Affordable 適格請求書 Extraction
for Japanese SMEs Starting at $9/Month

The Invoice System Cost Spike Nobody Budgeted For

The qualified invoice system didn't just add new fields to supplier invoices — it added a per-document compliance verification step that didn't exist in Japanese accounting before October 2023. Before that date, a SME's invoice workflow was straightforward: open the envelope, enter the vendor name, date, and total into freee or Yayoi (弥生会計) accounting software, and file the paper. The consumption tax credit was calculated from the company's own books — not from supplier documents. Now every supplier invoice requires a T+13-digit registration number to be verified against the NTA's registry before the buyer can claim full input tax credit.

Under the new rules documented in the National Tax Agency's official qualified invoice system overview, every supplier invoice now requires six specific fields — including a T+13-digit registration number that must be verified against the NTA's registry, a breakdown of amounts by tax rate (10% standard / 8% reduced), and the consumption tax amount calculated separately for each rate. A single missing or unverified registration number means the buyer loses part of their input tax credit. That verification didn't exist before 2023. Now it's a per-invoice step.

The Japan Chamber of Commerce and Industry quantified this in two consecutive annual surveys. JCCI's 2024 survey of 3,149 member businesses found 48.8% reporting increased costs and 82.2% reporting increased administrative burden. JCCI's 2025 follow-up of 2,710 businesses showed those figures at 45.8% and 73.4% — improving but still affecting nearly three-quarters of respondents. The single largest source of new work, ranked by 74.8% of businesses: "supplier registration status verification and management."

In the same week the system launched, Teikoku Databank surveyed 1,494 companies: 91% expressed concerns about the system, 71.5% cited "increased workload" as their top worry, and only 65.1% said they were "on track with complying." One in three businesses was already behind — on day one.

The economic backdrop makes this timing particularly painful. Japan has approximately 3.36 million SMEs, accounting for 99.7% of all businesses and roughly 70% of employment, according to the Small and Medium Enterprise Agency's 2025 White Paper. These businesses are operating in a rising interest rate environment for the first time in 30 years, with yen depreciation pushing up import costs. Adding a manual compliance step per invoice — at a time when every cost line item is under pressure — is why 91% of surveyed businesses said the system worried them. It wasn't resistance to change. It was the recognition that someone in the office now had a new part-time job: reading registration numbers off supplier invoices.

What Japanese OCR Tools Actually Cost

The obvious fix — buy an OCR tool that reads these fields automatically — runs into a pricing wall built for enterprise buyers, not SMEs. Japanese invoice automation software falls into two tiers: accounting software with built-in OCR (limited but affordable) and dedicated AI-OCR tools (capable but priced for finance departments at listed corporations).

The accounting software path: freee's Starter plan (¥1,980/month, annual billing) includes unlimited AI-OCR receipt scanning within the accounting plan. MoneyForward Cloud starts at ¥1,078/month for Mini but charges metered fees for OCR beyond a threshold — and full invoice payable management requires additional modules that push the real cost to ¥3,000-5,000/month. Yayoi (弥生会計), Japan's longest-established accounting platform with roughly 3.4 million users, offers desktop plans from ¥11,000-33,000/year. All three have supported the qualified invoice system since October 2023. The catch: their built-in OCR was designed for receipts (レシート) — short, single-format thermal paper slips — not multi-format supplier invoices where the registration number, tax breakdown, and line items appear in different places on every vendor's document.

Moving up to dedicated AI-OCR: WingArc SVF Archiver Cloud — the document management arm of WingArc1st, a publicly traded Japanese software company — starts at ¥200,000 initial fee plus ¥35,000/month for 10 users. Its invoiceAgent AI OCR module requires a separate sales conversation. DynaEye 11 Entry AI-OCR, PFU's on-premise AI-OCR software, lists at ¥2,016,000 for the first-year license plus ¥168,000/year ongoing — that's ¥1.68 million commitment before the first invoice is read, according to published pricing. SmartRead (Cogent Labs) starts at ¥30,000/month with annual contracts and annual page limits. DX Suite (AI Inside) charges ¥30,000-200,000/month with metered item-based billing that penalizes long invoices with many fields.

