KYC automation: structured fields, not raw text
A classic OCR returns text. A KYC file needs fields: surname, first names, date of birth, document number and expiry date, address, account number, company number, directors. Koncile relies on one template per document type, with over 99% accuracy on key fields, in several languages, on PDFs as on photos taken at an angle. Korint runs very different documents through the same flow: ID card, bank details, private hire licence, driving licence, no claims statement, vehicle registration, company extract, lease and photos.
KYC document verification: the customer's name on every document
The most profitable check is also the simplest: verify that each document is in the name of the person signing up. At Meelo, a French credit fintech, Koncile reads the supporting documents at onboarding (address, pay stubs, tax documents, bank details), checks they are in the customer's name and extracts the account number that will serve for the direct debit mandate. A document in a third party's name comes up as an anomaly, with the document concerned. The analyst decides: a partner hosting the applicant can be legitimate, a stranger much less so.
Alongside Onfido, Veriff, IDnow or Sumsub
These solutions verify the person: selfie, liveness, face compared with the photo, chip read when there is one. Koncile verifies the file: authentic or edited, issued or not by the issuer it displays. The two complement each other. The typical flow hands identity and selfie to the biometric provider, then sends Koncile the ID document for a second opinion on edits, and every other document in the file. 1stKYC, a KYC platform, integrates Koncile through the API inside its own product.
A readable reason, test by test
A compliance team must be able to explain a refusal, to the customer as to the regulator. 1stKYC set that requirement from day one: a fraud reason that can be explained and defended. Koncile returns a score from 0 to 100 and a reason for every test. Calibration matters as much: documents modified a few minutes after their creation by a scanning app were coming out on alert, when that is the normal behaviour of those apps. The check was recalibrated, not bypassed.
AI based KYC and AI for AML: what each model does
Three different jobs hide behind the words. AI identity verification is the biometric provider's work: matching a live face with the photo on the ID. KYC AI, as Koncile practises it, is file analysis: reading the fields of every document and judging whether the file was edited, generated or produced by another software than its issuer's. AML AI, finally, watches transactions and screens names against sanctions lists once the relationship exists; Koncile does not do that part. What it feeds into your AML tooling is cleaner: income and source of funds documents already read, recalculated and scored, with the reason attached.