Cycle Race: Keeping AI Fit for a Finance Function

Sometimes a technology comes along that seems perfect for the time. But AI, despite its many advantages, cannot be unleashed on any function, least of all treasury and finance, without first setting the rules of engagement. Aidana Zhakupbekova, CFO of Belgium-based expense management fintech Rydoo, explains how her finance team is addressing the issue.

New research from J.P. Morgan reveals that over half of its treasury respondents see geopolitical tensions as the number one barrier to long-term business planning, ahead of inflation and interest rates.  Improving the cycle of cash flow and working capital ranked as the top priority for the next 12 months. Implementing new technology was close behind as teams seek new ways of tackling persistent threats.

While trading activity is often knocked off course by geopolitical interventions, the main challenge with recent multiple overlapping events (the ‘polycrisis’) is that of chronic uncertainty, says Zhakupbekova. It makes mid- to longer-term planning almost impossible, while at the same time, traditional financial close periods and methods may deliver data too late to be of practical value in the short term.

The research results suggest that AI could have a major role to play in helping finance leaders keep abreast of almost daily changes in the global trading environment. As Zhakupbekova states, “we can’t necessarily wait for a month or a quarter to close before carrying out business analysis: decision-making in the current environment needs to become a lot more agile”.

AI is a powerful tool, with many actual and potential use cases in finance. But it’s understandable that one of its main use cases is in offering financial decision-makers increased visibility, accuracy, and speed over the fundamentals of cash flow and working capital.

In theory, AI can help finance professionals navigate global instability and gain a manageable level of certainty. This perhaps comes at the expense of exploring more experimental use cases, for now at least. But there is another more immediate AI risk lurking.

Too much information

While the more current the data, the more relevant it is, Zhakupbekova notes that data recency can become an issue in itself if financial leaders are given too much and too often. Indeed, an overabundance sets up the notion of analysis paralysis, where individuals struggle to make a decision because they constantly feel they should wait a little longer until the next update.

Mindful of the risk of analysis paralysis, Zhakupbekova argues that smart decision-makers would not keep wondering if they should wait for the next update, but instead take a different angle. “The real question that should be asked is what is expected to change significantly in a defined period if they don’t make a decision now. If that raises a valid concern, then maybe they should wait for an update. If not, then go ahead. But if persistent delays become standard business practice, it may signal a potential governance issue.” It’s a matter that she believes must be put in order before AI is used – a view underpinned by early experience.

“Even within our own finance teams – and we are naturally cautious around data – we forgot to pre-validate what was AI was processing. It took us about a week to spot the first issue, but we started to recognise that our analysis was not entirely correct. At that point we had to quickly roll back on some of the reports we had already communicated.”

Even with data integrity now being checked ahead of AI processing, as a matter of strict governance, all AI outputs are sense-checked before being published to the wider group. This approach is especially important where other functions are providing additional data, stresses Zhakupbekova. “You may know your own data is correct, but be careful to check that the overall output makes sense.”

After that first experience, and with Rydoo seeking to further leverage AI to enhance various internal processes, Zhakupbekova and her finance team worked quickly to align AI and data governance with their key processes and goals.

Rydoo’s own client-facing expense management platform is already AI-driven, as is the Semine AP platform it acquired in 2025, and both are subject to rigorous governance. The journey towards AI adoption for Rydoo’s internal use thus similarly required the foundation of robust policies and governance, says Zhakupbekova.

“Once these were properly tuned to AI, and people knew and understood what they can and can’t do with this technology, it gave us more comfort when exposing internal data to it.”

And that governance plan demands prudence to the last. “All employees who now look at AI outputs must develop and show critical thinking. They have to understand that AI is not a one-stop-shop where they just do whatever it tells them,” she explains.

It takes time to do it right

With the pressure on to perform under testing global conditions, it is of little surprise that Zhakupbekova sees more treasurers and CFOs using AI to stay afloat rather than push ahead with more experimental use cases. “It’s always going to be a question of priorities. We would all like to do everything at the same time, but unfortunately, it’s not possible,” she comments. Speaking of her own team, she saw that “implementing AI properly takes time, not because people didn't want to, but because they have demanding day jobs to do, too”.

In the setting of priorities, where teams need to “pick and choose their battles”, Zhakupbekova accepts that those implementing AI will be torn between addressing the here and now versus establishing plans for the longer-term. “If a business really needs something today, of course it should prioritise that. But I believe it should do that while at least trying not to lose the sight of its mid- to longer-term goals. We’re trying to do both at Rydoo.”

