AI is starting to take hold inside a handful of treasury teams that use it to write code, condense dense documents, accelerate analysis and even explore hedging models using quantum computing. These first movers stand out in a landscape where, as GenAI in Treasury: A Practitioner’s Guide—a report authored by Citi and Zanders with contributions from NeuGroup—makes clear, most companies remain early in their AI journeys. - The report shows that only about 18% of treasuries have progressed beyond pilots to using AI to assist in ongoing tasks and processes, with larger companies generally further along than smaller peers.
Why It Matters: What companies can take from AI pioneers’ early work. Members who are further along said the most impactful near-term benefits are not AI replacing staff, but reducing friction in work treasury already owns:
- Time-to-value is shrinking for everyday automation. Members said generative AI is dramatically shortening the cycle from idea to execution, allowing treasury teams to build, test and refine automations much faster while keeping humans firmly in control of review and decision-making.
- AI is becoming part of daily work, not a standalone project. Several members said personal AI assistants are now embedded in routine tasks such as research, drafting and analysis, freeing time for more judgment-based work rather than serving as a one-off innovation exercise.
- Ambitions are running up against data reality. Members said progress depends less on new tools than on the foundational work of consolidating and refreshing data, with dashboards offering a starting point but not the end state.
- Exploration without a map. As use cases multiply, members said the harder challenges include deciding what initiatives to pursue, understanding how AI reaches its conclusions, and ensuring that outputs can be explained and trusted by humans.
A coding assistant turns complex FX workflows into one-person tasks. One member described how treasury analysts at her company are using an AI tool that writes Python code based on prompts. “Some of our team members know how to write Python and others don’t, and anyone can just enter a prompt and it writes the code and explains it,” she said.
- A treasury director who reports to her described how generative AI has become a practical extension of coding work he has done for years, dramatically shortening the time it takes to build, test and deploy that code. “Before, if I ran into a problem, I’d search forums and piece together solutions,” he said. “Now I can describe what I want to do, review the code it generates and move forward much faster.”
- He uses Google’s Gemini AI assistant to help write and refine Python scripts that automate FX settlement and hedge processing. One example involves netting daily settlements across roughly 15 currencies, with trades moving in both directions. “In Excel, that’s pivot tables and manual work,” he said. “With Python, it’s a click of a button, and the operations team can execute the settlement.”
- He said generative AI acts less like an autonomous agent and more like a junior developer that accelerates work under human supervision. “You still need to know what outcome you want and how to review it,” he said. “But instead of a three-day project, you can get there in an hour.”
- To make the output usable by colleagues without programming backgrounds, he relies on the AI coding assistant to generate Python scripts that are broken into clear steps and include detailed comments explaining each part of the process. The scripts show both the logic and the results as they run, so even novice developers can follow what the program is doing. “It’s the same principle as a financial model,” he said. “You don’t hide everything in one formula. You make it transparent so someone else can review it.”
- The result, he said, is not fewer people, but more time spent on higher-value work. “I don’t see this as replacing roles,” he said. “It’s about getting to better outcomes faster and spending more time on what actually matters.”
AI assistants become part of daily routines. Another treasurer’s company is using personal AI assistants across finance following a push from the CFO for broader adoption of AI. The assistants help with routine tasks such as email and research, and the team monitors how frequently the assistants are used as adoption spreads. He said they are becoming part of everyday workflows and freeing employees to focus on more analytical work.
- In parallel, treasury has spent more than a year mapping where its data lives and now pulls it into Snowflake, with the goal of accessing the same information more frequently and in a consistent format. The work is intended to create a reliable foundation for future AI use by centralizing data that previously sat across multiple systems.
- That same Snowflake environment feeds the dashboards treasury uses for reporting. The member said the dashboards remain useful for showing what happened, but the team wants AI to help them spot patterns and relationships that dashboards alone do not reveal today. As the data becomes more complete and more current, the personal assistants may eventually play a role in pointing treasury toward insights it does not see on its own.
Other ways members are putting AI to work. Several members at the summit shared additional examples of how AI is showing up in their daily work, even in early stages:
- One team uses generative AI to simulate an analyst reviewing quarterly results, producing a set of questions that humans might raise and helping the group prepare more quickly for discussions with senior leadership.
- Another team has begun using AI to draft and redraft policies and intercompany agreements, reducing time spent on tedious document work that traditionally requires multiple rounds of manual editing.
- A member also described exploratory work that pairs AI with quantum computing to evaluate FX hedging decisions across factors treasury teams cannot manually process, an early signal of how emerging technologies may eventually influence risk management.
