How to Extract Japan Qualified
Invoice Data to Excel

3 New Fields per Invoice — The Extraction Problem Nobody Planned For

Before October 2023, a Japanese supplier invoice was a straightforward document: issuer name, buyer name, transaction date, item descriptions, quantities, unit prices, and a total. A finance clerk could key it into the accounting system in 2-3 minutes and move on.

Under the Qualified Invoice System — grounded in Article 57-2 of the Consumption Tax Act (消費税法第57条の2) — every compliant invoice must now carry six mandatory fields. Three of them are new: the issuer's registration number (登録番号, T + 13 digits), the total amount broken down separately by 8% standard rate and 10% reduced rate (税率ごとに区分した合計額), and the consumption tax amount for each rate category (税率ごとの消費税额等).

What this means in practice: the per-invoice data entry workload didn't grow by 50% — it roughly doubled. Each new field isn't just an extra number to copy. The registration number requires a format that doesn't appear on pre-2023 invoices at all. The tax-rate-split amounts require the clerk to scan the line items, mentally group them by rate, sum each group, and enter two subtotals instead of one. And the consumption tax field — mandatory now, in Japanese yen — must be entered separately even though it's mathematically derivable from the line-item totals.

As of March 2025, approximately 4.61 million businesses had registered as qualified invoice issuers, according to the National Tax Agency — 2.2 million sole proprietors and 2.41 million corporations. That means millions of businesses are now issuing invoices with these new fields, and every finance team that buys from them has felt the data-entry impact.

The Anatomy of a Qualified Invoice (適格請求書) — the 6 Fields Every Finance Team Needs to Capture

The NTA's official specification (refer to this NTA qualified invoice guide) defines exactly what a compliant document must contain. Here are the six fields, with the three post-2023 additions marked:

Fields 1, 4, and 5 are the three new data dimensions that turned invoice data entry from a single-total operation into a multi-field parsing task. The registration number follows a fixed pattern — T followed by exactly 13 digits — which makes it a strong candidate for automated extraction once you know what you're looking for. The dual-rate totals and consumption tax amounts, however, are positioned differently on every supplier's invoice template, making template-based OCR unreliable.

Why October 2026 Is the Inflection Point

The Qualified Invoice System includes a transitional period — and it's about to tighten:

Starting in October 2026, a ¥10,000 purchase at 10% tax from an unregistered supplier means ¥500 in lost input credit — up from ¥200 during the current phase. For a mid-size business processing 200 supplier invoices a month, with even 15% of suppliers unregistered, the difference between getting these fields right and getting them wrong compounds fast.

This is also the year when many foreign companies and exempt small businesses that delayed registration will face pressure from their B2B customers. Those customers need qualified invoices to maintain input tax credits — and if you can't extract the data from whatever invoice format you receive, whether it's a crisp PDF from a registered issuer or a handwritten slip from a sole proprietor counting on the transition period, you're leaving money on every transaction.

How to Extract Qualified Invoice Data into a Single Spreadsheet — Step by Step

ImageToTable.ai uses column-name extraction: instead of drawing bounding boxes around each field like template-based OCR tools, you type the column names you want — "Registration Number," "Total (8%)," "Consumption Tax (10%)" — and the AI locates each value anywhere on the page by understanding what it means, not where it sits. This matters for qualified invoices because every supplier positions these fields differently — one might put the registration number in the top-right corner, another in the footer next to the company seal.

Here's the extraction workflow for a batch of supplier qualified invoices:

The key advantage for qualified invoice processing is that the column names are the extraction instructions — the same spreadsheet header names you'd use for your internal invoice register. There's no separate template-design step, no training samples, and no per-supplier configuration.

Handling the 8% / 10% Tax Rate Split Automatically

This is where the Qualified Invoice System creates its most time-consuming manual task. Under the old system, a finance clerk entered one total per invoice. Under the new system, they need to separate line items into two tax-rate buckets — 8% reduced rate (for food and beverages, excluding dining out and alcohol) and 10% standard rate — sum each bucket independently, and record both subtotals plus their corresponding consumption tax amounts.

For a restaurant supply invoice with 15 line items mixing standard-rate kitchen equipment and reduced-rate food ingredients, that's a legitimate 3-4 minute data entry task. For 50 invoices: 3+ hours of tax-rate sorting.

ImageToTable.ai handles this through inferred columns — a mode where the AI reads document content and determines which category each item belongs to, even when that category isn't explicitly labeled on the document. You define a column like Category (options: 8% Reduced / 10% Standard), and the AI reads each line item, identifies whether the item is subject to the reduced rate based on Japan's consumption tax classification rules, and fills the result. This is not a rule-of-thumb guess: Japan's reduced rate classification — food and beverages not consumed on-premises and not alcoholic — is specific enough that a large language model can apply it from item descriptions alone.

The result is a single spreadsheet where separate columns for 10% and 8% totals are already populated, grouped correctly per invoice, without the finance team manually sorting through each line item.

Automated Registration Number Extraction: T+13 at Scale

The registration number is the most compliance-critical field on any qualified invoice — if it's missing or wrong, the invoice doesn't qualify for input tax credit. It's also the most structured: every valid registration number matches the pattern T followed by exactly 13 digits (e.g., T1234567890123).

