How to Extract Data from Japanese Delivery Notes (納品書) into Excel — Across Multiple Supplier Formats
Japanese delivery notes (納品書 nōsho) vary by supplier. Learn how AI extraction handles multi-format nōsho data extraction without per-supplier template setup.
Why Japanese Delivery Notes Break Template-Based Extraction
Template-based OCR works by memorizing where a field sits on a page. Train it on Supplier A's delivery note layout — delivery date top-right, PO number below the header, item table starting at row 12 — and it correctly extracts those fields from every future delivery note Supplier A sends. Then Supplier B sends a delivery note with the delivery date on the bottom left, the PO number embedded in the reference line, and the item table using a vertical layout. The template fails. To handle 30 suppliers, you need 30 templates — and one of them breaks every time a supplier updates their ERP and the layout changes without notice.
Japanese delivery notes amplify the template problem in three ways that generic delivery note tools — built for Western formats — do not handle. The dual consumption tax rate splits amounts across two tax brackets the tool does not recognize. The date is written in the Japanese era calendar (令和, Reiwa) that a Western-trained date parser misreads. And the format diversity in Japan is extreme: delivery notes generated by different accounting and sales management systems — 弥生販売, freee, Money Forward, 楽々シリーズ, custom ERP exports — share no common layout, plus handwritten carbon-copy delivery notes (still widely used by repair services, construction suppliers, and smaller vendors) introduce handwriting recognition challenges on Japanese characters that most OCR tools were never trained for.
The core issue is not OCR accuracy. A modern AI-OCR engine can read individual characters on a clean printed delivery note with high confidence. The problem is that the tool does not know which piece of text corresponds to which field. It sees "令和7年6月10日" and "T1234567890123" and "¥140,000" as text strings but cannot map them to "Delivery Date," "Qualified Invoice Registration Number," and "Total Amount (10% Taxable)" without understanding what each field means. This is the gap between OCR and Custom Column Extraction — an approach where you define the columns you want by their semantic meaning ("Delivery Date," "PO Number," "Supplier Name") and the AI locates the corresponding value anywhere on the page by understanding what it represents, not by remembering where it was last time.
The 納品書 Fields That Drive Procurement and Receiving Workflows
Before you can extract delivery note data into Excel, you need to decide which fields to extract — and a Japanese delivery note (納品書) carries 8–14 fields that matter for receiving, inventory posting, and three-way matching, depending on whether the note also serves as a Qualified Invoice under the invoice system.
A delivery note in Japan has no legally mandated format. The National Tax Agency (国税庁, NTA) defines five required fields for tax documentation purposes, and the Qualified Invoice System (インボイス制度, inboisu seido) — effective since October 2023 — adds three more when the delivery note serves as a qualified invoice. But procurement teams need more than the tax minimum. They need fields that feed the warehouse management system, the ERP's goods receipt module, and the purchase order reconciliation workflow.
The minimum extraction column set for a Japanese receiving dock:
| Field (English) | Japanese Label Variations | Downstream Use |
|---|---|---|
| Supplier Name | 発行者名, 会社名, 納入元 | Match to supplier master record; validate against PO issuer |
| Delivery Note Number | 納品書番号, 伝票番号, No. | Primary key for goods receipt record; audit trail |
| Delivery Date | 納品日, 発行日, 取引年月日 | Receiving log; determines which accounting period the receipt falls in |
| Purchase Order Reference | 発注番号, 注文番号, ご注文番号 | Three-way match against open POs |
| Item Name / Description | 品名, 商品名, 品目 | Inventory posting; verify against PO line items |
| Quantity | 数量, 個数 | Inventory update; detect over/under-delivery |
| Unit Price (if shown) | 単価 | Price verification; some delivery notes omit this |
| Amount per Line (if shown) | 金額, 税抜金額 | Aggregate receiving value; some delivery notes show only totals |
| Registration Number (QIS) | 登録番号, T+13桁 | Validate Qualified Invoice issuer status for input tax credit |
| Tax Breakdown (QIS) | 10%対象, 8%対象, 消費税額 | Separate 10% and 8% taxable amounts for consumption tax filing |
| Recipient Name | 宛名, 納品先, 御中 | Confirm delivery to correct entity; cost center routing |
If a delivery note doubles as a Qualified Invoice (適格請求書), the registration number (登録番号) and consumption tax amount split by rate (税率ごとの消費税額) are no longer optional — missing them means the buyer cannot claim input tax credit on that purchase.