An independent survey by Boxil, a Japanese B2B SaaS comparison site, aggregated pricing from 13 AI-OCR services as of October 2025 and found the typical cost range for a small-to-medium business: ¥100,000-200,000 initial + ¥10,000-50,000/month. Their survey of 613 actual buyers confirmed the median: about ¥30,000/month for a company with under 100 employees. The outlier at the bottom — invox受取請求書 at ¥980/month — exists, but its accuracy depends on operator-backed double-checking rather than pure AI extraction, which means turnaround time is measured in hours, not seconds.

For a Japanese SME processing 60 supplier invoices a month, the labor cost of manual data entry — at ¥2,040/hour (Tokyo minimum wage 2025) for roughly 5 minutes per invoice — comes to about ¥10,200/month. A ¥30,000/month OCR tool costs three times that. The automation doesn't pay for itself; it adds cost. This is the arithmetic dead-end that keeps most Japanese SMEs processing invoices by hand.

The Exchange Rate Arithmetic Japan's Enterprise Pricing Pages Don't Show

At ¥145 to the dollar (June 2026), a $9/month AI extraction plan costs ¥1,305 — less than 5% of what a domestic Japanese AI-OCR tool charges. The gap is not a currency anomaly the market will correct. It's the difference between a tool priced for self-serve small business buyers and a tool priced for enterprise IT procurement — and the exchange rate simply makes the structural gap visible.

A $19/month plan costs ¥2,755. A 30-credit pay-as-you-go pack at $30 costs ¥4,350 and never expires. This pricing architecture — low monthly fixed cost, no per-user fees, no minimum commitment — doesn't exist in Japan's enterprise document software market because that market was built around the assumption that the buyer is a finance department with a budget cycle, not a business owner with a credit card.

This isn't about buying cheaper because the yen is temporarily weak. It's that global AI extraction tools priced in dollars target the self-serve small business buyer — a sole proprietor, a freelancer, a small team — while Japanese domestic AI-OCR tools target the enterprise procurement cycle: initial license, annual maintenance, per-user add-ons, and mandatory onboarding support that shows up as a line item. The target buyer is different, so the entire pricing architecture is different. The exchange rate simply makes the gap visible.

The economics shift at SME invoice volumes. A tool that costs ¥30,000/month for 60 invoices works out to ¥500 per invoice. A tool that costs ¥1,305/month for 60 invoices works out to ¥22 per invoice. At 150 invoices a month — the volume of a mid-size SME with multiple suppliers — the domestic OCR cost stays roughly the same (most plans are flat-rate within a volume band), so the per-invoice cost drops to ¥200. The dollar-priced tool at the Pro tier ($19/month, 400 credits) processes those 150 invoices at ¥18 per invoice — with 250 credits left over for receipts or other documents in the same month. The subscription-vs-pay-as-you-go decision at these volumes depends on whether your invoice flow is steady (subscription wins) or seasonal — June and December spikes followed by quiet months (pay-as-you-go wins).

Dollar-Priced AI vs Yen-Priced OCR at SME Invoice Volumes

At three invoice volumes typical for Japanese SMEs, the monthly cost gap between dollar-priced AI extraction and domestic Japanese OCR is not marginal — it's an order of magnitude. The table below uses publicly listed prices: domestic tools at the Boxil survey median (¥30,000/month), enterprise AI tools at published entry pricing, and ImageToTable.ai at Personal ($9/month, 200 credits) and Pro ($19/month, 400 credits) — where one credit covers one page of extraction.

At 300 invoices a month, the domestic tool and the dollar-priced tool begin to converge — and that's where a Japanese SME should also factor in what each tool's extraction approach means for practical accuracy across a diverse supplier base. The ¥30,000/month domestic tool may require template definition per vendor layout. The dollar-priced AI tool uses Custom Column Extraction: you type the field names you need — "登録番号," "請求日," "10%対象金額," "消費税額" — and the AI locates each value by understanding what it means semantically, not by matching a pixel coordinate on a fixed template. For a company receiving invoices from 30 different suppliers in 30 different layouts, the template-free approach eliminates the ongoing maintenance cost that domestic tools' flat-rate pricing hides.

One thing the table can't show: the pay-as-you-go option. ImageToTable.ai's pay-as-you-go credits start at $30 for 300 credits ($0.10/credit), with bulk packs at $300 for 6,000 credits ($0.05/credit). Credits never expire. For a Japanese SME with seasonal invoice spikes — June settlement (六月決算) or December year-end (年末調整) — pay-as-you-go means the tool costs money only when invoices arrive, not in the quiet months when the accounting team is doing other work.

Does an English-Interface Tool Handle Japanese Invoices?

A dollar-priced AI extraction tool can read Japanese invoices — including the six NTA-mandated fields on a 適格請求書 — but its interface is in English. The decision for any Japanese SME comes down to whether the ¥28,695 monthly savings justify navigating a UI that wasn't localized for the Japanese market.

The first layer is about the extraction engine. ImageToTable.ai is built on a vision large model — the same class of AI that processes documents by understanding their visual layout and text content together, rather than running character-by-character OCR and hoping the pieces land in the right columns. This means it reads Japanese text as language, not as shapes. A 適格請求書 from a wholesale supplier — with the six required NTA fields spread across a header block, a line-item table, and a footer with the registration number — is read the same way a human reads it: by understanding what each piece of information is, not where it sits. The tool's Custom Column Extraction interprets the field names you write in English or Japanese and maps them to the semantically matching data on the page.

The second layer is about the interface. The tool's UI is in English. The column names you define can be in Japanese, English, or a mix — the AI reads both. The output spreadsheet will have the column headers you specified. But the menus, settings, and documentation are English-language. For a Japanese SME where the accounting clerk is comfortable reading basic English (a common profile among younger office workers in urban Japan), this is a manageable trade-off. For a company where no one reads English at all, it's a genuine barrier. Every Japanese SME needs to weigh this: is the ¥28,695/month savings worth navigating an English interface?

What Your Current Accounting Software Already Does — and Where the Gap Is

freee, MoneyForward, and Yayoi didn't build their OCR for supplier invoices — they built it for receipts. The distinction matters because the qualified invoice system turned supplier invoices into compliance documents, and none of Japan's major accounting platforms handle that document type at the extraction depth the new rules require.

freee's built-in AI-OCR was designed for receipts — thermal-paper slips from restaurants, taxi receipts, konbini purchase logs. These documents are short (typically 5-8 fields), single-format, and share a predictable structure: vendor name, date, amount, and a purchase category that freee can auto-classify. freee's AI-OCR reads the receipt, suggests a journal entry, and the user confirms — all within the accounting workflow. For this use case, freee's built-in OCR is genuinely good, and it's included in the plan with no per-scan metering.

An incoming supplier invoice (受領請求書) is a different document entirely. It arrives as a PDF attachment from a vendor whose invoice layout may have been designed in 2007 by someone who no longer works there. It contains 15-30 fields — registration number, invoice date, due date, line items with quantities and unit prices, subtotals by tax rate, consumption tax per rate, grand total, payment terms, bank transfer details — scattered across a layout that no two suppliers format the same way. freee's receipt-trained OCR can read text on an invoice, but it doesn't know which number on the page is the registration number vs. the PO number vs. the vendor's phone number. That distinction requires understanding what each field means, not just detecting the characters.

MoneyForward Cloud faces the same gap — its OCR is receipt-optimized and metered beyond a threshold, making it expensive for bulk invoice scanning. Yayoi's OCR capability is limited; most Yayoi users still key in invoice data manually or rely on their external accounting firm (税理士) to handle it — itself a cost item at ¥30,000-60,000/month for a typical SME retainer.

The gap, in short: freee and MoneyForward can already digitize your expense receipts. They were never designed to extract structured data from multi-format supplier invoices — the document type that the qualified invoice system turned into a per-item compliance task. For that, you need a tool whose extraction model was built for invoices from the start.

Two Ways to Fit Invoice Extraction Into a Japanese SME Workflow

How a Japanese SME integrates invoice extraction depends on what they already have in place. Two configurations cover the most common starting points:

Option A: Keep freee/MoneyForward for accounting, add ImageToTable.ai for invoice extraction. The accounting team continues using freee for daily bookkeeping, receipt scanning, bank reconciliation, and tax filing — all of which freee does well. For incoming supplier invoices, the team uploads PDFs to ImageToTable.ai with Custom Column names set to the qualified invoice fields: "登録番号 (T+13桁)," "請求日," "10%対象金額 (税抜)," "8%対象金額 (税抜)," "消費税額 (10%)," "消費税額 (8%)." The tool extracts all fields into one spreadsheet. The accounting clerk verifies the registration numbers against the NTA registry (a step they were doing anyway, but now it's a visual check of an extracted value rather than a manual read-and-type), then imports the spreadsheet into freee via CSV. The extraction cost: $9/month for up to 200 pages (Personal) or a $30 pay-as-you-go pack for 300 credits that lasts until used. This configuration adds extraction capability without changing the accounting system the team already knows.

