The fraud firewall
Catch the fraud before the money moves.
Advisory, never blocking — releasing money is always a human act.
Every signal is deterministic and explainable — no ML black box.
Three tiers of deterministic screening stand between an invoice and a payment — each one a set of rules a person can read and an auditor can question. What follows is the shape of the shield, not its blueprints: the full detector matrix is walked through in a demo, not published.
Even a born-clean invoice is corroborated — not trusted.
Pixels, arithmetic and conformance can all pass and the document can still be fabricated — an AI can generate a clean one. So when an invoice checks out on its face, Aurenta doesn't stop at the image: it corroborates the claim against out-of-document truth — public registries, your own buyer feeds, and your payment history. A first-time vendor paid into a first-time bank account raises an advisory REVIEW, fused with the checks below. Advisory, never blocking — releasing money is always a human act. Every signal is deterministic and explainable — no ML black box.
Three-way PO / GRN match
Each invoice is reconciled against your purchase-order and goods-receipt feed. A fabricated PO reference, billing beyond the authorised PO total, or billing before goods are received all surface — deterministic and advisory.
TRN-registry existence check
A seller's tax-registration number can be perfectly well-formed and still not exist. Aurenta checks it against the registry: format-valid is not the same as real.
Cross-tenant mule detection
One payee account quietly collecting for several unrelated sellers — across independent customers — is caught with a privacy-preserving one-way fingerprint. The account number itself is never stored.
First-seen payee prior
A first-time vendor paid into a first-time account is the classic set-up for a redirected payment. It raises an advisory REVIEW so a person looks before the money moves.
Tier 1 Per document
Every invoice is examined on its own before anything else sees it: the figures are recomputed exactly, the identifiers are checked for tampering and disguise, and the document's own story — dates, sequences, structure — has to hold together. An invoice that lies about itself never reaches a payment run.
Tier 2 Behavioural
Each document is then judged against history — what this vendor normally does, how invoices usually arrive, and what changed since the last time money moved. The quiet deviations that precede most payment fraud surface for a person to look at, with the reason written out.
Tier 3 Network & file
The widest lens looks across relationships and at the files themselves: patterns that only appear when your whole trading graph is considered at once, and the evidence a document carries about its own history. What no single invoice reveals, the network does.
The Payment-Run Firewall
Screen the payment proposal CSV thirty minutes before money moves — a per-line pay/hold recommendation, sealed as evidence.
It is the category most competitors don't have: not another approval step, but a deterministic screen at the exact moment the loss becomes irreversible. Advisory, never blocking — releasing money is always a human act.
On a 36-case adversarial benchmark — including evasions authored against our own detector matrix — precision 1.00, recall 1.00, zero false positives on look-alike legitimate controls. Detector verification on synthetic cases, not a field rate.
See your own exposure first.
The same detectors that guard live invoices can comb your history — ranked, evidenced, free to run.