How the Qualified Invoice System Changed Delivery Note Extraction
Before October 2023, extracting delivery note data was a pure efficiency play. You saved time by not typing. After October 2023, with the introduction of the Qualified Invoice System (インボイス制度), delivery note extraction acquired a compliance dimension that most extraction tools still have not caught up to.
Under the old rate-classified invoice retention system (区分記載請求書等保存方式), a delivery note needed five basic fields: issuer name, transaction date, transaction details (with reduced tax rate indication), total amount by tax rate, and recipient name. The new Qualified Invoice System adds three more: the Qualified Invoice Issuer registration number (登録番号, format T + 13 digits), the applicable tax rate per category, and the consumption tax amount broken down by rate. A delivery note that serves as a qualified invoice must carry all of these — and the buyer must retain it for seven years to support input tax credit claims under Article 30 of the Consumption Tax Act.
This means the extraction tool now needs to capture fields that were optional a few years ago. More importantly, the dual-rate structure — 10% standard and 8% reduced for food and newspapers — means a single delivery note may contain two separate tax subtotals, each with its own taxable amount and its own tax calculation. An extraction tool that reads the total as a single number, without splitting it by rate, produces a spreadsheet that cannot support consumption tax filing.
The Japanese tax authority (国税庁) explicitly allows delivery notes and invoices to work together as a single Qualified Invoice. Under guidelines from the Japan Federation of Certified Tax Accountants' Associations, multiple documents with clear cross-referencing — such as a delivery note carrying line-item details and an invoice carrying the tax breakdown — can together satisfy the Qualified Invoice requirements. This means extraction output from delivery notes must be structured to pair with invoice data downstream. Two spreadsheets with no shared key are useless.
For a deeper look at how the Qualified Invoice System affects document processing across procurement workflows, see our guide to extracting data under Japan's Qualified Invoice System.
Extracting Delivery Notes Across Multiple Supplier Formats: A Step-by-Step Workflow
This is where the approach shifts from theory to practice. You have delivery notes from 弥生販売 exports, freee-generated PDFs, Money Forward printouts, handwritten carbon copies, and whatever custom format the supplier's internal system produces. The extraction method must handle all of them with one column definition, because maintaining 30 templates defeats the purpose of automation.
Define your extraction columns once
Name the fields you need — not by pixel position, but by what they mean. "Delivery Date (納品日)" tells the AI to find the delivery date regardless of where it appears on the page. "Item Name (品名)" locates every line-item description, whether it's in a table, a vertical list, or printed beside a barcode. "PO Number (発注番号)" finds the reference even if the supplier labels it "ご注文番号" or "注文番号." The column names you type become the headers of your output spreadsheet. Save this column set as a template — define it once, use it for every supplier.
Upload all the day's delivery notes as one batch
Scan paper delivery notes at the receiving dock. Upload supplier PDFs from email attachments. Photograph carbon-copy notes with a smartphone at the truck door. Drop everything into one batch — the AI processes PDFs, JPEGs from a phone camera, and scanned images in the same run. Mixed formats, mixed suppliers, one upload.
AI reads each delivery note by field meaning, not field position
For each document, the AI reads the entire page, identifies which pieces of text correspond to each column definition, and extracts the values. Header-level fields — Supplier Name, Delivery Note Number, Delivery Date — are repeated into every line-item row so the spreadsheet stays relational. Line-item fields — Item Name, Quantity, Unit Price — are extracted row by row from whatever table or list structure the document uses. A 弥生 format with a 6-column table works the same as a handwritten note with three fields in a vertical list.
