‘We want to make bookkeeping less painful’: How Visma and its AI portfolio are reshaping accounting

Visma isn't chasing headlines with some flashy AI chatbot. The Norwegian software group, which sells accounting, payroll and tax tools to small and mid-sized businesses across Europe, has quietly built more than 600 AI initiatives across its portfolio, splitting them into early experiments and things ready to scale group-wide. That's the kind of number that only makes sense once you realize Visma owns dozens of local accounting software companies, each solving a slightly different version of the same problem: too many humans retyping invoices.

CTO T. Alexander Lystad breaks the value into four buckets: automation that saves time, like bank reconciliation; assurance, like catching payroll anomalies; advisory, surfacing patterns a human should look at; and agentic, meaning entire workflows run on autopilot. It's a tidy framework, and it maps almost exactly onto what two of Visma's portfolio leads are actually building. At Chaintrust, managing director Mikael Gandon says the whole project started because he watched accountants retyping supplier invoices into ledgers at night. Now the software scans, extracts and posts those entries automatically, and junior staff spend their time reviewing rather than typing.

Dinero, a Danish accounting platform also under Visma, is going further with something it calls a virtual CFO. It hooks into a company's bank account via open banking, handles routine bookkeeping and payments, and can flag things like an invoice likely to be paid late or a business leaning too hard on one client. Managing director Martin Thorborg frames it as leveling the field for small companies that can't afford an actual finance officer. Lystad's version of the same ambition: get agents keeping the books current enough that closing season becomes a formality instead of a scramble.

What's notable is how conservative these companies are about where generative AI is actually allowed to touch the numbers. Dinero uses large language models for advisory features but sticks to machine learning and deterministic algorithms for core bookkeeping, specifically because those can't hallucinate. Thorborg says Dinero runs roughly 20,000 tests on every update, and argues that being aggressive with AI accuracy claims is cheap in a demo and expensive in real client trust. Gandon makes a similar point about metrics: claiming 95% lab accuracy is easy, but the number that matters is how many entries get processed without a human ever touching them.

The bigger claim from Lystad, and it's a reasonable one, is that AI in accounting will eventually stop being a talking point altogether and just become table stakes, baked into every product. If that happens the way Gandon describes, accountants stop being data-entry clerks and become the people interpreting the numbers, handling messy human judgment calls like restructurings or bad-news conversations that no algorithm should be trusted with.