Over the weekend, I tackled a small but persistent frustration. My workspace in Google Gemini was becoming a chaotic, single-file list of conversations. So, with AI acting as my coding partner, I built a simple userscript to add a folder system to the interface. Now I can organize chats, use colors, and drag and drop everything into a logical structure. I didn’t change the tool’s core function, but I dramatically improved the process of using it.

This small act of process improvement got me thinking about the larger conversation around technology in my field, and how we might be missing the forest for the trees when it comes to AI for treasury operations.

A Practical Tool to Start

For those interested, the Gemini script is a straightforward fix for a common problem. It runs using the Tampermonkey browser extension, a fantastic tool for customizing your web experience.

You can find the script and installation details here: Gemini Conversation Folders, v1.0

A Misplaced Focus

Creating that script brought me back to my time at last year’s AFP Conference. Panelists would ask the audience of finance professionals how they were using AI, and a wave of silence would follow.

This silence, I believe, comes from a misplaced focus. When we hear “AI in Treasury,” our minds instinctively jump to the core asset: the money. We imagine feeding an AI our cash positions, investment data, and market variables, asking it to perform high-stakes analytical tasks. This immediately—and rightly—triggers fears about data security, model bias, and the infamous AI “hallucinations.” The risk of an error is simply too high, so the door to AI is shut before the conversation can even begin.

But this is a failure of imagination. We’re so concerned with the idea of AI touching the money that we’re ignoring the vast, complex, and often inefficient ecosystem of processes surrounding it.

The Opportunity Around the Money

The most significant and immediate opportunity for AI in treasury has nothing to do with analyzing the assets themselves. It’s about strengthening and automating the operational scaffolding that supports, moves, and reports on those assets.

Think about the daily reality of treasury work. It’s a world of:

• Logging into multiple banking portals to initiate, approve, and track payments — each with its own authentication and navigation quirks.
• Consolidating cash, investment, and FX data from disparate systems to meet compliance and regulatory reporting requirements.
• Parsing and validating monthly bank fee statements to identify billing errors or negotiate better terms.
• Matching and reconciling transaction data between ERP, TMS, and bank records to clear exceptions quickly.

These are the activities that consume countless hours. They are the friction in our financial engine. And they are the perfect proving ground for AI-assisted process improvement.

My weekend project is a microcosm of this principle. I didn’t ask AI to generate brilliant new ideas for me (the core “asset” of the tool). I used AI to help me build a better system for organizing my ideas (the process around the asset).

Instead of asking AI to create a cash flow forecast, what if we used it to help us write a script that automatically pulls report data from three different bank portals into a single dashboard? Instead of having it predict market movements, what if it helped us kick-off non-AI reconcilement functions? There are so many opportunities for AI in treasury operations when we start thinking outside the box.

This is the real, tangible value proposition. It shifts the focus from high-risk predictions to low-risk automation. It empowers treasury teams to solve their own workflow problems, making them faster, more accurate, and more strategic. The answer to the silence in the room isn’t to find a way to trust AI with our money. It’s to start using it to fix everything around it.