Affordable Tax Invoice Extraction
for Korean SMEs Starting at $9/Month

Korea's E-Invoicing System Is World-Class — and That's the Blind Spot

South Korea's electronic tax invoice mandate is among the strictest and most comprehensive globally. All corporate entities have been required to issue electronic tax invoices (전자세금계산서) since January 2011. Individual business owners with annual supply value exceeding 300 million won (₩300M) joined the mandate in July 2014, and the threshold was lowered to 80 million won (₩80M) as of July 2024, under Article 32 of the Value-Added Tax Act (부가가치세법). By most measures, the system succeeded: electronic tax invoices are transmitted to the National Tax Service (NTS) in real time, eliminating the paper trail for millions of transactions.

But the system's very success created an assumption that doesn't hold at the SME level: that all tax invoices are electronic. They aren't. A small manufacturer in Gyeonggi Province with 40 employees and 80 active suppliers processes about 120 incoming tax invoices a month. Roughly 40 of those arrive as PDFs attached to emails, scanned paper invoices from vendors below the electronic mandate threshold, or photos snapped by field staff. None of those 40 are captured by 홈택스. They land in an accounting clerk's inbox and get manually typed into whichever ERP the company uses — Douzone Smart A, Ecount, or an Excel spreadsheet someone built three years ago.

Most discussions of Korean tax invoice automation assume the invoices are already digital and the problem is issuing them correctly. The NTS system, ERP vendor marketing, and even third-party ASP services (바로빌, 하이웍스) focus on the issuer's workflow: generating electronic tax invoices, transmitting them to the NTS by the 11th of the following month, managing the certificate-based signing process. The recipient — the SME trying to get supplier invoice data into its accounting system — is left with manual typing as the default answer.

What 홈택스 Actually Covers — and What It Doesn't

HomeTax (홈택스) — and its mobile counterpart SonTax (손택스) — is a free government platform for issuing electronic tax invoices, filing VAT returns, and querying tax records. For businesses issuing invoices, it provides a direct channel to the NTS: log in with a digital certificate (공동인증서), enter the supplier and recipient details, and the invoice is transmitted immediately. The buyer can verify receipt through the same platform. For electronic tax invoices issued through certified ERP systems or ASP providers, 홈택스 serves as the central registry — all electronic invoices flow through it.

What 홈택스 doesn't do is read data from PDFs, scanned documents, or images. It doesn't extract fields from an invoice a supplier emailed as an attachment. It doesn't convert a paper tax invoice that was photographed with a phone into structured data fields like supply value (공급가액), VAT amount (부가세액), or supplier registration number (사업자등록번호). The system was designed for digital documents that were born electronic — not for digitizing documents that arrived in any other form.

This distinction matters because the Korean SME's incoming invoice workflow has two distinct streams. Stream one: electronic tax invoices received through 홈택스 — these are already in the system, queryable, and reconcilable through the government portal or ERP integration. Stream two: everything else — the PDFs, the scanned paper, the photos, the faxes from older suppliers, the overseas invoices from foreign vendors that don't participate in the Korean e-invoicing system at all. Stream two is where the manual labor lives. And stream two is invisible to the government platform.

What Korean Invoice Processing Tools Actually Cost

The Korean business software market offers tools at multiple tiers — from near-free accounting software with basic OCR to enterprise ERP suites with dedicated document automation modules. But across every tier, the pricing assumptions are the same: the buyer is either an ERP customer looking for an integrated solution or an enterprise with enough volume to justify a dedicated document processing investment. The SME that just wants to extract data from PDF tax invoices into a spreadsheet doesn't fit either profile.

The pattern is consistent: every Korean tool that includes document data extraction sits at the enterprise tier — priced for procurement departments with annual budgets and months-long implementation timelines. The tools priced for SMEs — Ecount at ₩40,000/month, JOBIS at ₩33,000/month — include accounting, payroll, and tax filing, but their OCR capabilities stop at receipts. A supplier tax invoice with six required fields across a varied layout remains a manual typing job.

Why a Dollar-Priced Extraction Tool Changes the Math

At ₩1,450 to the dollar (June 2026), a $9/month AI extraction plan costs ₩13,050. That's less than one-third of the cheapest Korean software that includes any form of OCR. The gap isn't a currency anomaly — it's a structural difference in who each tool was built for.

Korean business software vendors target the enterprise buyer: initial licensing fees, annual maintenance contracts, per-user add-ons, mandatory training sessions. The pricing architecture assumes a finance department with a budget cycle. Ecount's ₩40,000/month subscription — arguably the best value in Korean ERP — buys an entire accounting and inventory management suite, not standalone extraction. The extraction capability, to the extent it exists, is bundled inside a much larger product.

A dollar-priced AI extraction tool like ImageToTable.ai targets a different buyer entirely: the self-serve small business owner who wants to upload a PDF and get a spreadsheet back. No ERP swap. No implementation project. No per-user licensing. The $9/month Personal plan provides 150 credits — each credit processes one page — enough for 120 to 150 single-page tax invoices a month. At the Pro tier ($19/month, 400 credits), a 200-invoice workload costs ₩27,550 a month — roughly the price of Ecount's ERP but allocated entirely to extraction rather than a full accounting suite you may already have.

This isn't about finding a cheaper ERP. It's about recognizing that extraction can be separated from accounting — and the extraction step is the only one that was actually costing 10 hours of manual work. The SME that already uses Douzone or Ecount for accounting doesn't need to replace it. It needs to stop typing the same supplier registration number into a form that hasn't changed in years.

