Payslip Parsing

1. Overview

Provides Claude with extraction rules for converting Danish payslip images and PDFs into structured data. This skill covers:

• •Standard Danish lønsedler (salary slips)
• •Common payroll systems (Danløn, Proløn, Visma, Zenegy, Bluegarden)
• •Employer-specific variations

It also provides a workflow for looking up and applying employer-specific parsers from payslip-knowledge/.

2. Danish Payslip Structure

Legally Required Fields

Danish employers must provide lønsedler with these mandatory fields:

Common Additional Fields

Typical Lønseddel Layout

[Company name and logo]
[Employee name and CPR (masked)]
[Pay period]
---
INDTÆGTER (EARNINGS)
  Månedsløn                    {base_salary}
  Bonus/tillæg                 {bonus}
  Bruttoløn i alt              {gross_salary}
---
FRADRAG (DEDUCTIONS)
  AM-bidrag (8%)               {am_bidrag}
  A-skat                       {a_skat}
  ATP                          {atp}
  Pension (egetbidrag)         {pension_employee}
---
ARBEJDSGIVERBIDRAG (EMPLOYER CONTRIBUTIONS)
  Pension                      {pension_employer}
  Sundhedsforsikring           {health_insurance}
---
TIL UDBETALING                 {net_salary}
---
[Bank account info]
[YTD totals]

3. Employer Detection & Parser Lookup

Parser Lookup Workflow

When processing a payslip, follow this decision tree:

1. Detect employer (see Employer Detection below)
2. Normalize employer name to employer-id (lowercase, no spaces, no A/S suffix)
3. Check: does payslip-knowledge/{employer-id}/PARSER.md exist?
   ├── YES: Load PARSER.md → use employer-specific extraction rules
   │         Set confidence boost: +0.1 (parser-assisted extraction)
   └── NO:  Fall back to general rules (Sections 4-5 below)
            Set confidence: 0.6-0.8 (general extraction)
            After processing: suggest /smartspender:payslip learn

Employer Detection Signals

Detect the employer using these signals in priority order:

Employer ID Normalization

For employers not in this table: normalize to lowercase, replace spaces with hyphens, remove A/S, ApS, and I/S suffixes.

4. Field Extraction Rules

Core Fields

Pension Fields

Optional Fields

5. Common Payroll Systems

Danløn

• •Identifier: "Danløn" in footer or header
• •Format: PDF, clean tabular layout
• •Gross location: "BRUTTOLØN" centered header
• •Net location: "NETTOBELØB" or "TIL UDBETALING"
• •Pension: Separate "PENSION" section with employer/employee split

Proløn

• •Identifier: "Proløn" or "Proløn Online" in document
• •Format: PDF, two-column layout
• •Gross location: Left column under "Løn"
• •Net location: "Til udbetaling" at bottom left
• •Pension: Listed under "Fradrag" section

Visma Løn

• •Identifier: "Visma" logo or "Visma Løn"
• •Format: PDF, structured sections with boxes
• •Gross location: "Bruttoløn" box
• •Net location: "Netto til udbetaling"
• •Pension: Employer contributions in separate "Arbejdsgiver" box

Zenegy

• •Identifier: "Zenegy" branding or modern minimalist design
• •Format: PDF, clean modern layout
• •Gross location: "Bruttoløn" with large font
• •Net location: Highlighted "Udbetalt" amount
• •Pension: Expandable sections (may need to look for itemized view)

Bluegarden

• •Identifier: "Bluegarden" or "SD Løn" (legacy)
• •Format: PDF, traditional Danish format
• •Gross location: "BRUTTO" line
• •Net location: "NETTO" or "Til disposition"
• •Pension: Under "FRADRAG" with employer portion noted separately

6. Validation Rules

AM-bidrag Validation

AM-bidrag must be exactly 8% of the AM-bidrag base (usually gross salary minus certain exempt items).

Validation formula:

expected_am = gross_salary * 0.08
tolerance = 5  # kr
valid = |extracted_am - expected_am| <= tolerance

If validation fails:

• •Check if there are exempt earnings (some bonuses, pension contributions)
• •Ask user: "AM-bidraget ({extracted_am} kr) afviger fra forventet ({expected_am} kr). Er der fradrag før AM-bidrag?"

Sum Validation

The payslip should balance:

Formula:

expected_net = gross_salary - am_bidrag - a_skat - atp - pension_employee - other_deductions
tolerance_pct = 0.02  # 2%
valid = |extracted_net - expected_net| / gross_salary <= tolerance_pct

If validation fails with variance > 2%:

• •Flag for user review
• •Ask: "Lønsedlen summer ikke helt ({variance}% afvigelse). Tjek venligst om alle fradrag er aflæst korrekt."

