Dr. Veronika von Heise-Rotenburg is familiar with both worlds: traditional banking and the dynamic startup scene. As CFO of the German scale-up Everphone and a member of the supervisory board of the publicly traded Austrian bank Bawag, she spoke at an online event hosted by the AI Future Council’s on artificial intelligence in the finance function. In it, she makes a statement that really catches your attention: Anyone who still believes that AI in accounting is a thing of the future has already missed the boat.
As early as the 2022/2023 winter break, shortly after the launch of ChatGPT, von Heise-Rotenburg immersed herself in MIT courses on generative AI. By January 2023, Everphone was already using an AI model to generate management forecasts for its annual financial statements. She describes this early engagement as part of what’s known as a “K-shaped split”: A minority began early on to understand, shape, and integrate AI into their own processes and is thus on the “upward trajectory.” The majority, on the other hand, remains stuck in the old ways or, at best, uses off-the-shelf tools without truly grasping them—thereby reducing their knowledge and market value. According to a study she cites, just three percent of finance teams feel well prepared for the AI transformation; in Germany, only thirteen percent actively use AI tools. The reason, she says, is rarely a lack of will, but rather a perceived skills gap that must be closed urgently.
Why Finance Is the Ideal Playing Field
For von Heise-Rotenburg, there is no doubt: hardly any other corporate function is as data-intensive, as rule-based, and as dependent on accuracy as finance. Monthly closings, payment runs, reconciliations, all of these follow recurring patterns that lend themselves perfectly to automation. Added to this is the desire for real-time analysis, because no one wants to wait until the end of the month to find out whether a product or customer group is profitable. Ultimately, as with most back-office functions, it’s also about efficiency and cost savings. At the same time, she warns against blindly jumping on the bandwagon: Not every tool labeled “AI” delivers on its promises. Planning software, for example, has not yet convinced her in its current form, while specialized solutions for audit and compliance have already delivered solid results.
From residual value forecasts to your own chatbot
Heise-Rotenburg provides specific examples from her own company. Since 2017, the residual value forecast for smartphones leased by Everphone has been based on machine learning—specifically, the Random Forest model, which has now processed more than 400,000 completed lease agreements, long before generative AI became a topic of discussion. The internal chatbot Eesel, in turn, which is connected to Slack, Confluence, Google Drive, and various CRM systems, not only provides answers but also cites sources for verification, and, according to her, is used by 99 percent of the workforce. For the monthly variance analysis, she relies on a custom-built Gemini system that compares actual and budgeted figures and identifies the causes.
AI is also used when reviewing investment memoranda in the context of potential acquisitions: A system trained on company data compares incoming offers with the company’s strategy and expectations and independently suggests due diligence questions. In a test run with about seventy questions, von Heise-Rotenburg himself would have added only three and deleted five. The system is supplemented by in-house tools built on platforms such as Lovable or Claude Code. There are now over two hundred applications in use among nearly two hundred employees, ranging from pay equity analysis to an automated reminder system for pending invoice approvals.
The Downside: Deepfakes and the New CFO Scam
As clearly as the opportunities are outlined, the warning is just as clear. Von Heise-Rotenburg reports on the first documented case in the press in which a treasury manager was tricked via a deepfake video call into authorizing a wire transfer, with the voice and image supposedly coming from the company’s own CEO. In the past, such fraud attempts could be identified by poorly worded emails; today, systems mimic not only speech but also individual communication patterns.
During the discussion, event director Katrin J. Yuan who herself works on simulations of such attacks, confirmed just how much the combination of voice and personality analysis based on publicly available data increases the effectiveness of such attacks. Added to this is our growing dependence on technology: What happens if a key AI tool fails or providers drastically raise their prices? MIT studies show that the brain becomes accustomed to using AI. Reducing this dependence (“cognitive offloading”) would not be possible without some loss of efficiency, even with practice.
Why the CFO’s Job Will Remain
Despite all the automation, von Heise-Rotenburg does not see her own role as being threatened. What AI cannot do, she says, is exercise human judgment: selecting realistic scenarios, weighing geopolitical and macroeconomic risks, and taking responsibility toward the executive board and investors. A machine can provide a number, but it cannot provide the reasoning behind it or the personal trust that a leader builds in the boardroom. That is precisely why she advises strengthening both technical and human skills together: those who combine the two will be among the winners in the coming years.
Culture Trumps Control
Her most important advice to companies that are still hesitant: Don’t ban the use of AI, shape it instead. Companies that fail to provide employees with a legitimate framework risk having sensitive data end up in private AI applications without any oversight. At Everphone, they initially conducted a six-month pilot program with one person per department and a small but flexible budget, based on the principle that anything not explicitly prohibited is allowed. This pilot led to a publicly available AI policy as well as a tiered training program that accommodates different learning paces. She sums up her conclusion simply: AI won’t replace anyone. However, people who use AI will replace those who do not, and this applies particularly to Finance and Legal.
Incidentally, von Heise-Rotenburg, as editor of the book Financial Management in Startups and Scale-ups, has devoted an entire chapter to the topic of “AI in Finance” together with Nicolas Boucher and Stephanie Palero. It will be published in July 2026 by Schäffer-Poeschl Verlag, which previously published “AI for Executives,” featuring a guest contribution by Katrin J. Yuan.
See also: From Research to the Market and From Earth to the Moon
Search:
Sponsors:
