What makes treasury data AI-ready?
AI-ready treasury data is accessible, sufficiently complete for the use case, consistently defined and owned by accountable people. The process also needs clear rules for who reviews the AI output and who can act on it.
Treasury usually has enough data; the problem is getting it ready for decisions
Treasury data already exists across banks, ERP systems, treasury management systems and spreadsheets. The challenge is turning those sources into information that is timely, consistent and usable when a decision has to be made.
Many teams still spend a significant part of the day collecting, cleaning, reconciling and explaining data before they can analyse it.
AI can reduce that distance by classifying transactions, identifying unusual movements, supporting forecasts and explaining changes in positions. Those benefits depend on whether the model can reach the right data with enough context and quality.
AI readiness starts with the business problem
Giving treasury access to an AI tool is not the same as having an AI-ready function.
A stronger starting point is a defined decision or process that needs to improve. The team should know which data the use case requires, how success will be measured, who reviews the output and what happens when the AI is wrong.
Without those answers, a pilot may still produce an impressive demonstration, but it is much harder to turn into a sustainable treasury workflow.
Perfect data is not a prerequisite, but the relevant data must be controlled
Very few treasury functions have perfect data across every entity, system and process. Waiting for an enterprise-wide data programme to finish before testing any AI use case can therefore delay useful work indefinitely.
A more practical approach is to scope the data to the problem. Bank balances may be highly reliable even when forecast categories remain inconsistent. A pilot can begin with the dataset that is complete, timely and controlled enough for the intended decision.
The limitation has to be explicit. AI can help map, classify and clean imperfect data, but it cannot remove the need for common definitions, ownership and standards.
The operating model is often a bigger failure point than the model
AI initiatives can work technically and still fail to become part of day-to-day treasury.
The operating model determines who provides the data, who owns the process, how exceptions are handled, which access rights apply and who is accountable for acting on the output.
If those questions remain unresolved, a good recommendation can simply sit outside the existing workflow. Adoption then becomes a people and process issue rather than a technology issue.
AI can help with data fragmentation without making fragmentation disappear
Treasury data fragmentation is a long-standing problem. Different systems can use different formats, hierarchies and definitions for information that appears to describe the same thing.
AI can reduce some of the manual mapping effort and recognise patterns across banks, ERPs, spreadsheets and other sources. It can also help surface missing fields and inconsistencies.
What it cannot do safely on its own is decide that two differently defined business concepts mean the same thing. Data lineage, ownership and business context still matter.
Treasury should own the business meaning of treasury data
Data ownership becomes more important as AI uses information across a wider set of systems.
Treasury should own the business definition and required quality of data related to areas such as liquidity, intercompany funding, payments and derivatives. IT can own the technical architecture, security and integration standards. Banks and technology providers can supply or process information without becoming the ultimate business owner of it.
For every critical dataset, the organisation should know who defines it, who can change it, who checks its quality and who is allowed to use it.
The best AI use cases today support decisions and exceptions
Decision-support use cases are easier to control than a jump directly to autonomous treasury.
Examples include transaction classification, anomaly detection, forecast support, bank fee analysis, policy search and data mapping. These tasks are bounded, the output can be checked and there is usually a measurable amount of manual effort to reduce.
Funding actions, hedge execution, payment release and crisis liquidity decisions carry a different level of materiality and accountability. AI can support the analysis, but final approval should remain with the authorised person.
Shadow AI is a data-governance problem as much as a technology problem
Employees will use AI if it helps them work. A policy that only prohibits unapproved tools is unlikely to be enough if the approved alternative is too difficult.
Treasury needs clear guidance on which tools are permitted, which information can be entered and which data must never leave the controlled environment. Education, technical controls and contractual protection all play a role.
The safer behaviour also has to be practical. A governed enterprise AI environment gives employees a usable alternative to copying sensitive financial information into public tools.
The business case should start with measurable manual work
AI ROI is easiest to assess where the current process has recurring data, high manual effort and a measurable baseline.
Reduced effort is only one benefit. Treasury can also measure faster decisions, earlier identification of risk, improved forecast quality and better use of specialist time.
The business case becomes much clearer when it starts with the process cost and the delay or error the organisation wants to remove, rather than with a general ambition to 'do AI'.
The practical path is selective investment now
Treasury does not need to wait for perfect technology or perfect data. It does need to choose use cases carefully and build the data, controls and process around them.
Near-term AI can act as a copilot: finding information, summarising data, classifying transactions, challenging assumptions and preparing explanations for management.
That creates value today while building the foundation for more integrated and agent-based workflows later.
About the interview
This article is based on Nomentia's interview with Dennis Kiratli, Treasury Transformation Lead within KPMG's Global Center of Excellence for Corporate Treasury Advisory. He advises multinational organisations on treasury transformation, including centralisation, in-house banks, payment factories, cash and liquidity management, technology, data, automation and AI.
