Automated KYB: From Articles of Incorporation to Structured JSON with AI
Know Your Business (KYB) is the process by which a company verifies the identity and legitimacy of another company before doing business with it. It's mandatory in regulated sectors like fintech, banking, and insurance, and increasingly common in any B2B relationship.
The core of KYB is reviewing corporate documents: articles of incorporation, powers of attorney, legal representative IDs, and tax certificates.
The manual problem
In Mexico, articles of incorporation can be 80 or more pages long. Manually extracting relevant data — company name, entity type, corporate purpose, share capital, shareholder names, ownership percentages, representative powers, tax ID — takes hours of analyst work.
Multiply that by dozens or hundreds of corporate clients per month and you have an operational bottleneck that delays onboarding, generates data entry errors, and consumes resources that could be dedicated to actual risk analysis.
The difference between Mexican and US documents
Articles of incorporation in the United States are relatively short and standardized documents. Each state has its format, but the structure is predictable: corporation name, purpose, authorized shares, registered agent. They rarely exceed 10 pages.
Mexican corporate documents are a different story. Articles of incorporation include the full transcript of the constitutional assembly, with notarial formalities, identification data of all participants, complete bylaws, and frequently subsequent protocolized amendments. The result: dense, long documents with variable format between notary offices.
How automated extraction works
The system processes the document in three stages. First, specialized OCR converts the scanned document into text, handling stamps, signatures, and notarial typography. Second, a language model analyzes the text and extracts relevant fields. Third, the data is delivered in structured JSON format, ready for integration with any compliance system.
Processing a US document takes minutes. An 80-page Mexican acta constitutiva may take 40-60 minutes — significantly more, but compared to hours of manual work, the efficiency is clear.
What gets extracted
Standard fields include company name and entity type, tax ID (RFC in Mexico or EIN in USA), corporate purpose, share capital (fixed and variable), list of shareholders with percentages, governance bodies, legal representative powers, registered address, and entity duration.
The output is clean JSON that any system can consume via API.
Fullkro by Leeuwwolk processes corporate documents from Mexico and the United States. Fine-tuned OCR, AI extraction, JSON output. No external APIs — your client's documents never leave your server.