Forged, manipulated and AI-generated invoices, bank statements and contracts are becoming harder to recognise. Nio is a system designed to read a document’s forensics, its content and its context — return a calibrated risk score, and explain the decision in terms an auditor can verify. Intended for delivery as a B2B API, inside your workflow.
Currently in development · Not yet available
01 — The problem
Forensic and registry signals routinely disagree. A name differs by legal form or diacritics. A registry address is stale. An IBAN is new — but legitimate. Deciding when that’s normal business change and when it’s an attack — and explaining that decision to an auditor — is the unsolved part.
That decision is what Nio is built for.
02 — The output
Not a verdict.
A reasoned decision.
The design goal: a document arrives, and its score comes back with the work shown — every signal, every contradiction, every reason — ready for an analyst and verifiable by an auditor. The illustration shows the kind of output Nio is being built to return.
How it worksinvoice_2026-0912.pdf
Analysing…
Illustration — the explanation output Nio is designed to return.
03 — What Nio is designed to return
A risk score
A calibrated score per document — not a black-box yes/no, but a measure your policy thresholds could act on.
Forensic signals
File structure, metadata, fonts, compression traces — what changed after the document was made.
Contextual signals
Registry records, IBAN changes, legal forms, addresses — does the document’s story match the world outside it?
An explanation
Every score is to ship with its reasoning, written for a human reviewer. Explainability is the output — not an add-on.
04 — Who it’s for
Built for decisions that rest on documents. Factoring companies, lenders, financial institutions, audit and accounting firms, ERP platform vendors — API-first, inside the workflow you already run.
Calibrated for the Croatian stack
Diacritics, legal forms, registry lag, domestic formats — the details generic tools miss.
05 — The state of the art
Off-the-shelf detection barely beats a coin flip.
0
DocTamper AUC on AI-forged docs — Wu et al., 2026
0
Zero-shot GPT-4o, same benchmark. Chance = 0.5
0
Best multimodal LLM — Liang et al., 2025
0
Methods tested; none reliable out of the box — Zhao et al., 2026
06 — In its lane
Scores. Explains.
Never accuses.
Nio is designed to measure and explain — never to block payments or make accusations. The score and its evidence go to your analyst; the judgment stays where it belongs: with people.
- Never a black box
- Never an automatic rejection
- Every score comes with its reasons
07 — Built for trust
Our commitmentsExplainable by design
Every decision is written out, signal by signal, in language an auditor can follow. If we can’t explain it, we don’t ship it.
A full audit trail
Every check, every score, every version — logged and reproducible. Nothing happens off the record.
API-first, data-minimal
We process what verification needs and nothing more. Your documents are never sold or shared.
Accessible by commitment
The interface that shows scores and explanations is built to WCAG 2.2 AA — verified per criterion, not asserted.
08 — The intended integration
01
Connect
Designed as one REST API: documents sent from your ERP, loan system or back office — no core replacement, no IT project.
02
Score
Each document is to come back with a risk score and its full reasoning — forensic, content and contextual signals together.
03
Decide
Analysts act on explained evidence, and the audit trail writes itself.
09 — Why now
Documents used to be evidence. Generative tools ended that — a convincing forgery now costs a cent and takes a second. We’re building the verification layer that lets a lender trust a PDF again.
Currently in development. Not yet available.