Review and export to Excel
The output appears as a structured table — one row per line item, with header fields repeated and all values in standard format. Review edge cases: a handwritten quantity that was ambiguous, a date the AI flagged for manual confirmation. Once verified, export as XLSX. The file is ready for import into your WMS, ERP, or for matching against purchase orders in a spreadsheet. If your system accepts CSV, export that instead — no format conversion, no manual re-typing.
This approach works because the AI is not matching templates — it is matching meanings. A delivery note from 弥生販売 with "納品日" in the header and one from freee with "発行日" in the footer both contain a delivery date. The column "Delivery Date" finds both, without anyone needing to tell the tool where on the page to look. For a detailed walkthrough of batch delivery note processing, see our guide to combining packing slips and delivery notes into one receiving spreadsheet.
Files are processed securely and not stored.
Handling Japan-Specific Extraction Challenges
Japanese delivery notes present four extraction challenges that generic tools fail on — not because the OCR is inaccurate, but because the tools lack the Japan-specific context required to interpret what they read.
Japanese era calendar dates (和暦)
A delivery note dated 令和7年6月10日 is June 10, 2025 in the Western calendar. 令和 (Reiwa) began in May 2019. Before that was 平成 (Heisei, January 1989 – April 2019), and before that 昭和 (Showa, December 1926 – January 1989). Delivery notes from long-standing supplier relationships may carry any of these era names depending on when the note template was created. A tool that extracts "令和7年6月10日" as raw text leaves the spreadsheet user to manually convert every date. A Japan-aware extraction tool should recognize era names, perform the conversion, and output a standard date format — 2025-06-10 or June 10, 2025 — automatically. This is a 30-second-per-document correction when done manually. Across 100 delivery notes, that is 50 minutes on dates alone.
Dual consumption tax rates (複数税率)
Since October 2019, Japan applies a 10% standard rate and an 8% reduced rate to food (excluding alcohol and dining out) and newspaper subscriptions. A single delivery note from a food distributor may carry both rates — boxes of vegetables at 8% and cleaning supplies at 10%. The delivery note lists a subtotal and tax amount for each rate category. An extraction tool that sums everything into one "Total" column loses the tax split, making the data useless for consumption tax reporting. The column definition should include separate fields: "10% Taxable Amount (10%対象)," "8% Taxable Amount (8%対象)," "10% Tax Amount (10%消費税)," and "8% Tax Amount (8%消費税)."
Handwritten and carbon-copy delivery notes
Handwritten delivery notes (手書き納品書) are still common in Japan — used by repair services, construction material suppliers, small-scale manufacturers, and independent contractors. A carbon-copy delivery note (複写式納品書) produces multiple layers: the top sheet goes to the recipient, the second stays with the supplier. By the time the warehouse receives it, the copy may be a third-generation carbon impression with faded characters, smudged ink, and a company seal (印鑑) stamped over text. General OCR trained on clean Western layouts struggles with all three — the irregular handwriting, the degraded carbon transfer, and the red ink stamp overlay. A visual AI model that reads the document as a whole image, understanding the text as part of a document context rather than isolated character grids, can extract these fields. Accuracy on handwritten Japanese is lower than printed text — budget a quick review pass on any handwritten delivery note — but the gap between "AI reads 80% correctly and you fix 20%" and "you type 100% from scratch" is the gap that matters. For more on handling handwritten documents, see our guide on extracting handwritten delivery notes to Excel.
Company seals and stamps (印鑑 / 角印)
Most Japanese delivery notes carry a company seal — typically a square 角印 (kakuin) stamp in red ink, placed over the issuer name and address. The stamp is a mark of authenticity, not a data field, but its red ink overlay confuses OCR engines that try to read the text underneath. Worse, some suppliers stamp over the delivery date or the total amount line — the exact fields you need most. A visual AI model trained on document structures can distinguish between overlaid seal impressions and the text beneath them, extracting the original text while ignoring the seal. This capability is table stakes for Japanese document extraction; if a tool cannot handle 印鑑, it cannot handle Japanese delivery notes.