And there is plenty of potential for AI in the longer term, asserts Zhakupbekova. This is especially true within finance. But given the nature of the job, she accepts that the team has to start with “the low-hanging fruit, optimising what we’re most certain of so it doesn’t pose any real risk to the company”.

But as the low-hanging fruit is harvested, and confidence builds, Zhakupbekova is sure that more interesting or experimental use cases will naturally emerge. Indeed, her teams are now starting to develop their own limited AI-based tools for internal use. “These still needed to be approved by security, but we’re a huge leap ahead of where we were even just a year ago.”

Sharing the experience

Under the guidance of Rydoo’s clear rules of AI engagement, Zhakupbekova’s advice to her own and other teams was to start small. This facilitates their first exploratory steps with no undue pressure. “Within finance, we started with the simplest parts,” recalls Zhakupbekova. “We asked which were the tasks we were performing every day that were repetitive and no one liked doing? From there we just experimented to see what AI could do to help. We never set any goals. But we did insist on weekly sessions where the teams share what they are doing, what they want to achieve, and what has or hasn’t worked for them.”

Taking this approach has enabled other teams at  Rydoo to develop their own skills and apply recognised successful methods to their own projects. Having explored a number of AI opportunities, several internal teams are “slowly moving towards not just optimising their day-to-day work, but also building a deeper AI layer”.

It’s not all been plain sailing. With the exploratory work ongoing, there has naturally been a lot of additional data analysis generated within several of functions. Admirable though Zhakupbekova says the enthusiasm has been for AI exploration, the recipients of that output were at risk of falling into the analysis paralysis trap. The level of output had to be scaled back accordingly.

Nonetheless, she acknowledges that accelerating from a standing start on the day-to-day work, to reach the stage where multiple teams are now asking “what are the coolest AI tools we could build?” is encouraging. It demonstrates how AI can rapidly muster significant interest, moving Rydoo’s “mid-term” project discussions on from tackling only repetitive tasks, to tasks that team members felt were not adding value (and admittedly did not enjoy carrying out).

Rydoo’s employees are clearly up for the AI development challenge. But whether or not AI projects remain isolated within a specific function, or are tackled across the whole business, depends on the stage of development of the organisation, says Zhakupbekova. “When you start from scratch, it’s hard to tell the whole company that we’re going be doing this only with AI from now on, so we just said to each team individually, ‘experiment how you will, and let us know how it goes’.”

Having allowed free rein for a while, and gathered and shared interesting results, Rydoo is now moving towards a semi-centralised approach. Zhakupbekova reports that the firm is now slowly building a centralised AI management team. Each function is still free to experiment and build the tools that they need or want, but now there is company-wide leadership steering AI’s use. “And from time to time, the central team brings everybody to the table to collaborate and discuss how we could use AI to optimise any significant cross-functional touch points.”

Productivity bonus

While AI continues to develop at pace, and Rydoo progresses both as a business and a technology offering, forming long-term plans for the use of AI is challenging. “Rydoo as an organisation is changing rapidly, so it’s hard to say that within five years we want to be at a certain point with AI because even a year ago I would not have set the goals that we have reached today,” comments Zhakupbekova.

But she notes that many companies may not engage simply because they struggle to measure the ROI of AI. “It’s not always a monetary value that is achieved,” she cautions. As such, she has guided her own finance team to pay more heed to productivity gains. The reduction of the month-end close process from five to three days was, for instance, seen as a promising metric, especially in current circumstances. “It’s a very simple example, but it’s important to start with something small and build momentum,” she advises.

Keeping an eye on progress

As might be expected for a tech-focused team, Rydoo’s own R&D function has already seen significant productivity gains by using AI in-house, notably around coding. But AI is also being incorporated within the contract management platform that Zhakupbekova’s operations team use for its own procurement automation purposes. Rydoo’s sales team has likewise started using AI, enhancing its customer relationship management (CRM) system research, assisting in the production of slide decks, and in the capture and presentation of sales call data.

“There are many more possibilities because AI can touch so many more aspects in a larger organisation,” says Zhakupbekova. With Rydoo still growing, she says “we’re still picking and choosing how we apply it”.

Indeed, she adds: “We’re not interested in how fast we can accelerate from 0 to 100, but we are focused on keeping human expert eyes on our short-term progress, while keeping sight of the mid- to long term goals for ourselves and our customers.” As the corporate cash cycle continues to be buffeted by persistent headwinds, exploring AI in a safe and controlled manner seems like a sensible way to stay on course.