For CFOs and treasurers, the live question is where to trust AI first.
PwC’s 2025 Global Treasury Survey of 350 treasurers found that 74% were expanding or actively using AI, with predictive analytics a focus for 64%. Yet only 26% rated their AI capability as moderately or very mature. PwC 2025 Global Treasury Survey
When ambition runs ahead of capability, it is tempting to begin with the most sophisticated use cases: cash flow forecasting, scenario modelling or automated financial decisions.
In practice, a better place to start is work whose output can be seen, checked and reversed.
I use a simple progression for practical AI adoption in Finance and Treasury:
Automate → Personalise → Predict.
The stages can overlap. As AI moves from repetitive workflow towards judgement and prediction, the requirements around data, controls, validation and accountability should rise with the consequence of the decision.
Finance and treasury still contain a significant amount of repetitive, rules-heavy work.
Take reconciliations. A break appears between a treasury system, bank, custodian or accounting record. Someone identifies it, checks the underlying transaction, understands why the records disagree, routes the exception and follows it through to resolution.
Some of that requires judgement. Much of the surrounding workflow does not.
The same applies to confirmation matching, exception routing, data validation, information extraction and first-draft reporting. Conventional workflow automation may be enough for some tasks. AI adds value where the work involves interpreting less-structured information, identifying patterns or deciding what deserves attention.
In one use case I am advising on, AI is being used to surface potential unclaimed revenue: value already earned but not yet converted into cash. The system identifies where a person should look, while the underlying commercial and operational evidence determines whether financial action should follow.
This is the kind of early use case I favour. The technology improves visibility while accountability remains clear.
The same test applies to agentic AI. An agent that gathers bank balances, flags reconciliation breaks or drafts a morning cash position produces work a treasury team can inspect. An agent that moves money or changes a hedge is making a materially different decision.
The greater the consequence, the stronger the evidence, controls and human oversight should be.
NeuGroup ranks AI enablement as treasury’s No. 2 priority for 2026, behind cash flow forecasting. NeuGroup: AI Moves Up Treasury’s 2026 Priority List
My rule is simple: automate in production. Predict in parallel.
Automation creates capacity. Personalisation makes information more useful.
A Treasurer may need liquidity exposures, funding headroom and upcoming cash requirements. A business-unit leader may need overdue receivables, likely collection dates and the actions available to improve them.
The underlying data may be similar. The decision is different.
AI can increasingly surface the information relevant to each decision-maker rather than producing another dashboard containing everything.
For Finance, that creates an important shift: from distributing information to helping people see what requires action.
Prediction is where the potential value becomes larger. So does the cost of being wrong.
Cash flow forecasting is a good example.
A better forecast can improve liquidity, funding and working-capital decisions. It can also miss information an experienced treasury team understands immediately: unusual payments, business-unit behaviour, one-off events or weaknesses in the underlying data.
Run the AI forecast alongside the established process. Keep the existing forecast as the official position while the AI produces an independent view.
Compare forecast accuracy over time. Identify where the model performs well, where it misses context and which inputs need improvement.
This creates evidence before the technology is given more responsibility.
The real prize is the longer decision window.
If Finance can see several weeks ahead that a material amount of cash is likely to remain tied up in receivables, the CFO or Treasurer can work with the commercial team before liquidity pressure arrives.
Which receipts are delayed? Can disputes be resolved? Would an economically sensible early-payment incentive help? Could accelerating cash reduce the need to borrow?
Now Finance is helping the business change the outcome rather than explaining it afterwards.
Better prediction creates earlier choices.
Data quality matters more as the consequence of the use case rises.
A finance function can start before its data is perfect. The data requirement should match the decision.
A team may already have enough reliable information to classify exceptions or automate part of a reconciliation workflow, even if it is not yet ready to rely on an AI-generated liquidity forecast.
Early automation can also help improve the data foundation. Time released from repetitive work can be redirected towards cleansing data, clarifying ownership, improving processes and strengthening controls.
AI adoption and data transformation can therefore reinforce each other: create capacity through lower-risk automation, use some of that capacity to improve the foundations, and increase the sophistication of the use cases as confidence grows.
The 2024 Bank of England and FCA survey offers a useful reality check. While 75% of responding financial-services firms were already using AI, 62% of reported use cases were rated low materiality and only 2% were fully autonomous. Bank of England/FCA: Artificial Intelligence in UK Financial Services 2024
Before moving an AI use case into production, ask:
Can we tell whether the output is right?
Can the decision be reversed?
What happens if it is wrong?
These questions shift the conversation from what the technology can do to what the organisation is ready to trust it to do.
Start where the work is observable and the risk is manageable. Automate in production. Predict in parallel. Then earn your way towards the decisions where AI can create the most value.
Read More Productivity & Efficiency