For a broader look at how the Japanese document processing ecosystem handles these challenges, including a cost comparison of tools ranging from freee's built-in OCR to dedicated AI extraction services, see our analysis of affordable extraction options for Japanese SMEs.
FAQ
Does this work with handwritten Japanese delivery notes?
Yes — the visual AI model underlying column-name extraction reads both printed Japanese text and handwriting, including cursive and mixed kanji/kana script. Accuracy on handwritten fields is not identical to printed text, particularly for complex kanji with many strokes or heavily degraded carbon copies, but the review step catches edge cases rather than requiring complete manual re-entry. If a delivery note is entirely handwritten, expect to verify the extraction output rather than trust it blindly.
Can the tool handle the Japanese era calendar (令和/平成/昭和) for dates?
Yes. When you define a column like "Delivery Date (納品日)," the AI recognizes era-based dates and converts them to the Western calendar in the output. 令和7年6月10日 becomes June 10, 2025 or 2025-06-10, depending on your preferred format. This conversion happens automatically during extraction — no manual date correction in Excel afterward.
What if some delivery notes show prices and others do not?
In Japanese procurement, pricing is not mandatory on a delivery note (納品書). Some suppliers include unit price and amount per line; others — particularly those who bill monthly with a consolidated invoice — omit pricing entirely and show only item names and quantities. Your column definition can include fields like "Unit Price (単価)" and "Amount (金額)" — the AI will extract them when present and leave the cell empty when not. The spreadsheet remains consistent, and price matching happens downstream against the invoice.
Can I extract data from delivery notes that also serve as Qualified Invoices?
Yes. Define columns for the Qualified Invoice registration number (登録番号), the 10% tax breakdown (10%対象額, 10%消費税額), and the 8% tax breakdown (8%対象額, 8%消費税額). The AI extracts all fields into the same spreadsheet row. If a delivery note does not carry Qualified Invoice fields, those columns remain empty — the extraction handles both types of delivery notes in the same batch without separate templates.
Does the extraction handle delivery notes written vertically (縦書き)?
Some Japanese delivery notes — particularly those using traditional formats — arrange text vertically, with recipient information on the right side and issuer information on the left. The AI's visual language model reads the document as an image and understands layout structure, not just text flow direction. It locates fields by their semantic role regardless of orientation. That said, accuracy is higher on horizontal (横書き) layouts; if a supplier consistently uses vertical formatting, include a few samples in your initial test batch to evaluate accuracy on that supplier's specific layout.
How does this compare with the OCR built into freee or Money Forward?
The OCR engines built into freee, Money Forward, and 弥生 (Yayoi) were designed primarily for receipts (レシート) — short, single-format thermal paper slips with a known structure. They excel at reading a convenience store receipt and auto-categorizing the expense. A multi-format supplier delivery note — particularly one with a line-item table, dual tax rates, and mixed horizontal/vertical layout — is a fundamentally different document. The built-in OCR may extract some fields correctly but typically fails on the line-item table structure, the era date conversion, and the tax rate split. Dedicated delivery note extraction that reads by field meaning rather than document template is designed for this exact scenario.
What WMS or ERP systems can the Excel output be imported into?
The XLSX or CSV output can be imported into any system that supports spreadsheet import for goods receipt — including SAP, Sage, Microsoft Dynamics, and Japan-specific platforms widely used in Japanese procurement: 弥生販売, freee, Money Forward, and 楽々シリーズ. If your system supports CSV import for receiving records, the output is ready without additional transformation.
The Japanese delivery note is not harder to read than any other delivery note. What makes it harder is that most extraction tools were trained on a different set of assumptions about what a delivery note looks like, what a date looks like, and what tax information belongs on the page. When the tool reads by field meaning instead of field position, those assumptions no longer matter. The same column definition that extracts data from a 弥生販売 PDF works on a freee printout, a Money Forward export, and a handwritten carbon copy from a supplier who still uses a company seal and a fountain pen. The spreadsheet comes first. The manual typing stops. The three-way match starts.