Option B: Use ImageToTable.ai as the primary extraction layer for all incoming documents. A small business with 50-100 invoices a month and no complex payroll or depreciation tracking can run almost entirely on ImageToTable.ai plus a spreadsheet. The Pro plan ($19/month, 400 credits) covers invoices and receipts. The To Table mode — where defined column names become the spreadsheet structure — handles invoice extraction. The Collection Link feature lets external parties (a remote bookkeeper, a part-time経理 assistant, or even suppliers themselves) upload documents directly to the company's processing queue through a shareable URL without needing their own accounts. This matters for SMEs with distributed operations — a small construction company where the site supervisor in Saitama collects supplier delivery notes and the accounting person works from the Tokyo office, for example. The Collection Link gives the field person a single upload point that feeds directly into the back-office extraction queue.

For both options, the exit strategy matters as much as the start. If a company grows from 100 to 500 invoices a month and the English interface becomes a bottleneck, the data is always exportable as Excel — the tool doesn't lock you into a proprietary format. Upgrading to a domestic enterprise OCR later doesn't mean re-extracting historical data.

Frequently Asked Questions

Can ImageToTable.ai read handwritten Japanese invoices?

Yes — handwritten characters, including Japanese kanji, hiragana, and katakana, are within the vision model's recognition capability. Accuracy is lower than printed text (as with any OCR tool), but the semantic extraction approach means the AI uses context to disambiguate — if it reads "¥12,000" next to the characters "合計," it recognizes the field as the total amount even if individual strokes in "合計" are imperfect. For invoices that are entirely handwritten — common among small traditional suppliers in regional Japan — accuracy is best when the handwriting is clear and the fields are well-separated.

Does the tool verify registration numbers against the NTA database?

No. ImageToTable.ai extracts the T+13-digit registration number from the invoice — it doesn't validate it against the NTA's qualified invoice issuer registry. The accounting clerk still needs to run this check, which was a manual step before automation and remains one afterward. What changes: instead of reading the number off the physical invoice, squinting at a small-font footer, and typing it into the NTA lookup tool character by character, the clerk copies the extracted value from the spreadsheet. The time saving is in the data entry, not the validation step itself.

What about the 2026 transitional measure change — 80% deduction dropping to 50%?

From October 1, 2026, the transitional deduction rate for purchases from non-qualified invoice issuers drops from 80% to 50% of the input tax credit (and to 0% from October 2029). This means manually tracking which supplier invoices are from registered vs. unregistered issuers — and applying different deduction rates — becomes more consequential: getting it wrong costs twice as much in lost credits. Having the registration number extracted and visible in a spreadsheet column makes this classification step a filter operation rather than a per-document manual check. For more on the cost implications, see our breakdown of what the qualified invoice system costs Japanese finance teams by invoice volume.

Can the dollar-priced plan be paid for in yen?

Payment is in US dollars via credit card or PayPal. Most Japanese business credit cards handle dollar transactions automatically, using the card network's exchange rate (typically within 1-2% of the mid-market rate). At ¥145/$1, a $9 charge appears as roughly ¥1,305-1,330 on the statement.

Is there a free trial to test with actual Japanese supplier invoices?

Yes — the guest demo processes files without requiring sign-up, so you can upload a real 適格請求書 from one of your suppliers and see the extraction quality before committing to a paid plan. No credit card is required for the demo.

The Arithmetic That Changes the Buy Decision

The qualified invoice system added work that the Japanese accounting software ecosystem wasn't priced to absorb. freee, MoneyForward, and Yayoi handle receipt scanning within their plans because receipts are short, uniform, and the OCR was designed for them. The missing piece — multi-format supplier invoice extraction with registration number recognition — is what domestic AI-OCR tools solve at ¥30,000/month and above. That number makes sense for a finance department processing 1,000 invoices. It doesn't make sense for an SME processing 60. At 60 invoices a month, ¥30,000 for automation costs more than the labor it replaces. The only way the math works for an SME is if the tool costs less than the labor — and at $9/month (¥1,305), that condition holds at any volume above roughly 15 invoices a month.

The cost difference isn't about one tool being "cheaper." It's about what happens when a dollar-denominated AI product designed for self-serve small business buyers enters a market where the nearest equivalent is priced for enterprise IT procurement. The pricing tier gap isn't $10 — it's a structural difference in who the tools were built for.