How to Extract Data from Korean
Payslips into Excel

A Payslip that Compliance Built — and Extraction Left Behind

The Korean payslip was redesigned by regulation, not by convenience. When the National Assembly amended Article 48 of the Labor Standards Act (근로기준법) in May 2021—effective November 19 of that year—it elevated the payslip from an optional courtesy to a mandatory document backed by penalties of up to 5 million KRW per violation. Suddenly, every employer in Korea, including businesses with fewer than five employees, had to issue an itemized statement containing ten specific data points for every worker every pay period.

The compliance outcome was swift: virtually every employer in Korea now issues payslips. The data extraction outcome is less obvious. Unlike a Korean tax invoice (세금계산서), which follows a heavily standardized format under Article 32 of the VAT Act, a Korean payslip has no government-prescribed layout—only prescribed content. The same ten fields can appear in radically different arrangements depending on which software the employer uses.

Douzone Smart A (더존 스마트A), the dominant ERP in Korea's mid-market, places deduction items (공제내역) in a compact two-line block after earnings. ECOUNT (이카운트), widely adopted by smaller firms for its web-based accessibility, spreads deductions across a wider grid with sub-section headers. PAYZON, Shiftee (시프티), flex, and a dozen other HR platforms each render the payslip differently. Even within the same software, upgrades change the layout. The result: what looks like a single document type to an HR manager is actually a family of visually distinct layouts to a computer.

Most Korean SERP content on "급여명세서 + Excel" teaches you how to create payslips inside Excel—building VLOOKUP formulas, linking 간이세액표 tables, embedding VBA macros to auto-generate PDFs. That solves the generation problem. It does not solve the aggregation problem: you already have the payslips. You need the data out of them and into one spreadsheet.

The Ten Fields the Law Requires — and Why They Matter for Extraction

Before building an extraction workflow, you need to know exactly what you are looking for. Under Article 27-2 of the Enforcement Decree of the Labor Standards Act (근로기준법 시행령 제27조의2), every Korean payslip must contain ten items. Missing even one exposes the employer to progressive fines. For extraction purposes, these ten items define your target columns.

The deduction section is where the heaviest lifting happens during extraction. Korean payslips carry five to seven deduction line items, each calculated against a different rate. The 2026 rates—effective January 1 of this year—are:

Source: 2026 Korean Social Insurance Rates (HRSide). Income tax (소득세) uses an eight-bracket progressive schedule from 6% to 45%; local income tax (지방소득세) is a flat 10% surcharge on the income tax amount.

These rates are not background trivia—they are your post-extraction sanity check. If your extracted National Pension amount for an employee earning 3,000,000 KRW/month is not within a few thousand won of 142,500 KRW (3,000,000 × 4.75%), you know something went wrong upstream—either the extraction missed a digit or the payslip itself contains an error. More on this verification step later.

Step 1: Collect Your Payslip Files — from Douzone, ECOUNT, Paper, or Camera

Korean payslips arrive in HR departments through three main channels, each with different extraction characteristics.

Software-generated PDFs. Douzone Smart A, ECOUNT, Shiftee, and PAYZON all produce payslips as downloadable PDFs. These are your best-case scenario for extraction—text is machine-readable, layout is consistent within a given software version, and you can batch-download a month's worth of payslips at once. In Douzone, navigate to 인사급여 → 근로/연말 → 근로소득관리 → 급여대장(명세), then use 기능모음 → 일괄인쇄 and select PDF as the output. In ECOUNT, the payroll report can be saved as Excel, though the layout still varies by your template settings.

Employee-submitted files. Some employers email payslips as PDFs; some still hand out printed copies. If your workforce includes part-time or contract workers paid through multiple channels, you'll receive a mix of crisp PDFs, smartphone photos of printed pages, and screenshots from mobile banking apps showing the deposit notice (입금확인증). All of these can be extracted—but a clear, straight-on photo will always yield better results than a tilted, low-light shot.

Scanned or photographed printouts. Smaller businesses that use in-house Excel templates often generate payslips as printed paper. A flatbed scan at 200–300 DPI produces extractable text; a shaky smartphone photo at an angle may drop recognition accuracy by 15–20 percentage points. If you expect to process scans regularly, investing in a document scanner with auto-feed (ADF) pays for itself in extraction quality within the first batch.

Step 2: Define Your Columns Once — Let the AI Handle the Rest

This is where the extraction approach diverges from traditional OCR. Template-based tools—the kind that ask you to draw bounding boxes around where each field appears on a page—fail at Korean payslips because the layout changes between Douzone, ECOUNT, and PAYZON. You would need a separate template for every software output your company receives.

Instead, you define your columns by what they mean, not where they sit. Type the field names you want into the column definition panel, and the AI locates each value by understanding the semantics of the payslip—reading "기본급" as Base Pay, "국민연금" as National Pension, "차인지급액" as Net Pay—regardless of whether those labels appear top-left, center-stacked, or nested inside a sub-table. This is Custom Column Extraction: the column names you enter become the headers of your final Excel sheet, and the AI fills the rows by matching meaning to content, not coordinates to pixels.

Here are the columns a typical Korean payroll reconciliation workflow needs:

You can also add inferred columns—fields the AI derives from context that may not appear explicitly on the payslip. For example, a column defined as Pay Frequency (options: Monthly/Weekly/Biweekly) lets the AI determine whether the employee is paid monthly (월급) or weekly (주급) based on the pay period and wage structure on the document. Another useful inferred column is Deduction Verification (options: OK/Warning)—the AI can flag rows where the extracted social insurance amounts deviate significantly from the expected rates.

If your reconciliation spreadsheet already has formulas that reference specific column names downstream, using these exact column names in the extraction means the extracted data feeds directly into your existing calculations. No column renaming, no re-mapping. Payroll register reconciliation is a common downstream use case that benefits from this kind of direct integration.

