This May I came back to treasury after a six-month break. To my surprise, I had missed it more than I thought I would. Last year, I felt like I had achieved my ambition. I had built out what I thought was the best treasury advisory team with the broadest capabilities. On top of that I'd helped build the succession of my team. I was proud of what I did, and it felt like a good time to take a break.
The biggest lesson I learned while away is that not much changes in treasury! It’s the how of treasury that changes. Yes, the geopolitical volatility and economic shocks are perhaps more heightened than we’re used to, the risks more concentrated, and some new risks rising such as with private credit. But technology and tools continue to evolve and have been a key driver of change in the profession. Now, that driver of change is artificial intelligence, and how treasurers use it.
I set myself the challenge of building it with AI, and within an hour I had a working prototype that would ordinarily have taken two days to build
While I was away, I wanted to test how far AI could take a real treasury problem, not just a hypothetical example. I'd come across a case where a company was buying imports priced in sterling rather than in the supplier's own currency, and they wanted to work out the cost of pushing the currency risk on smaller suppliers. The question was to calculate the cost of the supplier building a hidden risk premium into the sterling price and if so, whether it would be cheaper to buy in the foreign currency and hedge it yourself.
Normally building such a model is a two-day job. I set myself the challenge of building it with AI, and within an hour I had a working prototype that would ordinarily have taken two days to build. It wasn't perfect: I had to correct assumptions about how to align the currency data, and certain logics a couple of times before I trusted the output. But that back-and-forth was the interesting part, and it demonstrated how AI collapses the build time so I could then spend more of my time on the judgement, interpretation and fine-tuning it.
I have not yet run this on live client data: it stayed a proof of concept. But building it took an hour, not two days.
It's not that AI gets you a better answer; it's that it changes which questions are cheap enough to bother asking in the first place
In my break things have certainly moved quickly with AI, and I think most treasury teams are using it in one form or another. I think it needs a layered approach to be successful.
The first layer is personal use: I now run every client call through an AI summariser the same day to help me manage actions and follow ups. The key is to double-check facts and figures carefully. The second layer pertains to team processes. One team I worked with built a shared prompt template for drafting board risk papers, so junior members could produce a consistent first draft that a senior person just had to sense-check. This process allows more junior staff to own the delivery and hence get the critical experience.
The third layer is tooling investment. I've seen teams start building their own AI-powered cash forecasting tools: pulling bank and ERP data automatically and using AI to flag anomalies; effectively a lightweight alternative to parts of a TMS. It's a bigger lift because you need someone with both treasury and technical know-how, but the payback is real for teams still consolidating manually in Excel.
From my small FX analysis exercise, I have learned that this is one shift I think most treasurers are underestimating. It's not that AI gets you a better answer; it's that it changes which questions are cheap enough to bother asking in the first place. When testing a hunch costs an hour instead of two days, a lot more of the small, easy-to-ignore questions suddenly become worth chasing.
Karlien Porré is director of KP Treasury
