Payslips to structured JSON, EU-wide

What we extract

Every field, with confidence and citations

Every field comes back with a confidence score in [0,1] and a normalized bounding box on the source page. Net pay is recomputed from gross minus deductions and surfaced as a validator your agent can branch on.

{
  "schema_version": "payslip.v1",
  "result": {
    "employee_name": "Pieter Janssens",
    "employer_name": "Acme NV",
    "employer_address": "Havenlaan 86C, 1000 Brussel, BE",
    "period_start": "2026-04-01",
    "period_end":   "2026-04-30",
    "currency": "EUR",
    "gross_pay": { "amount": "3850.00", "currency": "EUR" },
    "deductions": [
      { "type": "rsz_employee", "label": "RSZ werknemer (13.07%)",          "amount": { "amount": "503.20", "currency": "EUR" } },
      { "type": "income_tax",   "label": "Bedrijfsvoorheffing",             "amount": { "amount": "742.15", "currency": "EUR" } },
      { "type": "meal_voucher", "label": "Maaltijdcheques eigen bijdrage",  "amount": { "amount": "21.78", "currency": "EUR" } }
    ],
    "net_pay": { "amount": "2582.87", "currency": "EUR" },
    "ytd_gross":      { "amount": "15400.00", "currency": "EUR" },
    "ytd_deductions": { "amount": "5068.52",  "currency": "EUR" },
    "ytd_net":        { "amount": "10331.48", "currency": "EUR" },
    "validation": {
      "net_pay_match": {
        "valid": true,
        "computed_net":  "2582.87",
        "declared_net":  "2582.87",
        "diff":          "0.00",
        "tolerance":     "0.01"
      }
    }
  },
  "field_metadata": {
    "net_pay.amount":      { "confidence": 0.98 },
    "deductions.0.amount": { "confidence": 0.96 },
    "employer_name":       { "confidence": 0.99 }
  }
}

Domain-specific validation

What makes this a specialist

Format coverage

Tested across 16+ EU and US payroll systems

16+ payroll systems · 6 jurisdictions · 4 languages (NL · FR · DE · EN)

Code recipes

From document to JSON in five lines

curl -X POST https://eu-api.tryparsr.dev/v1/parse/payslip \
  -H "Authorization: Bearer $PARSR_API_KEY" \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{
    "document_url": "https://example.com/april-payslip.pdf",
    "wait": 60
  }'

import os, httpx

resp = httpx.post(
    "https://eu-api.tryparsr.dev/v1/parse/payslip",
    headers={"Authorization": f"Bearer {os.environ['PARSR_API_KEY']}"},
    json={
            "document_url": "https://example.com/april-payslip.pdf",
        "wait": 60,
    },
    timeout=70,
)
result = resp.json()["result"]
match = result["validation"]["net_pay_match"]
if not match["valid"]:
    raise ValueError(f"Net pay does not reconcile — diff {match['diff']}")
print(result["employee_name"], result["net_pay"]["amount"], result["currency"])
for d in result["deductions"]:
    print(" -", d["type"], d["amount"]["amount"])

const resp = await fetch("https://eu-api.tryparsr.dev/v1/parse/payslip", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.PARSR_API_KEY}`,
    "Content-Type": "application/json",
    "Idempotency-Key": crypto.randomUUID(),
  },
  body: JSON.stringify({
    document_url: "https://example.com/april-payslip.pdf",
    wait: 60,
  }),
});
const { result } = await resp.json();
if (!result.validation.net_pay_match.valid) {
  throw new Error(
    `net pay does not reconcile — diff ${result.validation.net_pay_match.diff}`,
  );
}
console.log(result.employee_name, result.net_pay.amount, result.currency);

from langchain_parsr import ParsrToolkit
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

tools = ParsrToolkit.from_env().get_tools()
agent = create_react_agent(ChatOpenAI(model="gpt-4o"), tools)

result = await agent.ainvoke({
    "messages": [(
        "user",
        "Parse this April SD Worx payslip and tell me the take-home pay and "
        "total statutory deductions: https://example.com/april-payslip.pdf"
    )]
})
print(result["messages"][-1].content)

What teams build with this

Three integration patterns

Compared

How parsr's payslip parsing compares

FAQ

Common questions

• How do you handle different jurisdictions' deduction codes?

  Each parsed deduction returns both a canonical type (rsz_employee, lohnsteuer, loonheffing, paye, fica, …) and the raw label as printed. Belgian RSZ/ONSS, German Lohnsteuer / Solidaritätszuschlag / Kirchensteuer, Dutch loonheffing, French CSG/CRDS, UK PAYE/NI, and US federal / FICA / state lines are all canonicalized. Unknown lines come back as type='other' with the raw label preserved — never silently dropped.

• What languages are supported?

  NL, FR, DE, and EN — covering Belgian (NL/FR), Dutch, German, French, UK, and US payslips end-to-end. Trilingual Belgian payslips (NL + FR + DE column headers) are handled out of the box; the canonical deduction type is the same regardless of the printed language.

• How is year-to-date treated?

  ytd_gross, ytd_deductions, and ytd_net are returned when the payslip prints them. Mid-year hires whose first payslip has no YTD column return null for those fields rather than zero — null is honest, zero is misleading. If you parse a sequence of payslips for the same employee, validation.ytd_progression checks that month-over-month YTD values reconcile.

• What about a mid-year hire or a partial period?

  period_start and period_end reflect what's printed — a partial period (e.g. 2026-04-15 to 2026-04-30) is returned faithfully. Pro-rated gross_pay is returned as printed; we don't re-derive a full-month equivalent. If a downstream system needs an annualized figure, your code does that math with the period dates as inputs.

• Scanned vs digital PDF payslips?

  Vision LLMs handle PDF (text or image-based), JPEG, PNG, and HEIC. Digital PDFs hit ~95%+ field accuracy on Tier-1 systems (SD Worx, DATEV, Sage Paie, Loket, ADP) where fixture coverage is deepest. Phone-photo scans work down to ~150 DPI equivalent; deep skew or partial occlusion drops confidence. field_metadata.<field>.confidence < 0.85 is the recommended escalation threshold for review.

• When should I use doc_type='payslip' vs 'invoice'?

  Use 'payslip' for any employer-issued earnings statement showing gross pay, deductions, and net pay for a worked period. Use 'invoice' for goods or services billed by a vendor. Contractor invoices that look payslip-shaped (e.g. monthly retainer with VAT) should still be parsed as 'invoice' — they don't have the gross/deductions/net structure the payslip schema validates against.