Korean Document Extraction Pricing
Government vs AI Tools in 2026

Three Price Tracks, Three Different Buyer Assumptions

South Korea has roughly 8 million small and medium enterprises, according to the Ministry of SMEs and Startups' 2022 survey, employing 18.96 million people and generating ₩3,309 trillion in combined revenue. These businesses process an enormous volume of documents — tax invoices, receipts, bank statements, purchase orders, delivery notes — across an economy where the electronic tax invoice mandate (부가가치세법 제32조) makes Korea one of the world's most digitized document environments.

But "most digitized" doesn't mean "fully automated at the SME level." The gap between the infrastructure (electronic tax invoices flowing to the NTS in real time) and the daily workflow (a finance clerk typing PDF data into Douzone) is where the extraction pricing question lives. And the answer depends entirely on which track you're evaluating.

The three tracks don't compete on the same dimension. Track 1 is free but narrowly scoped. Track 2 is fully featured but priced for buyers who are either buying an entire ERP or have enough volume to justify the per-user licensing. Track 3 is purpose-built for one thing — extracting data from documents — and priced for the self-serve buyer who just wants the spreadsheet. The pricing tier architecture of the Korean market makes Track 2 the default for anyone searching in Korean; Track 3 requires knowing it exists.

Track 1: 홈택스 — Free, Government-Built, and Strictly Limited

HomeTax (홈택스) and its mobile version SonTax (손택스) are the default starting point for any discussion of Korean document data. The system handles electronic tax invoice issuance, transmission to the NTS, and query of received invoices — all at zero cost. For an SME that only deals with registered suppliers who issue electronic tax invoices through the NTS system, 홈택스 covers the entire tax invoice data workflow. The data is already structured, verifiable, and accessible through the government portal or ERP integration.

The limitation is scope. 홈택스 was designed as a tax administration platform — a registry and transmission system for electronic documents — not a general-purpose document extraction tool. It doesn't process PDFs. It doesn't read scans. It doesn't extract fields from photos. A supplier who issues a paper tax invoice or emails a PDF is invisible to the system. A foreign vendor invoice from a Chinese or Japanese supplier is invisible. A receipt, a bank statement, a delivery note, a purchase order — none of these exist in 홈택스 because they were never meant to.

For businesses processing only electronic tax invoices through fully compliant suppliers, 홈택스 is effectively the complete solution — plus their ERP for accounting. But for the typical SME with 80 to 200 supplier relationships spanning registered and unregistered vendors, domestic and foreign partners, electronic and paper formats, 홈택스 covers maybe 60% of the incoming document volume. The remaining 40% lands in Track 2 or Track 3 territory.

Track 2: Korean Commercial Tools — What ₩30,000 to ₩50,000+/Month Buys

The Korean commercial document software landscape divides into three layers, and the extraction capability varies dramatically across them.

Layer 1 — Accounting/ERP suites with basic OCR. Ecount (이카운트) at ₩40,000/month (+ ₩200,000 initial) serves over 80,000 businesses with a full ERP stack: accounting, inventory, payroll, groupware — all with unlimited users. JOBIS (자비스) at ₩33,000/month provides AI-powered bookkeeping with bank transaction aggregation and 4대보험 payroll integration. Both include receipt-level OCR optimized for short, standardized slips (영수증). Neither includes the semantic document extraction needed for multi-field supplier invoices where the layout changes with every vendor.

Layer 2 — Expense management platforms with receipt OCR. BizPlay (비즈플레이) automates corporate card expense processing across 17 Korean card company networks, with per-user pricing (도입비 + monthly per-user). Strong at collecting receipt images and matching them to card transactions — a real efficiency gain for employee expense workflows. Not designed for extracting structured data from supplier tax invoices, purchase orders, or any document type beyond the expense receipt use case it was built for.

Layer 3 — Standalone OCR engines. Naver Clova OCR leads in Korean-language accuracy at 97–99%, priced at approximately ₩50 per page on a usage basis. NHN Cloud OCR and Kakao Enterprise OCR offer similar per-page pricing with platform-specific strengths (receipt field extraction, KakaoWork integration). Lido (a US-Korean startup) offers a flat-rate ₩39,000/month option for no-code structured extraction with Korean-language support. These are genuine extraction tools — they read Korean documents and output structured data. But they occupy a narrow segment of the market: the buyer who specifically searches for "OCR" or "문서 추출" rather than "회계 프로그램" or "ERP."

