Thank you for your message
It has been sent successfully. If you would prefer to book a meeting right away, choose your preferred time.
29 / 09 / 2026
At the 8th CFO Meetup on 24 September, we spent the whole evening talking about AI in finance. The speakers showed what they actually have running in their companies and were open about what hasn’t worked.
AI clearly resonates in financial management too, and the turnout at the 8th meetup showed it. That is why we made this edition more presentation-driven than usual. Jan Gabriel admitted in his opening that he wouldn’t want to give a talk on AI himself. So he invited Dušan Antoš and Michal Polesný, who work with it in finance every day.
The opening round brought a range of expectations. Some people wanted to know where the profession is heading. Others are dealing with staff shortages and data security in their companies. And some CFOs admitted they can’t get enough of Claude and are most afraid of the moment they run out of tokens and are left without it for a few hours.
Jan then shared numbers that sparked the first discussion of the evening. According to an international survey of CFOs, 76% of FP&A work is still done in spreadsheets, more than a year ago. At the same time, two thirds of finance teams now use AI. Someone in the audience suggested the two are linked. People ask Claude to build Excel models for them, so Excel is, paradoxically, making a comeback.
Dušan, CFO at Apify, opened with a question to the room: “Has anyone here already banned the word agent in their team?” He admitted he feels overwhelmed by AI and could use a break from the bubble. Then he went on to show probably the most concrete examples of the whole evening.
His problem has been the same for a while. Apify processes more than a thousand transactions a day, and for three years the team hasn’t managed to shorten the monthly close. The process is documented, but information about what’s done gets lost in emails, Slack threads and hallway conversations. So Dušan had Claude build him a dashboard that now runs on a screen in the finance team’s office. When someone posts on Slack that a reconciliation is finished and tags the company bot, the task ticks itself off on the dashboard. The senior controller no longer has to chase people every day. Dušan calls it version 0.1, and it has only been live for a few days.
What caught the room’s attention even more than the dashboard was how Dušan organises the work around it. The team has written up the whole context of the finance function on GitHub, from how the close runs to who is responsible for what and which files they work with. Each team keeps its own skills there and anyone can run them, even from a phone. He then stated one rule in a way the discussion kept coming back to: “I’ve banned creating any more Excel files.”
That rule came from a specific experience. A colleague in controlling built a neat skill for budget variance analysis, but it only worked for him because it ran on a spreadsheet stored on his personal drive. Dušan rebuilt it so anyone on the team can use it. Every month the skill now goes through tens of thousands of general ledger lines and shows variances down to individual suppliers. Soon it will get a “wall of shame and wall of fame” with the five best and five worst budget owners.
Asked whether he trusts the outputs, he said simply that he checks them. Payroll data never goes into AI, and he still goes through the management report line by line. What Claude reads as a seasonal swing is sometimes just an invoice that was posted incorrectly twice. He also still models plan scenarios in Excel, because he wants to see the formulas.
Michal Polesný followed with a lighter topic, vibe coding. He described himself as naturally lazy, someone who wants things done quickly and to a reasonable standard. That’s why he uses Lovable, a tool where you put a web app together in a chat and don’t have to worry about hosting or security. By his own account, he has never once looked at the code.
His horse racing project Horsebeat got the biggest reaction. When he was looking for sponsors, he had Lovable prepare a separate version of the pitch deck for each of the fifty companies he approached, with an offer tailored to their business.
The most honest story, though, was one that went wrong. Michal asked Lovable to run a security audit of the site, and the tool played it safe by hiding all the horses, because it judged the data about them to be sensitive. For ten days the site didn’t sell the only product it offers. Michal found out only when one of the members stopped him at the races and told him the page was empty.
That led to his main warning. Vibe coding isn’t maintenance-free, and whoever builds an app can easily end up looking after it forever. So Michal gradually brings colleagues into his projects. He also admitted that about 90% of what gets built in Lovable has no long-term value, and it takes courage to shut such projects down.
The discussion after the talk turned to briefs. One participant argued that only about 3% of people could do what Dušan and Michal had shown that evening, and that the business side is generally bad at defining what it wants. Michal agreed and added that this is exactly where finance comes in. Finance should ask what the result is supposed to be and whether the project will bring the company anything, before someone spends weeks on it.
The evening closed with Symfio. Michal Belan and Daniel Kmoch demoed Symfio 2.0 with its AI assistant Conrad a week before the commercial launch, live from the beta.
The demo ran on a fictional group called Karlínfornia Group. Michal asked Conrad what had happened in the group in the first half of the year. Conrad spotted a one-off income of CZK 120 million at the holding company, most likely a dividend, and a Slovak subsidiary that had no revenue for four months. It then found inconsistencies in the account mapping across the companies and suggested fixes. But it didn’t touch the data until Michal had confirmed each change.
Daniel then showed how someone outside finance could work with the numbers. In a few sentences he had Conrad estimate the full-year result, recalculate a scenario with three new design engineers, and finally turn it all into a plan for 2027.
We got the most valuable feedback right there in the room. One participant pointed out that people outside finance can ask Conrad to clarify a report, which makes their conversations about the numbers with the finance team better informed and more precise. For power users, the interesting part is using Conrad for setup. If you only deal with mapping once a quarter, you forget where it’s configured every time, and instead of reading the manual you email support. Conrad would save you that step.
The programme ended after six in the evening. Thanks to everyone who came, and to the speakers for also showing what didn’t work for them. See you at the next CFO Meetup.
By subscribing you accept the Privacy Policy