Skip to main content

Field Processing Studio

Field Processing Studio is the layer between OCR/AI extraction and verification. It transforms raw recognized values into business-ready values.

Accounting sandbox administration

Why it exists​

OCR can recognize a value, but the business may need:

  • trimming;
  • replacing text;
  • extracting part of a string;
  • normalizing dates;
  • converting decimal separators;
  • validating IBAN;
  • mapping supplier name to ERP supplier code;
  • enriching value from a reference list;
  • calculating or checking VAT;
  • rejecting impossible values.

Older enterprise OCR products used scripting for this. eDocify uses a safer two-level model.

Rule levels​

Low-code rules​

Recommended for most users:

  • Trim;
  • Regex replace;
  • Substring;
  • Split;
  • Concat;
  • Normalize date;
  • Normalize amount;
  • Normalize VAT rate;
  • Lookup;
  • Map value;
  • Required field;
  • Conditional validation.

Advanced script​

Reserved for system administrators. It should run in sandbox, with timeout, without filesystem or network access, and with full audit.

Rule scope​

Rules can be scoped by:

  • tenant;
  • client group;
  • company;
  • document type;
  • OCR/AI provider;
  • field key;
  • supplier;
  • version.

Processing order​

Rules run in defined order. A rule can be active, draft, published, or rolled back.

flowchart LR
  A["OCR / AI raw value"] --> B["Field Processing Engine"]
  B --> C["Validation Engine"]
  C --> D["Verification UI"]
  D --> E["Approval and ERP export"]

Test before publish​

Before publishing, a rule profile should be tested with real OCR text:

  • paste OCR text;
  • run draft rules;
  • inspect before/after values;
  • inspect warnings;
  • compare against expected fields;
  • publish only after validation.

AI assistance​

AI can propose rules from examples:

  • "supplier name contains extra spaces";
  • "invoice number includes prefix that ERP does not need";
  • "date format is DD.MM.YYYY";
  • "amount uses comma decimal separator";
  • "IBAN is present but has spaces".

AI suggestions should remain draft until an admin tests and publishes them.

Audit and rollback​

Every published change should store:

  • author;
  • timestamp;
  • reason;
  • previous version;
  • new version;
  • affected client and document type;
  • test sample;
  • rollback option.