AI in e-invoicing: why the hype can be dangerous in accounting

Every second product announcement in the finance space now leads with AI. Some of it is genuinely useful; some of it is a source of silent data-quality problems. The distinction is not the technology, it is where in the pipeline probabilistic output is allowed.

Great for extraction, risky for arithmetic

Reading a messy supplier PDF and proposing a structured draft is a strong fit for a model: a human reviews the result before anything downstream depends on it, and mistakes are caught early. Computing totals, or asserting that a document conforms to EN 16931, is the opposite case. Here a confident-looking wrong value is more harmful than an obvious failure, because nothing downstream re-checks it.

The validation gate does not guess

This is why eunormia keeps the compliance boundary deterministic. Totals are recalculated server-side from the line items. Every generated invoice must pass a veraPDF check (PDF/A-3) and the EN 16931 business-rule schematron; anything that would not pass is never delivered. No model is asked to believe the result is valid — it is mechanically proven against the standard, or it is rejected.

Used that way, AI is a capable assistant at the edges of the workflow. It just never gets to sign off on correctness.