcile messages, flag errors and follow up on missing information. This will reduce manual workload and shrink team sizes over time. Some activity will remain, driven by exceptions, regulatory interpretation and the need to resolve system anomalies. Full automation is unlikely, but the direction is unmistakable.

Preparing for adoption

The KPMG report stresses that successful AI deployment begins long before selecting a vendor. The foundations lie in data architecture, governance and skills.

The first requirement is centralised, high-quality data, typically in the form of a modern data lake that aggregates treasury-relevant information in a structured and consistent manner. Without this, even the most advanced models will produce unreliable outputs.

Next is defining the indicators and drivers that matter for a company’s specific context, such as cash cycle dynamics, volatility patterns or FX sensitivities. Machine learning techniques can then be applied to uncover relationships and anomalies.

Crucially, early results still require human interpretation. Treasury professionals must understand how models behave, which parameters influence outputs and when results warrant adjustment. Only once this oversight is in place should organisations begin automating tasks within their TMS environments.

Clear reporting is also essential so that stakeholders understand how AI-derived conclusions are reached. Transparency supports better decision-making and reinforces trust internally and externally.

Recognising the limits

The report emphasises that AI’s potential is significant but not limitless. Certain challenges stand out. For example, data quality and security remain the foremost barriers. Incomplete or inconsistent datasets lead to weak insights. Compliance, privacy and cybersecurity considerations must also be tightly managed.

Implementation complexity is often underestimated. AI is not plug-and-play. Meaningful adoption requires integration work, model training, ongoing monitoring and alignment with existing architectures.

Treasury teams also face evolving skills requirements. New roles such as data stewards, AI model operators and workflow designers will become essential. Effective change management will determine whether teams can adapt.

Finally, compliance and explainability are critical. Treasury decisions carry financial implications, so organisations must be able to trace how algorithms arrive at recommendations. Without this, both operational governance and regulatory assurance are at risk.

An evolutionary path for the years ahead

The paper concludes on a pragmatic note. AI will not transform treasury overnight. Instead, adoption will follow a staged evolution, shaped by data maturity, system readiness and organisational capability. Companies that structure their data, develop skills and partner closely with their TMS providers will be best placed to scale successfully. Pilot projects are the ideal starting point, providing insight into where AI creates genuine value and where human oversight remains indispensable.

KPMG’s report highlights that the true promise of AI is not in replacing treasury teams but enhancing them. The biggest gains will come from combining technology with experience, judgement and control. Treasurers who take a measured, informed approach today will influence not only their own operational performance but the broader shape of finance functions over the decade ahead.