Can a finance team trust an AI answer if the systems underneath it disagree about what revenue, cost or performance actually mean?
In this episode of AI at Work, I speak with Matej Trbara, co-founder of Farseer, about the foundations required to use artificial intelligence in financial planning and analysis. Finance teams are being asked to forecast faster, model economic uncertainty and provide answers while decisions are being discussed. At the same time, many of those teams continue to reconcile data across ERPs, BI products, warehouses, planning systems and large collections of spreadsheets.
Matej begins by defending Excel. He describes it as a programming platform for people who are not programmers, which helps explain why it remains so deeply embedded in finance. A user can start with a blank sheet, add numbers and formulas, and build a model that reflects the way a business operates. The difficulty begins when that flexibility is stretched across an enterprise. Hundreds of contributors may change structures, formulas and definitions before a finance team attempts to combine everything into one report.
This leads to the central argument of the episode. According to Matej, finance often has a data problem before it has an AI problem. Definitions can drift between systems, actuals require reconciliation and different teams may work from different versions of the financial model. Adding AI to that environment can magnify existing inconsistencies rather than correct them.
Farseer’s approach separates the natural-language interface from the calculation engine. Matej says AI should translate a request, such as asking for a sales variance analysis, into machine instructions. The company’s Rama database then performs deterministic calculations against a governed financial model. AI reasons over those results, while the user can trace an answer through each level and return to the original system of record.
We also discuss how this model changes forecasting and scenario planning. Annual or quarterly forecasts can become outdated quickly when tariffs, energy prices, wars and market conditions affect a business. Matej describes a move toward monthly and continuous planning, where new actuals update the model, responsible people receive a notification and a human either approves the result or adjusts the plan. Farseer estimates that AI-supported workflows could reduce the cost of producing a forecast by between ten and one hundred times. That is a company estimate, but it gives finance leaders a concrete claim to test against their own process.
Another practical example comes from board reporting. Matej says some clients now use Farseer AI Analyst during meetings to answer questions that were not anticipated in the prepared deck. Instead of promising a response several days later, finance can question the live model, inspect suspicious results and continue the decision while everyone is in the room.
The opportunity is faster planning and greater access to financial analysis. The challenge is resisting the temptation to let a language model become an ungoverned calculator. Could your finance team explain where every AI-assisted number came from, and would that answer survive the scrutiny of the board or auditor? Share your thoughts with me at Tech Talks Network.