At the enterprise end, Samsung SDS Brity Works offers document AI with 94–96% structured document accuracy and RPA integration — but the pricing starts in the tens of millions of won (₩15,000,000+) with months-long implementation timelines. Korea Deep Learning (한국딥러닝) offers VLM-based DEEP OCR+ with on-premise deployment and enterprise-grade security — a genuine Korean alternative to overseas VLM tools, but priced and positioned for enterprise buyers with compliance requirements, not SMEs.

The common thread across all three layers: the tools that extract data well are either bundled inside an ERP you may not need, priced per-page at rates that add up at SME volumes, or positioned at an enterprise tier with implementation costs that dwarf the monthly subscription. For the SME that already has an accounting system and just needs data extracted from documents into a spreadsheet, none of these tracks is optimized for that exact workflow. The comparison between subscription and per-page tools reveals different breakeven points for different volumes.

Track 3: Dollar-Priced AI Extraction — The Track Nobody in Korea Searches For

At ₩1,450 to the dollar, a $9/month extraction plan costs ₩13,050 — less than a third of Lido's ₩39,000/month and roughly the same as processing 260 pages through Naver Clova OCR's per-page pricing. The gap isn't about Korean tools being expensive — it's about them being priced for a different buyer.

ImageToTable.ai represents the third track: a dollar-denominated, self-serve extraction tool that targets the small business owner who wants to upload a PDF and get a spreadsheet back. No ERP to install. No implementation project. No per-user licensing — the $9/month Personal plan (150 credits) covers an individual business owner's monthly document volume, and the $19/month Pro plan (400 credits) covers a typical SME's monthly document processing load.

The technical approach differs from Korean domestic OCR in one critical way: it uses semantic AI extraction rather than position-based OCR. Traditional OCR engines like Clova OCR read text from a page and return it as a string — you then need to parse that string to find field values. Semantic extraction reads the document the way a human would: understanding that a 10-digit number following the pattern XXX-XX-XXXXX next to the characters "사업자등록번호" is the supplier registration number, regardless of where on the page it appears. This means no template configuration per supplier — define the columns once and the AI matches values across every document format. For the SME with 40 different suppliers using 40 different invoice layouts, this is the difference between automated extraction and giving up.

For tax invoices specifically, the 세금계산서 extraction economics favor the third track even more dramatically — because 홈택스 already handles electronic invoices for free, the extraction tool's value is concentrated on the PDFs and scans the government system can't touch. And for document types beyond tax invoices — bank statements, purchase orders (발주서), delivery notes, contracts — 홈택스 offers nothing at all. Those documents live entirely in Track 2 or Track 3 territory.

Monthly Costs at Three SME Document Volumes

The following comparison assumes a Korean SME processing a mix of document types: tax invoices (세금계산서), receipts (영수증), purchase orders (발주서), and bank statements. Three volume scenarios reflect different business sizes and document intensity. All prices include VAT where applicable. The dollar exchange rate used is ₩1,450/$1.

The cost comparison reveals a counterintuitive pattern. At 50 documents a month, Naver Clova OCR's per-page pricing is actually the cheapest option — ₩2,500/month. But that pricing requires API integration: you need a developer to connect Clova OCR to your workflow, build the parsing logic to extract data from the raw OCR output, and maintain the integration. The ₩2,500 price tag doesn't include the development cost, which for a one-time integration might run ₩500,000–₩1,500,000 (roughly $350–$1,000). Spread that over a year and the effective monthly cost jumps to ₩44,000–₩127,500 — putting Clova OCR in the same range as Ecount or Lido, but with ongoing maintenance overhead.

At 150 documents a month, the no-code flat-rate tools (Lido at ₩39,000, ImageToTable.ai at ₩27,550) become the clear economic winners — they don't require development, don't charge per page, and don't bundle extraction inside an ERP you may not need. The Ecount price stays flat at ₩44,000/month, but the extraction capability doesn't scale with volume — more documents just means more manual typing inside the ERP.

At 300 documents a month, the per-page API tools start to lose their price advantage against flat-rate alternatives. But the more important shift at this volume is operational: a business processing 300 documents a month can no longer afford to have extraction accuracy vary by supplier layout. Template-based OCR breaks down at this diversity. Semantic extraction — whether from a dollar-priced tool or an enterprise-grade Korean VLM solution like DEEP OCR+ — becomes the minimum viable approach.

