Bureau report parsing
Each bureau has its own parser that knows the structure of its report. As a result, data is extracted in seconds and identically on every run. Text recognition and an AI model are used only for scans and non-standard documents.
Plan All plans. The OCR page limit depends on the plan: up to 20 in Starter, up to 50 in Growth.
What’s inside
- Dedicated parsers for NBKI, OKB, Scoring Bureau, FCB and SCB reports
- The NBKI payment calendar is read directly from the PDF graphics
- The PDF text layer goes to the parser, a scan goes to text recognition
- OCR in Russian, Kazakh, Kyrgyz, Uzbek, Mongolian and English
- PDFs up to 100 pages and 25 MB, PDF, JPG, PNG, HEIC and TIFF formats
- Confidence for every page and coordinates for every field
- Operator field corrections that keep the original value and the reason
- AI model output is checked against its own totals, otherwise it gets low confidence
Limitations
Delinquency is stored in ranges, as the bureau reports it, with no invented exact day counts. If a field could not be read with confidence, it is flagged rather than filled in with a guess.
FAQ
Why not recognize everything with an AI model?
A parser built for a specific bureau gives the same result on every run and works in seconds. An AI model is more expensive, slower and can make mistakes, so it remains a fallback.
What if a bureau changes its report format?
Bureau templates are versioned and updated without a new platform release. While a template is being updated, the document goes through text recognition and manual review.
Use cases with this feature
Other features
Shall we run a check on your report?
Send an anonymized sample report from the country you need: we will break it down in a demo and show the flags, the summary and the connection to your system.