Tax return, the AI as devil's advocate

Request: „Challenge every item like a tax auditor would.“

1. Rate every item from my own cost register with a traffic light: benefit? evidence? audit risk?
2. Find levers the software never asks about, e.g. the home-office allowance instead of a rejected study, first-ever rental schedule for the let flat.
3. Match receipts from the document archive; a missing receipt yields a red row, not a guess.
4. Have six review agents read the finished draft field by field against 76 form pages.

The tax software only asks what is on the form.

Every call on deductibility, the data entry into the software, and filing.

One €340 item had been claimed twice, caught only in the agent check of the finished draft, not during entry.

A devil's advocate that challenges every item with “benefit? evidence? risk?” plays the same role a good controller plays against a budget.

The AI is not the tax advisor; it is the devil's advocate. For every item it asks: benefit? evidence? audit risk? The figures come from my own cost register, the receipts from the document archive. 356 items were rated one by one with a traffic light. Before filing, six review agents read the draft field by field against 76 form pages and found three genuine errors, including a double deduction. Filed five days ahead of the deadline.