Retail, business and institutional

A bank accepts more customer evidence than anyone, through more channels

Cheque images, invoices, payslips, dispute evidence, collateral photos, onboarding selfies. Every channel that accepts a document is a place fabricated evidence can enter, and generative AI now produces all of it in minutes.

The problem

The systems read it perfectly. That is the problem

SURFACE

Six doors, one lock missing

Lending origination, KYC, invoice and trade finance, remote deposit, disputes, and asset-backed lending all accept customer-supplied evidence. Each is assessed by a different team on a different system, and none of them can establish where a file came from.

ARTEFACTS

Generated, not edited

Forensic tools look for the traces of editing. A document produced whole by a model has none to find - it is internally consistent, correctly formatted, and extracts cleanly. The checks pass because there is nothing for them to catch.

TIMING

The loss lands months later

It rarely appears in a fraud column. It appears as a credit loss at first default, a covered cheque, or an advance that cannot be recovered - by which time the money is gone and the document is old news.

How CapSeal applies

Move the trust decision to the moment of capture

SUBMIT

Sealed at submission

The customer captures through the channel they already use. Whether a real lens saw a real document is measured on the device and bound into the file before it reaches you.

PROCESS

One verdict, every product

The same call serves origination, onboarding, invoice finance and disputes. One integration rather than six, and one basis your teams learn once.

ACT

Stop it before the money moves

Advances, deposits and approvals proceed on evidence that settled. Anything else reaches a human with the report attached, before funds leave.

Inside your workflow

Where CapSeal sits in the bank's evidence flow

CapSeal does not replace your processing - it adds two touchpoints to the channels you already run. Evidence is sealed at submission, a verdict attaches during processing, and each file routes down one of three paths. Every outcome lands back in your core banking or origination systems with the proof stored on the file.

YOUR PROCESS • STAGE 1 Evidence submitted Customer or business submits documents in your banking channels CAPSEAL: sealed at submission YOUR PROCESS • STAGE 2 Processing Handled in your core banking or origination systems as normal CAPSEAL: verdict attached to file ATTESTED-GENUINE Fast-track Straight-through processing - advance, deposit or approval UNVERIFIED Normal checks Your existing controls, unchanged. Never auto-declined. TAMPERED / SYNTHETIC Held for fraud team Advance or payment stopped; fraud team gets the file with proof YOUR PROCESS • STAGE 3 Action & audit Every outcome lands back in your systems. The proof object is stored with the file - replayable for AML review, dispute, internal audit or regulator, years later, without calling CapSeal. INTEGRATION FOOTPRINT Runs beside core banking, origination and dispute platforms - an SDK at submission, one API call in processing. No core migration. Branch, mail and non-digital channels simply route as unverified.

Note the middle path. Evidence submitted by email, post or in branch routes to your existing controls exactly as it does today - never auto-declined, and no customer is worse off for having used a channel that cannot seal.

What this does not do

Where the evidence stops

CapSeal establishes that a document or photograph reaching the bank was captured by a real device from a real thing, at a known moment, and has not been altered since.

  • It does not establish that a genuine invoice is for work actually performed, or that a real payslip states a real salary. A truthful-looking document that is simply untrue is beyond any provenance tool.
  • It does not detect first-party fraud where the customer submits real evidence of a real event and misrepresents what it means.
  • It sits beneath your IDV vendor rather than replacing it: it attests the capture, it does not match the face.
  • Evidence arriving by email, post or branch carries no provenance. That is common, it is not suspicious, and it routes to your existing controls unchanged.

The full position is in the Product Disclosure. We would rather you knew the boundary before you bought than after.

Industry brief

The banking numbers, with every input cited

A bottom-up model of one product line - a bank financing 60,000 invoices a year at typical 80 to 90 per cent advance rates, the product where a single fake document converts to a near-total loss. Seven pages, every assumption on the page.

$20.7M
advance losses avoided per year, invoice finance alone
$26.6B
global cheque fraud losses in a single year
88%
of banks avoid invoice financing over document authenticity
~$400M
remote-deposit-capture fraud in a single year

The honest boundary, stated in the brief itself: the model sizes invoice and trade finance only, which is one of six entry points and deliberately the narrowest. It does not add lending origination, deposits, disputes or onboarding, and it does not claim fraud where the document is real and the story behind it is not. The figures are modelled from published ranges, not results from a named bank.

Banking brief (PDF)
Integration

Three things to wire up

  1. Capture Drop the iOS or Android SDK into the flow your users already complete. Our capture-integrity engine measures the scene on the device and seals the result before it leaves the handset.
  2. Verify One authenticated call at intake returns a verdict, a basis and a proof object. See the SDK.
  3. Route Automate on proven and declared. Send everything else to a human with the report attached.

Try the verification half right now, with no signup, at capseal.ai/check.

Start with the product where a fake document costs you most.

Invoice finance is the usual entry point. One line, one API call, measured against your own loss rate.

Book a pilot Check a file first