Every week, I speak with treasurers who tell me the same thing:

“Management wants us to use AI.”

But when I ask why, the answer is often surprisingly vague. Perhaps competitors are talking about AI, the CFO recently attended a conference, or a vendor presentation made implementation look easy.

Whatever the reason, the message is usually the same:

“We need to do something with AI.”

The problem is that AI is not a treasury objective. It is a tool and, like any tool, its value depends entirely on the problem you are trying to solve.

The Wrong Starting Point

Many treasury teams begin their AI journey by focusing on the technology. They evaluate large language models, AI agents, forecasting engines, copilots, chatbots, and automation platforms.

While this may seem logical, it often puts the process in reverse.

The first question should not be:

“How can we use AI?”

It should be:

“What problem are we trying to solve?”

Treasury Already Has Hundreds of Potential Use Cases

The reality is that AI has potential applications across almost every area of treasury. These include:

• Summarising large volumes of emails
• Drafting treasury reports
• Analysing bank fee statements
• Reviewing contracts and banking documentation
• Answering questions about treasury policies
• Supporting hedge accounting analysis
• Improving cash flow forecasting
• Identifying payment anomalies
• Automating routine operational tasks
• Supporting treasury help desks and internal knowledge management

One example comes from Delivery Hero, which developed a custom large language model to answer questions about treasury policies and procedures.

For treasury professionals, a solution like this can significantly reduce the time spent searching through documentation while improving the consistency of answers across the organisation.

But the key point is this: the starting point was a specific business problem; employees needed faster and easier access to treasury knowledge.

AI simply happened to be the right tool for solving it.

AI Is Often Step 4, Not Step 1

When discussing AI with treasury teams, I often use a simple framework.

Step 1: Identify the use case

Begin with the business problem. Ask:

• Which tasks consume the most time?
• Where are errors most likely to occur?
• Which processes cause the most frustration?
• Which activities require significant effort but create little value?

Without a clearly defined use case, AI risks becoming a solution in search of a problem.

Step 2: Determine Whether the Process Can Be Automated

Not every process requires AI.

In fact, many treasury processes can be improved through standard automation, workflow redesign, APIs, integrations or better system configuration.

If a process can be solved with traditional automation, that may be simpler, cheaper and more reliable.

Step 3: Assess Data Quality

This is where many projects run into trouble.

Treasury teams often underestimate how much AI depends on data quality.

Poor forecasts, incomplete datasets, inconsistent naming conventions and fragmented data sources will not magically improve because AI has been added.

The old saying remains true:

Garbage in, garbage out.

Before introducing AI, organisations should understand:

• Where data comes from
• Who owns it
• How reliable is it
• How complete is it
• How frequently is it updated

Step 4: Select the Right AI Tooling

Only after completing the first three steps should technology selection begin. The right solution might be a standard LLM, a treasury copilot, a custom chatbot, an AI forecasting tool, an agentic workflow, a machine learning model, or something much simpler. The technology should always follow the use case, not the other way around.

Step 5: Run, Measure, and Improve

The first version is rarely perfect. Successful treasury AI initiatives typically improve through continuous iteration: monitoring outcomes, measuring accuracy, collecting feedback, refining prompts, improving data quality, and expanding the scope where appropriate. AI should therefore be treated as an ongoing improvement process rather than a one-off project.

The Most Common Treasury AI Use Case

When speaking with treasury teams, cash flow forecasting remains the most frequently mentioned use case. The logic is understandable: forecasting is data-intensive, patterns can be identified, and accurate predictions matter. However, forecasting projects also demonstrate why AI should not be the first step. If business units submit forecasts inconsistently, ERP data is incomplete, or underlying assumptions are unreliable, AI will not magically produce accurate results. In many cases, the greatest improvements come from strengthening processes and improving data quality before introducing an AI model.

The Real Opportunity

The treasury teams achieving the greatest success with AI are not necessarily those investing the most money. They are the teams that understand their processes, know their data, and clearly define the problems they want to solve.

AI can undoubtedly transform treasury, but treasury does not need AI for its own sake. It needs practical solutions to real business problems. The best AI projects do not begin with technology. They begin with a clear use case.

If your treasury team is exploring AI but is unsure where to begin, Pecunia Treasury & Finance can help you identify the right use cases, assess your processes and data, and turn ideas into practical solutions. Get in touch to discuss how AI can create real value for your treasury function. Contact us