How should treasurers be using AI?

Artificial intelligence has quickly become a regular topic in treasury discussions. Most teams agree it will change their work, but many still don't know exactly how. A big gap remains between interest and practical use, where mistakes can be more costly in treasury than in other finance areas.

This guide explains where AI can really help treasurers today, such as with cash investment decisions. It also covers when to be cautious, how to manage AI before using it more widely and how to get started safely.

Where treasury stands on AI today

AI adoption in treasury is happening, but it is still in its early stages. PwC's 2025 Global Treasury Survey found that 74% of treasury teams are expanding or using AI, but only 26% feel their skills are advanced. Coalition Greenwich research from early 2026 showed that less than one in 10 teams use AI for their main tasks.

Governance is lagging behind the interest in AI. The 2026 AFP Treasury Benchmarking Survey gave AI policies the lowest score among all treasury policy areas, at 2.9 out of 5. Our own AI in Treasury Report found a similar trend: treasurers often trust AI least for the tasks they most want help with.

Where AI adds value in treasury

AI works best on tasks that involve lots of data, happen often, and are easy to review. This fits much of the treasury workload, so the biggest early benefits are in day-to-day operations rather than big-picture strategy.

Cash flow forecasting

Forecasting is the most common use of AI in treasury, according to major surveys. Machine learning can spot patterns in past receipts and payments that spreadsheets might miss, including seasonal trends or customer payment habits. AI can also show why forecasts differ from actual cash, helping teams see where their assumptions may be off.

Human judgement is still needed for unusual events in forecasts. Things like acquisitions, large tax payments, or changes in supplier terms can confuse models trained on normal situations. Many teams use AI forecasts alongside their usual methods as a backup until they see consistent accuracy.

Reconciliation and exception handling

Matching bank statements to ledger entries is repetitive and follows clear rules, so it suits automation well. AI can handle the routine matches and send only the unusual cases to a person for review. This saves analysis time and helps prevent errors from piling up.

Reporting and portfolio queries

Generative AI can turn raw holdings data into easy-to-understand answers and draft reports. Treasurers can ask questions like which investments mature this month or how exposure is split by counterparty, without exporting data to a spreadsheet. Our article on using AI to query portfolio data explains how this works in practice.

Fraud and payment anomaly detection

Payment fraud remains a major risk. AI can flag unusual transactions before they go through. The AFP's 2026 Payments Fraud and Control Survey found that 76% of US organisations faced attempted or actual payment fraud in 2025. Only 17% were using AI to help prevent it, so there is plenty of room for more teams to adopt these tools.

Using AI in cash investment decisions

AI could be most valuable in cash investment decisions, but this is also where treasurers need to be most cautious. Decisions about surplus cash involve risks with counterparties, liquidity needs, and policy limits. Mistakes here can have serious consequences.

Supervising policy and concentration

AI is good at checking if a portfolio follows its rules. It can compare current holdings to the contents and limits of the investment policy. AI can also warn if too much is invested with one bank or sector before a limit is reached.

This type of monitoring requires clear, detailed data. Pooled products that mix many exposures are harder for any tool to analyse, while single-issuer investments make each exposure easy to see. The more transparent the structure, the more reliable AI oversight becomes.

Keeping people in charge of the decision

AI can perform a variety of high-concept tasks like modelling liquidity scenarios and comparing investments based on risk. However, the treasurer should always make the final decision about where to put cash, with an approval process that can be checked later. Our research found that only 4% of treasurers want AI to execute transactions, showing that most prefer caution.

Most teams want to know how AI can help improve their judgement without giving up control. If a model suggests a counterparty, it can’t be held responsible for the choice, but the treasurer who approves it is.

Where treasurers should hold back

Some tasks are too risky for AI to handle alone. Releasing payments, executing hedges, and making investments all need close human supervision because mistakes can be expensive and hard to fix.

Generative AI can sometimes give different answers to the same question, which doesn’t fit well with treasury's need for clear audit trails. Treasurers shouldn’t rely on AI to predict interest rates or market changes, since confident answers can hide real uncertainty. Sensitive data like bank balances and counterparty positions should never be put into public AI tools unless proper controls are in place.

Governance comes before scale

Most treasury teams already have solid policies for cash, counterparties, and payments. AI should follow these rules, with proper guidelines on which tools are allowed and what data they can use.

Building AI into existing policy

A treasury AI policy should state who can use each tool, which decisions require human approval and how results are recorded. It should also explain how models are tested before use and reviewed as things change. Involving all key teams early helps prevent informal tool use.

Data comes first

AI is only as effective as the data it uses. Poor data quality is still the biggest problem. PwC found that 76% of treasurers say poor data quality makes forecasting harder. Using consistent cash flow categories, reliable bank connections, and a single source of truth for positions will improve AI results more than switching models.

Where to start with AI in treasury

The best way to start is small and build trust gradually. The Association of Corporate Treasurers calls this a ladder, where teams move from improving personal productivity to supporting bigger decisions as controls get stronger.

Good first steps with low risk include summarising market commentary, writing policy language, and asking questions of existing data in plain language. After that, teams can automate reconciliation and compare AI forecasts to their current methods. Support for liquidity and investment decisions should come later, once the data, controls, and confidence are solid. 

For teams starting with plain-language queries, our guide to the TreasurySpring MCP server shows how to connect investment data to an AI assistant with read-only access. Measure results at every stage. Setting a starting point for forecast accuracy or time spent on reconciliation helps you see if AI is making a difference.

Putting AI to work in treasury

AI is, for better or worse, here to stay. While we’re still a long way from systems acting fully autonomously, there’s definitely potential for treasurers to build real use cases here. Start small, replacing simple, repetitive tasks with an engine to sort them out. You might start seeing time savings within days.

If you’re interested in seeing how AI can help with your cash investment process, contact our team. We can show you how our platform links your portfolio data to the tools your team already uses.

For anyone who can't tell the difference between their COP and their CCP, we've introduced our Treasury & Cash Investment Glossary.

*TreasurySpring’s blogs and commentaries are provided for general information purposes only, and do not constitute legal, investment or other advice.