Step 3: Batch Process and Sanity-Check with Social Insurance Rates

Upload all the payslips from one pay period in a batch. The AI processes each file independently, populating the same set of columns regardless of whether one file came from Douzone and the next from ECOUNT. When processing completes, you export one merged Excel file—one row per employee, all columns populated.

Processing speed: approximately 5–10 seconds per page. Manual data entry, by comparison, averages about 3 minutes per payslip—and that's before you double-check the numbers. For a 30-employee company, batch extraction turns a monthly 90-minute data entry session into roughly 3–5 minutes of processing time, plus a verification pass.

These three checks catch the vast majority of extraction anomalies before they propagate into your payroll reconciliation or tax filing. The 2026 rate increases—0.5pp for National Pension, 0.1pp for Health Insurance—make this verification step more important now than it was last year, since some software packages may have updated their formulas while others still use 2025 rates.

Try It: Extract a Sample Korean Payslip

The demo below loads with payslip-specific extraction columns pre-configured. Upload a sample Korean payslip—a Douzone PDF, a screenshot, or any format—and watch the AI pull Base Pay (기본급), National Pension (국민연금), Net Pay (차인지급액), and all other deduction fields into a structured table. No login required.

What Changes After Extraction: Reconciliation, Audit Prep, and Year-End Filing

Once the data is in Excel, the bottleneck shifts from "getting numbers into a spreadsheet" to "doing something useful with them." For most Korean HR and payroll teams, the downstream tasks fall into three buckets.

Monthly payroll reconciliation (급여대사). The extracted spreadsheet becomes your truth table. Compare each employee's extracted gross pay against the payroll register exported from your accounting software. If Douzone says the total payroll outflow was 45,230,000 KRW and your extracted spreadsheet sums to 45,228,500 KRW, the 1,500 KRW gap points to a specific row—and a specific field on that row—rather than requiring you to re-trace the entire payslip stack. Pay stub extraction with side-by-side comparison columns turns reconciliation from a needle-in-a-haystack problem into a sort-by-difference exercise.

Year-end tax settlement prep (연말정산). Korea's year-end tax reconciliation in January/February requires aggregating twelve months of income, deductions, and social insurance contributions per employee. If each month's payslip data lives in a separate PDF, assembling the annual picture means opening 144 files (12 months × 12 employees) and transcribing values. With extracted data, you already have twelve monthly sheets—merge them into one annual summary with a pivot table, and the 연말정산 preparation that used to take a full day takes the length of a coffee break.

Audit and compliance documentation. Under the Labor Standards Act, employers must retain payslip records for three years. An auditable trail of extracted data—with source filenames, extraction timestamps, and the original payslip files archived alongside—satisfies both the recordkeeping requirement and any future labor board inquiry. If the audit question is "show us the gross pay, deductions, and net pay for employee X across all of 2025," you have it in one Excel file instead of a filing cabinet.

These three use cases share one property: they require the data to be portable. You can answer payroll questions, run year-end calculations, or satisfy an auditor only if the values can move freely between software systems. Extraction creates that portability. The cost of not having portability—measured in reconciliation hours, missed filing deadlines, and audit prep scrambling—is the hidden line item on every manual payroll department's budget.

Frequently Asked Questions

Does this work with payslips written entirely in Korean?

Yes. The AI reads the semantic meaning of Korean-language field labels—"기본급," "국민연금," "공제내역"—and maps them to whatever English column names you define. The language on the document does not need to match the language of your column headers. This is especially relevant for Korean companies whose HR teams define columns in Korean while extracting from payslips that use the same Korean terms.

What about handwritten payslips or small-business formats?

Handwritten text (수기 기입) has a lower recognition ceiling than printed text—expect roughly 80–90% accuracy for clear handwriting in dark ink on a clean background, versus up to 99% for printed fields. If your company receives handwritten payslips from part-time or daily workers, batch them separately and schedule extra review time for those rows. The same verification formulas (rate proportion checks for social insurance, net pay = gross − deductions) remain your best quality guard.

Can I process payslips from both Douzone and ECOUNT in the same batch?

Yes. The column definitions are source-agnostic—you define "National Pension" once, and the AI locates the corresponding value whether it appears on a Douzone PDF, an ECOUNT spreadsheet export, or a Shiftee electronic payslip. This is the practical advantage of semantic extraction over template-based approaches: adding a new payslip source does not require a new template.

How do I handle employees with the same name?

Include the Employee ID column (사원번호 or 생년월일) in your extraction definition. Dong-soo Kim from the engineering team and Dong-soo Kim from the sales team will be disambiguated by their employee numbers. If the payslip uses birth dates instead, define Employee DOB as a separate column. The extracted table then contains enough identifying information to correctly assign each row to the right person in your downstream spreadsheet.

What if the social insurance rates change mid-year?

Korean social insurance rates are announced annually and take effect January 1. The 2026 rates listed in this article (National Pension 9.5%, Health Insurance 7.19%) apply to all payslips with a pay date from January 1, 2026 onward. If a rate change occurs mid-year—rare but possible through emergency legislation—update your verification formulas and re-check any flagged rows. The extraction itself does not need recalibration; the AI reads the actual printed amounts, not the expected amounts.

Can the Google Sheets add-on handle this workflow?

Yes. ImageToTable.ai's Google Sheets add-on extracts payslip data directly into your active spreadsheet tab, eliminating the download-upload-reimport cycle entirely. Define your columns once in the sidebar, upload payslip files, and the extracted rows append directly below your existing reconciliation formulas. For companies already running payroll in Google Sheets, the add-on reduces the extraction workflow to a single platform.