Review KYC and AML files at volume
Point Hexagone AI at a batch of onboarding files. The AI cross-checks documents, flags inconsistencies and gaps, while identity papers stay on your machines.
Onboarding review is a queue that never empties, and the reviewer is the bottleneck. It is also the single worst candidate for a cloud AI tool: a KYC file holds identity documents, proof of address, beneficial ownership chains and source-of-funds evidence, covered at once by GDPR, banking secrecy and your sectoral obligations. Most files also arrive as scans, so tools that only read digital text quietly miss the passport page entirely.
How it works
Protect the batch, scans included
Point Hexagone AI at the folder. Local OCR reads image-only PDFs, so photographed ID pages and scanned utility bills are covered like native documents.
Cross-check the file
Ask for inconsistencies between declared and documented information, missing pieces, dates that do not line up, ownership chains that stop short of a natural person.
Keep the evidence trail
Every detection and substitution is logged locally and exports as CSV, which is what a regulator or an internal audit will ask to see when you explain how the analysis was run.
List the inconsistencies between the supporting documents in [c.KYC_2026] and the declared beneficial owner.
What gets masked here
- Client names, dates and places of birth
- Passport, ID card and residence permit numbers
- Home addresses and proof-of-address details
- IBANs, account numbers and tax identifiers
- Beneficial owners and shareholding chains
- Transaction amounts and counterparties
What you get back
A first-pass review across the backlog that reaches the analyst already annotated, so human judgement is spent on the files that actually need it rather than on the ones that were complete all along.
How accurate is the detection?Questions on this use case
- If names are masked, can the AI still link the same person across files?
- Yes, and this is the part that makes the exercise worth doing. The mapping is held for the whole folder rather than per document, so one individual keeps one placeholder across every file in the batch. That is exactly what lets the model notice that a beneficial owner in one file is a declined applicant in another. Masking each document independently, which is what a generic tool does, destroys precisely the signal an AML review is looking for.
- Our policy forbids any client data leaving our network. Does that rule this out?
- No, that is the deployment it was built for. The desktop app processes everything locally and works offline after installation, and the Enterprise plan deploys on-premise or fully air-gapped with no outbound network at all, with SSO and a command-line interface for batch pipelines. Because nothing is transmitted, there is no transfer to a third party and no processing agreement to negotiate for the local product.
Try it on this exact task.
One week free, no credit card. Nothing is uploaded, so you can run it on a real file without asking anyone's permission.
macOS Apple Silicon and Windows 10/11