The Extraction That Works Across Every Supplier Layout

Traditional OCR — the technology underlying most receipt-scanning features in Korean accounting software — works by recognizing characters in fixed positions. It's reliable when the layout is predictable: a receipt from a convenience store chain, a standardized card slip, a tax invoice issued through the NTS system in a fixed XML format. But supplier tax invoices don't arrive in a fixed layout. The supplier registration number (사업자등록번호) appears top-right on one vendor's form and bottom-center on another's. The tax breakdown by rate sits in a sidebar box on some, in a line-item column on others. Template-based OCR needs a template per supplier — and maintaining 40 supplier templates is exactly the kind of overhead that makes SME finance teams give up and type by hand.

ImageToTable.ai uses semantic extraction rather than position-based OCR. Instead of looking for text in a specific coordinate on the page, the AI reads the document the way a human would — understanding that "공급가액" means supply value regardless of where it appears, that "123-45-67890" is a registration number regardless of whether it's next to a label or embedded in a header. You define the columns you want — "Supplier Registration Number (사업자등록번호)", "Supply Value (공급가액)", "VAT Amount (부가세액)", "Invoice Date (작성일자)" — and the AI locates each value by understanding what it means, not where it sits. This is the fundamental difference from template OCR: 30 different supplier layouts cost the same as one.

This semantic approach also means the tool works on any format a supplier might send: the clean PDF generated by an NTS-compliant ERP, the scanned paper invoice from a vendor below the electronic mandate threshold, the photo of a tax invoice taken on a phone at a trade show, the fax-to-PDF from a supplier still running a legacy system. The extraction quality depends on the document's legibility, not its origin — because the AI reads meaning, not pixels on a grid.

Monthly Costs at Three Korean SME Invoice Volumes

At three invoice volumes typical of Korean SMEs, the cost gap between a dollar-priced AI extraction tool and Korean domestic alternatives is not marginal — it's in a different pricing category. The table below uses publicly listed prices and an exchange rate of ₩1,450/$1. Domestic tool costs represent the cheapest tier of each that includes any extraction capability; 홈택스 is free but limited to electronic-only invoices.

The numbers reveal a structural gap. Ecount at ₩44,000/month is affordable as an ERP — it replaces an entire accounting, inventory, and payroll stack. But even at that price, the extraction part of the workflow remains manual. The ERP organizes the data once entered; it doesn't eliminate the entry step. A dollar-priced extraction tool at $9/month doesn't replace the ERP — it feeds it. The two tools solve different problems, and the extraction tool solves the one that was costing 10 hours of manual typing a month.

For an SME that processes 150 supplier invoices monthly, the manual data entry cost — at South Korea's 2025 minimum wage of ₩10,030/hour for roughly 4-5 minutes per invoice — runs approximately ₩100,000–125,000/month in labor. A tool that automates that step for ₩27,550/month cuts the labor cost by over 75% — and the person who was typing invoices can spend Monday morning doing the exception handling that actually requires human judgment.

FAQ

Does ImageToTable.ai work with Korean-language tax invoices?

Yes. The AI reads Korean text on tax invoices — including supplier registration numbers (사업자등록번호), supply values (공급가액), VAT amounts (부가세액), dates (작성일자), and recipient information. Korean-language document support is built into the vision model's training, not layered on top of an English-only engine.

Can it handle handwritten tax invoices?

Yes, though accuracy depends on legibility. Handwritten fields on printed forms (e.g., a handwritten amount on a pre-printed 세금계산서 form) work reliably. Fully handwritten documents with irregular layouts produce lower accuracy and may require manual review of extracted values. The tool is strongest on printed and typewritten Korean documents, which covers the vast majority of supplier tax invoices.

Do I need to stop using Douzone or Ecount to use an AI extraction tool?

No. ImageToTable.ai exports to Excel (XLSX) and CSV — formats every Korean ERP can import. The workflow is: upload supplier PDFs → AI extracts fields → download spreadsheet → import into your existing ERP. The extraction step is independent of your accounting system. You keep your current ERP for accounting, tax filing, and financial reporting exactly as before.

How is this different from the OCR built into Ecount or JOBIS?

Ecount and JOBIS include receipt-level OCR optimized for short, single-format documents (영수증). A supplier tax invoice has six or more fields spread across a page with significant layout variation between vendors. The built-in OCR in these tools wasn't designed for that use case — it was designed for the short, standardized receipts that come from corporate card transactions. ImageToTable.ai uses semantic AI extraction that reads documents by understanding what each field means, not where it's positioned on the page.

Can it extract from 홈택스-downloaded electronic tax invoices?

홈택스 electronic tax invoices are already structured XML data — they don't need extraction. You can query them through the NTS system or your ERP's integration. The extraction tool is for the documents 홈택스 doesn't cover: PDFs, scans, photos, and paper invoices from suppliers outside or below the electronic mandate.

What about security? Can I upload sensitive tax documents?

Files are processed through encrypted connections and are not stored after processing. For businesses with strict data governance requirements — 한국딥러닝's DEEP OCR+ offers on-premise VLM-based extraction — cloud-based AI extraction may not be suitable. ImageToTable.ai is designed for the SME use case where cloud processing is acceptable and the priority is getting data into a spreadsheet quickly and affordably.