Invoice Guardian: AI That Checks Every Detail

A German company asked us to build an internal AI that reads invoices and checks them against German tax rules. The AI categorises items, applies the tax, and flags errors, missing items, wrong tax numbers and missing mandatory sections. The rules stay current on their own: whenever German tax rules change, the system reads the new rules from government sites and updates itself.

• Client: a German company, name withheld

• Work: internal AI invoice checker

• Knowledge: a RAG base built from German government taxation sources

• Outcome: errors flagged before invoices go out

The challenge

German tax rules change when budgets pass and new rules come in. Every invoice must carry the mandatory tax details, and a mistake means rejected paperwork and tax risk. Checking invoices by hand is slow, and tired eyes miss things. The company needed every invoice checked against the current rules, every time, without adding headcount.
A mistake on an invoice is not a small thing. A wrong tax number or a missing mandatory section can mean rejected paperwork, a tax dispute, or a fine, and the error is usually found weeks after the invoice went out, when it costs the most to fix. The company needed the check to happen before the invoice left the building, not after.

What we built

We built an internal AI that reads each invoice and checks it against the current German tax rules. The AI categorises the items on the invoice, applies the tax, and flags what is wrong: errors in the invoice, missing items, wrong tax numbers, and mandatory tax sections that are missing. The system checks the latest law and applies it to each invoice. If anything is missing or misinterpreted under the law, it raises a flag. The rules stay current because the system crawls government taxation sites whenever the rules change or a new budget passes, on the day or within a week, and the RAG knowledge base updates itself, so the new rules apply automatically.
Every invoice goes through the same check, in the same order. The AI reads the invoice, sorts the items into categories, applies the tax, and compares the result with the law as it stands today. When something is wrong, the flag says what is wrong: the error, the missing item, the wrong tax number, or the section the law requires and the invoice does not have. The team sees the problem and the reason in one place, and can fix it before the invoice goes out.

The outcome

Mistakes are caught before invoices go out. Tax numbers and mandatory sections are verified on every invoice. The company spends its time fixing real issues instead of hunting for them. The law stays current without anyone in the company reading tax bulletins. When the rules change, the system reads the new rules and updates its knowledge, so the next invoice is checked against the new law, not the old one. The company gets the same careful check on every invoice, whether it is the first of the day or the hundredth.