Can AI do the bookkeeping?+
Not on its own, and not with us by default. Connections are set read-only, so the agent analyses, reconciles and prepares the basis but posts nothing. If something is to be written back we require human approval per action, and the approval lands in the log. A mistake in the finance system costs real money, so that is the right order.
Which AI is best for bookkeeping?+
The wrong question to lock yourself into, in our view. The models trade places a couple of times a year, and the one best at reading an invoice is not necessarily best at summarising a year-end close. With Walma you connect the systems once and then run Claude, GPT or Gemini against the same data, so you can switch when something better arrives without redoing the integration work.
Does it work with Fortnox or Visma?+
Stripe, Xero and NetSuite have official servers. Spiris has one in beta, one company at a time. SAP and Business Central go through third parties and need review. For QuickBooks, Pleo, Ramp and Brex we build the connection ourselves.
How does AI in accounting actually work?+
The agent pulls from the source instead of guessing. You ask about overdue invoices, variances against budget or the basis for accruals, and it runs the query against the system and answers with rows it has actually read. Every call is logged with the question asked and what came back, so the answer is traceable the way an audit trail is. That is the difference from pasting an export into a chat.
Where does our financial data end up?+
The gateway runs in your own Azure tenant in an EU region, so prompts, arguments and logs are processed there. If a vendor's server sits outside the EU we tell you before you approve the connection.