Verification of supporting documents
The documents sent to you were not produced by you.
A care invoice, a till receipt, a certificate, an expense claim, the photo of a barcode or a damaged item: these documents are produced, altered and uploaded by third parties. And you pay on their appearance alone.
Detection of alterations and reuse, calibrated on your own specification.
Document fraud watch · 9 September 2026
The fake receipt has become a €9 monthly subscription.
Receipt generators are sold online, with the right template and the right VAT number. Human checking is out of the game.
Nine euros a month to produce the document. At the other end, what it releases on your side is counted in refunds, settlements and grants paid out.
01The contextWhat has changed
The fake document is no longer amateur work.
Retouching a PDF no longer requires any particular skill. Receipt generators are sold online, by subscription, with the right issuer template and the right VAT number. What used to be isolated fraud is now tooling.
On the other side, human checking is wholly ineffective. A case handler decides over a far longer span, on a screen, with no way to compare the document against the thousands already received. The same proof of purchase serves ten claims as long as nobody looks at portfolio scale.
The principle
There is no universal detector. So there is no generic VigiProof.
A dental quote, a supermarket receipt, a supplier invoice and an insurance certificate have nothing in common: no shared issuers, templates, fields, validity rules or grounds for refusal. A model trained for everyone gets it wrong for each.
VigiProof is therefore never delivered off the shelf. Each deployment is built on the client's specification, and its engine learns on the client's corpus. That is what makes it accurate, and what makes it improve.
02ImplementationFour steps
How a deployment is built.
01
Framing
We start from your specification: which documents you receive, which issuers are legitimate, which rules make a document acceptable, which grounds for refusal must remain enforceable.
02
Corpus
Your real documents, genuine and fraudulent alike, form the training set. It is your own material that trains the engine, not a generic dataset built for someone else.
03
Calibration
Thresholds are set to your tolerance: what you want blocked, what you only want flagged, what you accept letting through so as not to slow your flow.
04
Continuous learning
Every decision your handlers make, assisted by the agent, feeds back into the engine. Detection improves with your portfolio instead of ageing as the fakes get better.
03The engineA cluster, not a single signal
What the machine looks at, and nobody has time to look at.
File integrity
Retouching, residual layers, generation inconsistencies, metadata that does not match the claimed issuer.
Document consistency
Issuer, numbering, amounts, dates, mandatory wording: what a handler would check if they had time to check everything.
Reuse
The same document, the same number, the same image, resurfacing in another case; including under another name and months apart.
The cluster
No single signal is enough. It is their conjunction that grounds the opinion, and that makes it defensible.
VigiProof does not replace a case handler. It presents a reasoned opinion and the evidence behind it. The machine becomes autonomous once it has acquired the competence, but the decision can remain human, and it stays traced: how far you go is your call.
04The groundWherever a document triggers a payment
Wherever a document triggers a payment, a refund, a credit note.
Health insurers and top-up schemes
Care quotes and invoices, coverage requests, document-based reimbursements.
Online retailers
Proofs of purchase, return requests, damaged parcels, disputes and goodwill gestures.
Promotional platforms and rebates
Receipts, barcodes and proofs of purchase: the ground most exposed to industrialised reuse.
Accounting SaaS
Invoices, receipts and expense claims uploaded by end users.
Insurers
Claim declarations, proof of loss, third-party certificates.
Anyone who requires an original
Landlords, lenders, social bodies, human resources departments.
05Some applicationsWho receives documents, and what they trigger
Nine segments, one shared blind spot.
| Segment | What is uploaded | What the document triggers | Why it is a blind spot |
|---|---|---|---|
| Promotions, cashback and rebates | Till receipt, barcode, form, bank details | Deferred refund, cashback, voucher | The same receipt serves ten claims before anyone notices. |
| E-commerce, after-sales and warranties | Order, invoice, photo of the product or parcel, return slip | Refund, replacement, covered repair | The photo of the broken item travels from one dispute to the next; the invoice is retouched in two minutes. |
| Delivery, freight and travel | Costed claim, ticket, reservations, invoices for expenses incurred | Compensation, credit note, expense refund | The loss is declared on paper, and the side costs are backed by receipts nobody cross-checks. |
| Property insurance, mobility and rental | Report, inventory, quote, invoice, serial number or IMEI | Compensation, covered repairs, deposit withheld | A favour quote and an item already compensated look very much alike, except at portfolio scale. |
| Health, protection and assistance | Paid invoice, care form, quote, prescription | Health cost reimbursement, coverage | A paid invoice for care never delivered passes every formal check. |
| Employers, expense claims and works councils | Receipt, expense claim, named invoice, proof of payment | Employee reimbursement, grant, social benefit | A receipt is generated online, and the same expense is claimed twice at two desks. |
| Supplier purchasing and payments | Invoice, delivery note, order, bank details | Supplier payment, credit note, bank recredit | The fake invoice and the changed bank details are the same attack: a document taken at its word. |
| Grants, premiums and subsidies | Quote, paid invoice, completion certificate, proof of status | Premium, subsidy, allowance, final grant payment | The completion certificate arrives before the work; the final payment leaves straight away. |
| Platforms, delegates and third-party verifiers | Their clients' documents, checked on their behalf | Approval of a refund, payment for an assignment | They process for others: a single deployment covers their whole book of clients. |
06The benefitsWhat the system changes
Clean cases move fast. The others are reasoned.
Clean cases move fast
Human review focuses on the cases that deserve it. Average processing time falls while control is substantially increased.
Reuse becomes visible
A recycled proof of purchase is never visible at case level. It is visible at portfolio level.
A refusal is reasoned
Every opinion rests on documented elements, available if the decision is challenged.
The system does not age
Fakes improve; so does the engine, because it learns on your cases and not on a frozen model.
A deployment is scoped before it is priced.
Set-up covers framing, building the corpus and calibration. That is what separates an accurate engine from a generic detector; and it is the one part of the project that cannot be ordered from a catalogue.
07PricingIndexed on document volume
A package indexed on the volume of documents checked.
| Documents per year | Year 1 package | Price per document | Following years | Discount |
|---|---|---|---|---|
| − de 5 000 | 2 400 € | 0,4850 € | 1 680 € | −30 % |
| − de 10 000 | 3 900 € | 0,4050 € | 2 652 € | −32 % |
| − de 20 000 | 5 840 € | 0,3340 € | 3 913 € | −33 % |
| − de 40 000 | 9 960 € | 0,3070 € | 6 574 € | −34 % |
| − de 70 000 | 13 870 € | 0,2410 € | 8 738 € | −37 % |
| − de 150 000 | 19 880 € | 0,1630 € | 11 928 € | −40 % |
| − de 300 000 | 25 870 € | 0,0996 € | 14 746 € | −43 % |
| − de 1 000 000 | 62 440 € | 0,0712 € | 33 718 € | −46 % |
| − de 2 000 000 | 83 600 € | 0,0452 € | 43 472 € | −48 % |
| Au-delà de 2 000 000 | sur devis | sur devis | sur devis | −50 % |
Annual prices excluding tax. The package covers the volume of its band; the per-document price applies with no volume commitment. Framing, corpus building and calibration are covered by a set-up from €400, based on your specification. The “following years” price applies from renewal.