Kbis extract data extraction

A Kbis extract is the official proof of registration for a French company, issued by the Registre du Commerce et des Sociétés through the local commercial court (or via Infogreffe). Finance teams request it during vendor onboarding, banks and insurers request it during KYB checks, and legal teams file it during contract signature. Anyone who touches French B2B onboarding ends up collecting hundreds of these documents, each with slightly different formatting depending on which greffe issued it and how old the extract is. Automated extraction turns a scanned or downloaded PDF into structured fields you can match against your CRM, your AML screening tool, or your accounts payable master file, without someone retyping a SIREN number by hand.

Fields to extract

Why this document is hard to process

The Kbis is not a single fixed template. Each of the roughly 150 commercial courts (greffes) in France can produce a slightly different layout, and Infogreffe's digital version differs again from the paper-stamped original a company scans and emails. Some extracts run to two pages when a company has multiple establishments or a long history of legal representatives; others fit on one. Older extracts, especially ones scanned from paper archives, arrive as low-resolution images with faded stamps and skewed text, which breaks naive OCR.

Fields also shift position depending on legal form. A SARL extract lists gérants where an SA extract lists a président and a conseil d'administration, and the extraction logic has to know which role labels are legitimate for which form. Dates appear in multiple formats and sometimes in handwritten annotations added by court clerks. Fraud is a real concern too: altered share capital figures or forged registration dates are a known vector in vendor onboarding fraud, so extraction needs to flag inconsistencies rather than just read text at face value. A system trained only on clean digital PDFs will fail on the scanned, stamped, decade-old version that shows up in a real onboarding queue.

How Sygnet processes it

Sygnet first classifies the document as a Kbis extract, distinguishing it from similar registry documents like a K-bis historique or a statuts filing. It then infers the schema based on legal form (SARL, SAS, SA, EURL) rather than applying one rigid template, since field position and terminology vary by form. Every extracted field carries a confidence score and a bounding-box reference back to the source page, so a reviewer can see exactly where "SIREN: 483 927 615" was read from, a form of field-level provenance that matters when a compliance officer has to justify a decision later.

Validation rules run automatically: SIREN checksum, SIRET-SIREN consistency, date logic, and legal form-to-role matching. Where useful, Sygnet supports cross-document checks, comparing the SIREN on the Kbis against the SIREN on an invoice or a RIB submitted by the same vendor, which catches mismatches that manual review often misses. Output comes as clean structured output (JSON), ready to post via webhook into a CRM, KYB workflow, or vendor master file. Human review is reserved for low-confidence fields or fraud-detection flags, not for every document, which is what keeps a high-volume onboarding pipeline moving.

Validation rules worth applying

• Verify the SIREN checksum and confirm SIREN-SIRET consistency
• Cross-check the RCS court name against the list of active French commercial courts
• Confirm the extract issue date is recent enough for your policy (many KYB programs require under 1 or 3 months)
• Flag legal representative names that don't match role conventions for the stated legal form
• Compare the SIREN against an external registry lookup (Infogreffe or INSEE) to catch forged or defunct entities
• Check that share capital format and currency are consistent with the legal form (some forms have minimum capital rules)
• Run cross-document validation against invoices, RIBs, or contracts tied to the same vendor
• Look for inconsistent fonts, misaligned stamps, or resolution artifacts that suggest a digitally altered scan

FAQ

Can Kbis extraction catch forged documents?

Extraction alone reads text; it does not inherently detect forgery. But paired with document fraud detection techniques, such as checking font consistency, stamp placement, and registry cross-lookups, it can flag suspicious extracts for human review rather than passing them through silently, which is the more realistic goal.

How is Kbis extraction different from invoice or receipt extraction?

Kbis extracts have far fewer numeric line items but far more structural variation by legal form and issuing court. There's no total to reconcile like on an invoice or receipt; instead, validation leans on registry checksums and cross-document identity matching, which is closer to the logic used in KYB workflows generally.

Does this replace a live registry lookup?

No. Extraction gets you structured data from the document a vendor actually submitted, but a live SIRENE or Infogreffe lookup confirms the company still exists and matches. The two are complementary: extract first, then verify against the registry as part of your KYC onboarding or KYB process.