API for extracting requisites: 99% accuracy and integration with 1C

# Automating Requisite Extraction: How to Reduce Errors in Documents by 99%

Manually entering INN, KPP, and bank accounts from PDF or DOCX files is a straight path to errors, stalled payments, and fines. A third of all documents contain inaccuracies. The solution is contextual NER parsing with checksum validation. Below is a technical description of the API, benchmarks, and an example for 1C.

How Automatic Requisite Extraction Works

The process starts with uploading a text document. Supported formats include PDF (with text layer), DOCX, DOC, TXT, RTF, and HTML. Scanned images without OCR are not processed—the system works only with machine-readable text. After receiving the file, a pipeline of five stages launches:

• Text Extraction—uses an internal parser for each format, including recovery of encodings like CP1251 without BOM.
• Normalization—fixing word breaks, merging hyphenations, cleaning up layout artifacts.
• NER Recognition—proprietary model identifies legally significant entities: INN, KPP, OGRN, bank account, BIC, bank name.
• Checksum Validation—each field is checked against Federal Tax Service and Bank of Russia algorithms. Invalid data is not returned.
• Structured Response Formation—JSON with clean fields ready for import into 1C, CRM, or ERP.

The system doesn't rely on simple regular expressions. Instead, it uses contextual analysis: for example, "bank account" should be near the bank name, and INN in the counterparty details block. This reduces false positives.

Accuracy and Performance: Real Numbers

On a test sample of 10,000 documents, the following metrics were achieved:

• INN / OGRN: 99.7%—errors occur only with extremely distorted layouts.
• Bank Account: 98.9%—main cause of errors is merged digits in poorly formatted text.
• BIC: 99.5%—thanks to BIC's strict structure and check digit.

Processing speed depends on format and size:

• Light DOCX (2 pages): ~1.5 seconds.
• Heavy PDF with graphics: up to 45 seconds.
• Maximum file size: 2 MB.
• Request timeout: 120 seconds.

Limitations stem from the need for deep document structure analysis. For high-load systems, asynchronous processing via webhook is recommended.

Integration with 1C: Practical Example

For accountants and developers working with 1C, a ready-made processing module is available. Below is a simplified API call code from 1C Accounting 3.0:

Function IzvlechRekvizityIzFayla(FilePath)    
  URL = "https://api.sb-poisk.ru/v1/extract";    
  Headers = New Withotvetstvie;    
  Headers.Insert("X-API-Key", "VASh_KLYuCh");        
  FormData = New Withtruktura;    
  FormData.Insert("file", New DvoichnyeDannye(FilePath));
  HTTPConnection = New HTTPConnection(URL);
  Answer = HTTPConnection.SendForObrabotki("POST", Headers, FormData);
  Result = ProchitatJSON(Answer.PoluchitTeloKakWithtroku());        
  Return Result;
EndFunction

The output is a structure with fields: inn, kpp, ogrn, account, bic, bank_name. These can be used immediately to create a counterparty record, generate an invoice, or act.

Where It's Deployed and What Results It Delivers

The API is already integrated into several types of business systems:

• B2B Platforms and Online Stores—auto-filling requisites during legal entity registration, reducing rejections due to payment data errors.
• Banks—automatic client record checks before account opening, eliminating manual entry.
• Accounting Services—creating counterparties in 1C from contracts without operator involvement.
• CRM and ERP—synchronizing requisites between systems, eliminating duplicates and typos.

Economic impact: reducing document processing time from 5–10 minutes to 1–2 seconds, cutting error rates by 99%, minimizing fines and losses from failed deals.

Key Points

• Text Documents Only—scans without OCR not supported.
• Validation Is Mandatory—API doesn't return invalid INN or accounts, even if found.
• CP1251 Support—proper handling of old documents with problematic encodings.
• Ready Solutions for 1C—saves time on integration development.
• Free Trial—100 requests for 30 days to test on your data.

Test access is activated with promo code 1MHABR2026 until May 30, 2026. Get your API key via web form or email. The system is in production and used by the authors for their own automation—this is not an MVP, but an industrial-grade solution.

— Editorial Team