NCB Credit Report OCR

Reads a National Credit Bureau report and returns the report's metadata, the data owner's identity, and a summary per credit account.

POST/ocr/v3/ncb-credit-report/fileTry it

• Price1 ICper page
• Measured95.6%
• Served29Kcalls

• v1.0
• Active
• Benchmark paper (PDF)

Input

OutputPOST /ocr/v3/ncb-credit-report/file

report_number

123456789

reference_number

MOBILE_APP-001

query_date

24 ต.ค. 2562 11:09:43

id_card_number

1100000000001

full_name

นาย สมชาย ตัวอย่าง

date_of_birth

11 เม.ย. 2529

total_accounts

6

accounts_under_dispute

0

accounts

lender_name: ธนาคารตัวอย่าง credit_type: สินเชื่อที่อยู่อาศัย account_status: ปกติ credit_limit: 3000000 outstanding_balance: 2500000

Every field, under Response

Request

curl -X POST "https://api.iapp.co.th/ocr/v3/ncb-credit-report/file" \
  -H "apikey: YOUR_API_KEY" \
  -F "file=@/path/to/document.pdf" \
  -F "target_lang=th" \
  -F "include_confidence=true"

Response

200OK. The fields are explained under Response.

{
    "docs": [
        {
            "message": "success",
            "raw": {},
            "processed": {
                        "report_number": "123456789",
                        "reference_number": "MOBILE_APP-001",
                        "query_date": "24 ต.ค. 2562 11:09:43",
                        "id_card_number": "1100000000001",
                        "full_name": "นาย สมชาย ตัวอย่าง",
                        "date_of_birth": "11 เม.ย. 2529",
                        "total_accounts": 6,
                        "accounts_under_dispute": 0,
                        "accounts": [
                                    {
                                                "lender_name": "ธนาคารตัวอย่าง",
                                                "credit_type": "สินเชื่อที่อยู่อาศัย",
                                                "account_status": "ปกติ",
                                                "credit_limit": 3000000,
                                                "outstanding_balance": 2500000
                                    }
                        ]
            },
            "confidence": {
                        "report_number": 0.95,
                        "reference_number": 0.95,
                        "query_date": 0.95,
                        "id_card_number": 0.95,
                        "full_name": 0.95,
                        "date_of_birth": 0.95
            },
            "process_ms": 4120
        }
    ],
    "docs_count": 1,
    "process_ms": 4300
}

Request

multipart/form-data only; there is no Base64 form. Every page of a PDF is read and billed.

Headers

apikeystringrequired

Your API key, created under API Keys in the control panel.

Body

filefilerequired

The report. PDF, JPEG, PNG or WEBP, up to 50 MB. Repeatable: several documents in one request.

return_ocrboolean

true to include the raw OCR text.

target_langstring

The language of the extracted values: th, the default, en, or auto to keep the document's own language.

include_confidenceboolean

Default true: a confidence from 0.00 to 1.00 per field.

/v3/store/ocr/ncb-credit-report is the same endpoint under the store prefix.

Response

The iapp-input-pages header carries the number of pages billed.

docsarray

One entry per document read

docs.processedobject

The fields read from the document

docs.processed.report_numberdocs.processed.reference_number

The report and reference numbers

docs.processed.query_datestring

When the report was requested

docs.processed.id_card_numberdocs.processed.full_namedocs.processed.date_of_birth

The data owner: the 13-digit national ID, the name, the date of birth

docs.processed.total_accountsdocs.processed.accounts_under_dispute

How many credit accounts, and how many under dispute

docs.processed.accountsarray

One entry per account

docs.processed.accounts.lender_namestring

docs.processed.accounts.credit_typestring

docs.processed.accounts.account_statusstring

docs.processed.accounts.credit_limitinteger

docs.processed.accounts.outstanding_balanceinteger

docs.messagestring

Status message

docs.rawobject

The OCR text, when requested

docs.confidenceobject

Confidence per field, 0 to 1

docs.process_msinteger

Processing time, in milliseconds

docs_countinteger

Number of documents read

process_msinteger

Processing time, in milliseconds

Status codes

Code	Meaning
400	Unsupported file extension, or a file over 50 MB
404	Unknown document type in the URL
422	No file attached
500	Processing error; the response carries a code to quote to support
502	Temporary processing failure, also when any page of a multi-page file fails (code: PROCESSING_FAILED). Not charged; retry
503	Service busy (PROCESSING_BUSY). Not charged; retry, after the seconds in Retry-After when that header is present
504	Processing timed out (PROCESSING_TIMEOUT). Not charged; retry

Details

Measured

On the production service, 28 October 2026, against 40 synthetic credit bureau reports: every person, ID number and lender invented, every page watermarked "specimen", half degraded as office scans and half as phone photographs.

Measure	Result
Report, identity and summary fields exactly right, 320 fields	95.6% (95% CI 93.8–97.5)
Scan / phone photograph	96.2% / 95.0%
Reports with every report, identity and summary field right	65.0%
Account rows: lender, type, status, limit, balance (860 values)	99.2%
Weakest field: reference number	67.5%
All 11 business-document types in the benchmark, 5,092 fields	97.9%
Median time per page through the API, 44 documents across the 11 types	10.8 s (19.9 s at the 95th percentile)

Most errors were in the reference number: an underscore came back as a space on 13 of the 14 reports that print one, so treat the two as equal when matching. The national ID number, dates and account counts were right on every report. Real reports score lower than this synthetic set. Method and every field: the benchmark (PDF).

Limits

• Only the fields above are extracted.
• Degraded scans, glare and handwriting over printed values lower accuracy. Prefer the original PDF, a square-on photograph in even light, or a scan at 300 dpi or more.
• The per-field confidence does not mark errors: across the 11 document types in the benchmark it averaged 0.957 on wrong fields and 0.967 on right ones, and 97.4% of wrong fields scored 0.90 or more. Choose fields for review with arithmetic checks (totals, VAT, balances) and identifier check digits instead.

Data handling

Consent and notice under PDPA and GDPR remain with you as the data controller.