When to Use Each Track: A Decision Framework

The right extraction track depends on three variables: what kind of documents you process, how many, and whether you already have an ERP you're committed to.

Use Track 1 (홈택스) exclusively if you deal only with electronic tax invoices from registered Korean suppliers and you already query them through 홈택스 or your ERP's NTS integration. The system is free, reliable, and the data is already structured. You don't need extraction — you need query and reconciliation.

Use Track 2 (Korean commercial tools) as your primary platform if you need a full ERP or accounting suite and the built-in OCR covers a meaningful share of your document types — typically receipts and standardized invoice formats. Ecount, JOBIS, or Douzone serve as the backbone of your financial operations. But expect manual data entry for PDF tax invoices, purchase orders, bank statements, and any document format that varies by supplier.

Layer Track 3 (dollar-priced AI extraction) on top of your existing ERP if you already use Douzone, Ecount, SAP, or any other accounting system and just need to stop typing data from documents. The extraction tool exports to Excel, your ERP imports from Excel — the integration is the file format itself. This is the lowest-friction approach for the SME that has a working accounting system and wants to eliminate the data entry step without replacing anything.

Use Track 3 as your extraction layer even for 홈택스-covered invoices when you're aggregating data across multiple sources for analysis or reporting. 홈택스 shows you individual invoices. An extraction tool lets you batch-process 50 supplier invoices — including the PDFs — into a single spreadsheet for reconciliation, cost analysis, or audit preparation. The 세금계산서 deep-dive explores the economics of this specific workflow in detail.

FAQ

Which Korean document types can AI extraction tools handle?

Tax invoices (세금계산서), receipts (영수증), purchase orders (발주서), bank statements, delivery notes, contracts, payslips (급여명세서), and most printed Korean business documents. The accuracy is highest on printed/typewritten documents with clear field labels. Handwritten documents work with reduced accuracy depending on legibility. Documents in mixed Korean-English or Korean-Chinese formats are supported — the AI reads all languages present on the page.

Does the exchange rate make dollar-priced tools unreliable long-term?

The ₩1,450/$1 rate used in this article reflects mid-2026 levels. Exchange rate fluctuations will shift the won-denominated price — a move to ₩1,300/$1 drops the $9 plan to ₩11,700; a move to ₩1,600/$1 pushes it to ₩14,400. Even at ₩1,600, the dollar-priced tool remains less than half the cost of the cheapest Korean flat-rate alternative (Lido at ₩39,000). The pricing gap is structural, not currency-dependent.

Can I use a dollar-priced tool if my ERP is entirely in Korean?

Yes. ImageToTable.ai processes Korean documents and outputs Excel (XLSX) or CSV files. These formats are language-agnostic — your Korean ERP imports them the same way it imports any spreadsheet. The extraction output columns can be named in Korean, English, or any mix — the tool doesn't enforce a language on your column headers. The web interface is in English, which is the only friction point for Korean-only users.

What's the difference between Naver Clova OCR and AI semantic extraction?

Clova OCR reads text from a page and returns it as a string with position data — you then need to parse that output to identify which text is the supplier name, which is the amount, and so on. This parsing step requires either custom development or a pre-trained template per document layout. AI semantic extraction reads the document holistically — it understands that a number next to "공급가액" is the supply value regardless of where on the page it sits. No per-supplier template configuration is needed. The trade-off: Clova OCR's per-page cost is lower (₩50/page vs ~₩87/page at $9/150 credits), but the integration cost and template maintenance overhead can eliminate that advantage at SME volumes.

What if I need on-premise document extraction for compliance reasons?

Cloud-based AI extraction isn't suitable for every Korean business. Companies handling sensitive personal data under Korea's Personal Information Protection Act (개인정보보호법) or subject to ISMS-P certification requirements may need on-premise solutions. Samsung SDS Brity (enterprise-grade, on-premise capable) and Korea Deep Learning's DEEP OCR+ (VLM-based, on-premise) serve this segment — but both are priced at the enterprise level. For less sensitive documents — supplier invoices, purchase orders, bank statements — cloud extraction is widely used by Korean SMEs across existing cloud ERP platforms like Ecount and JOBIS.