But the structural predictability ends with the format. On an actual invoice, the registration number might appear:

• In the top-right header next to the company logo
• In the footer next to the company seal (印鑑)
• Below the issuer's name in a separate "registration number" (登録番号) line
• Inside a pre-printed box with other business registration details
• Completely absent — if the supplier hasn't registered or is still using transitional rules

A template-based OCR tool trained on one supplier's layout will fail on another's. An AI that reads document content semantically, on the other hand, can identify "T followed by 13 digits" wherever it appears — the same way a human clerk scans a page looking for that specific pattern. The extracted numbers can then be cross-referenced against the NTA's public Qualified Invoice Issuer registry (NTA Qualified Invoice Issuer Publication Site), which allows anyone to search by registration number or business name to verify that the issuer is legitimately registered.

For the finance team processing incoming invoices, this transforms registration number handling from "manually type a 14-character code and cross-check it on the NTA site" to "review the extracted T-number in your spreadsheet and spot-check a sample against the registry." The difference is minutes per invoice versus seconds.

What Japanese Accounting Platforms Cover — and the Gap They Leave

Japan's three dominant cloud accounting platforms — freee, MoneyForward Cloud, and Yayoi (弥生) — were all updated for the Qualified Invoice System before the October 2023 deadline. Each supports registration number management on outgoing invoices, dual-rate consumption tax display, and the 7-year electronic archive requirement under the Electronic Books Preservation Act (電子帳簿保存法).

But these platforms are designed for issuing compliant invoices, not for extracting data from the hundreds of invoices that arrive from external suppliers in PDF, paper, or screenshot form. A supplier sends you a PDF invoice — maybe a scanned copy, maybe a mobile photo of a paper original — and your accounting platform can't read it. The data entry still falls to a person with a keyboard.

This is the precise gap an AI extraction tool fills. It sits between "receiving a supplier invoice" and "the data is in your accounting system," handling the step that freee and MoneyForward don't: turning a non-machine-readable document into structured fields. For companies using international ERPs — SAP Japan, Oracle Japan, or Microsoft Dynamics 365 with Japan localization — the gap is even wider, since these platforms require manual configuration just to display qualified invoice fields, let alone extract them from incoming documents.

For teams already using Google Sheets as their invoice register — a common pattern among foreign subsidiaries in Japan that need to share data with headquarters — ImageToTable.ai's Google Sheets add-on writes extraction results directly into the active sheet. You specify the column names once, upload the invoices through the sidebar, and the extracted data populates the rows without leaving the spreadsheet where the rest of your reporting lives.

Frequently Asked Questions

Can AI extract handwritten qualified invoices?

Yes, but with realistic expectations. ImageToTable.ai's vision model can read handwritten registration numbers, amounts, and dates — even handwritten Japanese characters. However, accuracy on handwriting is inherently lower than on printed text, and badly degraded thermal paper receipts (common in Japan's restaurant and retail sectors) will produce lower-quality results than cleanly handwritten invoices. The tool is best suited for the mixed landscape most finance teams actually face: 80% printed/PDF invoices and 20% handwritten slips, rather than a pure handwritten-only workflow.

What if a supplier hasn't registered and their invoice lacks a registration number?

If an invoice from an unregistered supplier doesn't contain a registration number, the column extraction simply leaves that field empty — it won't hallucinate a number. During the current transition period (until September 2026), you can still claim 80% of the input tax credit on purchases from unregistered suppliers as long as proper bookkeeping entries are maintained. Starting October 2026, that drops to 50%. The tool doesn't make compliance decisions for you — it extracts what's on the document, leaving your finance team to apply the transitional rate rules based on the extraction results.

Does the tool handle era-date formats (令和X年Y月Z日)?

Yes. When you define a column as "Invoice Date," the AI extracts whatever date format appears on the invoice — including Japanese era dates like 令和6年5月28日. The extraction output preserves the original format. If your downstream process requires Gregorian dates, the tool's post-processing layer can convert era dates automatically during export.

Can I batch-process 100+ supplier invoices at once?

Yes. Batch processing — uploading multiple files simultaneously and receiving a single merged Excel spreadsheet where each row corresponds to one document — is a core capability. There's no per-batch file limit, though processing time scales with the total page count. A batch of 100 typical single-page invoices completes in well under 10 minutes. All files in a batch share the same column definitions, so you define your qualified invoice fields once and they're applied across the entire batch.

What about simplified qualified invoices (適格簡易請求書) that omit the recipient name?

Simplified invoices — used by retailers, restaurants, and taxi operators who sell to many unspecified customers — are valid qualified invoices that omit field 6 (recipient name). The tool handles these without issue: if the document doesn't contain a field you've defined as a column, that cell is left blank in the output spreadsheet. No error, no confusion.

Does the tool verify registration numbers against the NTA database?

No. The tool extracts the T+13 registration number from the invoice itself — it does not connect to the NTA registry to verify whether the number is active and valid. Verification against the NTA's Qualified Invoice Issuer Publication Site remains a separate step. What the tool eliminates is the manual typing of those 14-character codes into a browser-based lookup tool: you have the extracted T-numbers in your spreadsheet and can verify a sample or the full list against the NTA site in bulk.

Extract First, Verify Once

The Qualified Invoice System didn't create data entry — it made data entry materially harder by adding three mandatory fields that didn't exist before October 2023. Every finance team in Japan processes these fields already; the question is whether they're typing them in one by one, invoice by invoice, or extracting them programmatically.

The October 2026 transition from 80% to 50% input tax credit makes accuracy on these fields non-negotiable. The difference between extracting a registration number correctly versus mistyping one digit is real money — not in processing cost, but in lost tax credit on audit.