ATP Validation

ATP contributions are fixed amounts based on employment level (2025 rates):

If extracted ATP doesn't match standard amounts, note as non-standard but don't flag as error.

7. Output Schema

Extracted Payslip JSON

{
  "pay_period": "2026-01",
  "employer": "Virksomhed A/S",
  "employer_id": "virksomhed",
  "gross_salary": 45000.00,
  "am_bidrag": 3600.00,
  "a_skat": 12500.00,
  "atp": 94.65,
  "pension_employer": 4500.00,
  "pension_employee": 2250.00,
  "pension_total": 6750.00,
  "pension_pct": 15.0,
  "feriepenge": 5625.00,
  "net_salary": 26555.35,
  "benefits": {
    "sundhedsforsikring": 150.00,
    "fritvalgskonto": 450.00
  },
  "extraction_confidence": 0.9
}

CSV Row Mapping

Map extracted fields to payslips.csv columns per skills/data-schemas/SKILL.md:

8. Privacy Handling

CPR Number Masking

Danish CPR numbers (DDMMYY-XXXX) appear on payslips. NEVER store or display CPR numbers.

Masking rule: If CPR detected, replace with XXXXXX-XXXX in any logs or confirmations.

Bank Account Numbers

Bank account numbers may appear. Don't store in extracted data — only the net_salary amount matters.

9. Confidence Scoring

Rate the quality of each extraction:

10. Examples

Example 1: Standard Danløn Payslip

Input: PDF payslip from Danløn system

Extracted data:

pay_period: 2026-01
employer: Teknologi A/S
employer_id: teknologi
gross_salary: 42000.00
am_bidrag: 3360.00
a_skat: 11200.00
atp: 94.65
pension_employer: 4200.00
pension_employee: 2100.00
pension_total: 6300.00
pension_pct: 15.0
feriepenge: 5250.00
net_salary: 25245.35

Validation:

• •AM-bidrag: 42000 * 0.08 = 3360 ✓
• •ATP: Standard full-time rate ✓
• •Pension: 15% of gross ✓

Confidence: 0.95

Example 2: Partial Extraction (Blurry Image)

Input: Photo of payslip, partially obscured

Extracted data:

pay_period: 2026-01
employer: [unclear]
gross_salary: 38000.00
am_bidrag: [estimated] 3040.00
a_skat: [unclear]
net_salary: 24500.00
pension_total: [not visible]

Confidence: 0.5

User prompt: "Nogle felter var svære at aflæse. Tjek venligst bruttoløn (38.000 kr) og nettoløn (24.500 kr). Pension kunne ikke aflæses — vil du tilføje det manuelt?"

Example 3: Unknown Employer (No Parser)

Input: PDF payslip from new employer

Employer detection:

1. •Header shows "Startup ApS" → employer-id: startup
2. •Check payslip-knowledge/startup/PARSER.md → does not exist
3. •Fall back to general extraction rules

After processing: Suggest:

Tip: Kør /smartspender:payslip learn for at gemme udtræksregler
for Startup ApS, så fremtidige lønsedler bliver mere præcise.

11. Edge Cases

Multiple Pay Periods in One Document

Some employers provide YTD summaries alongside monthly slips:

1. •Extract only the CURRENT period data
2. •Ignore YTD columns
3. •Look for "Denne periode" or "Aktuel måned" labels

Supplementary Payments

Bonus, vacation payout, or other irregular payments:

1. •Extract the total gross (including supplement)
2. •Note irregular items in benefits_json
3. •Set confidence slightly lower (0.7-0.8) for unusual payslips

Foreign Currency Elements

For employees with foreign income components:

1. •Only extract DKK amounts
2. •Ignore foreign currency lines
3. •Note in extraction if payslip contains non-DKK items

Correction Slips (Rettelsesseddel)

Some months have correction slips instead of full payslips:

1. •Detect by "Rettelse" or "Korrektion" in header
2. •Extract the net change amount
3. •Store with special note in benefits_json: {"type": "correction"}

Related Skills

• •See skills/data-schemas/SKILL.md for the payslips.csv file structure
• •See skills/transaction-matching/SKILL.md for how payslips are matched to salary transactions
• •See skills/danish-finance-guide/SKILL.md for pension percentage recommendations (15% target)