How to Batch 150 Japanese Receipts
into Excel for Kakutei Shinkoku

Why One-at-a-Time Receipt Processing Falls Apart at Tax Season

The National Tax Agency reports that 23.39 million tax returns were filed for 2024 income, with 13.29 million filed via e-Tax from home. Of those, 8.24 million were filed by taxpayers themselves — not by accountants or payroll departments. Japan's freelance population, according to Cabinet Secretariat estimates, has reached 4.62 million, spanning sole proprietors (‮個人事業主‬), one-person company presidents, and parallel workers with side businesses. For every one of them, receipts are the most tedious part of the kakutei shinkoku process — not because any single receipt is difficult to process, but because the total volume resists linear effort.

The per-receipt data entry time for a Japanese ‮領収書‬ isn't uniform. A standard ‮コンビニ‬ (convenience store) receipt with clearly printed fields might take 30 seconds. A handwritten ‮領収書‬ from a small restaurant, with faded kanji, an era date (‮令和7年6月15日‬), and no tax-rate breakdown, can take 90 seconds or more. A user on Japan's Q&A platform Qiita tested freee's OCR on a handwritten ‮領収書‬ and reported approximately 20% accuracy on the amount field alone — four digits reduced to two, with only one correct. The immediate reflex is to blame the OCR, but the real problem is structural: a processing model built for one-at-a-time, per-receipt interaction doesn't fail on any single receipt. It fails when you multiply it by a year's accumulation.

What Japanese Accounting Software Gets Right — and Where It Falters at Scale

Japan's big three cloud accounting platforms — freee, ‮夥生‬ (Yayoi), and MoneyForward (‮マネーフォワード‬) — handle the ongoing bookkeeping workflow exceptionally well. They auto-import bank and credit card transactions, apply machine-learning to suggest the correct expense category (‮勘定科目‬), and generate the ‮青色申告決算書‬ (Blue Return Financial Statement) ready for e-Tax submission. For a freelancer who scans receipts as they arrive — one taxi ride, one Lawson receipt, one client lunch — the mobile app + OCR pipeline works.

The gap appears in February. MoneyForward's ‮クラウド経費‬ (Cloud Expense) supports batch upload of up to 50 receipt images at once for OCR processing, extracting date, vendor, amount, and registration number. freee's ‮ファイルボックス‬ (File Box) supports PDF, JPEG, and PNG uploads under 10MB each, with OCR that extracts key fields and highlights them for verification. Both are genuine productivity tools — for the steady-state workflow of one receipt at a time, spread across the year. But when you face a year-end backlog of 150 receipts accumulated in a shoebox, the platform designed for "scan as you go" becomes a bottleneck.

The distinction isn't about OCR accuracy. It's about what happens after extraction. In accounting software, each receipt produces one journal entry. Processing 150 receipts means 150 OCR-triggers, 150 category confirmations, 150 individual records. The output is a general ledger, not a spreadsheet — useful for the accounting software's internal engine but not designed for you to scan, sort, and validate 150 rows at a glance.

A batch extraction approach inverts this. Instead of uploading receipts into an accounting workflow one at a time, you define the columns you want — date, vendor, total amount, 10% taxable subtotal, 8% taxable subtotal, ‮但‬し書き description, expense category — and upload all 150 receipts as a single batch. The tool processes them in parallel and outputs one spreadsheet where each row is a receipt and each column is a field you named. This is column-name extraction: you tell the AI what you want to find (by naming columns), and it locates the matching values on each document by understanding their meaning — not by matching coordinates like template-based OCR. The extraction logic is format-independent because it's semantic, not positional. One column definition works across a ‮コンビニ‬ thermal slip, a handwritten restaurant ‮領収書‬ with a revenue stamp, and an Amazon Japan email PDF — no per-vendor templates, no per-format configuration.

What the 2024 ‮電子帳簿保存法‬ Changes About Your Receipt Workflow

Since January 2024, Japan's Electronic Bookkeeping Act (‮電子帳簿保存法‬) has been in full enforcement with a provision that directly affects receipt processing: any receipt received electronically — as a PDF email attachment, a website download, or an app screenshot — must be stored as electronic data. You cannot print it out and keep only the paper copy. The two-year grace period that allowed paper fallback ended January 1, 2024.

The law also sets three searchability requirements for electronically stored documents: they must be retrievable by transaction date, amount, and counterparty. A properly structured batch extraction spreadsheet automatically satisfies this: sort by date column, filter by vendor column, scan amounts by range. The spreadsheet itself, combined with organized digital folders of original receipt images, creates a searchable archive that meets the NTA's requirements — without a dedicated document management system. Paper receipts that you digitize by scanning fall under the Act's scanner preservation rules: once scanned, the physical original can be discarded if the scan was completed within two months plus seven business days of receipt and your office has adopted the required internal processing rules. For receipts originally received as email attachments or downloads, there is no paper original to keep — electronic storage is the only compliant option.

For a freelancer who files independently (‮青色申告‬ filers keep records for 7 years; ‮白色申告‬ filers for 5 years), the combination of batch extraction output and organized digital files addresses both the extraction need and the compliance requirement in a single step. If you need a deeper dive on the extraction mechanics for individual Japanese receipts, including exactly what fields to extract and how to set up columns for specific formats, see our guide to extracting ‮領収書‬ data into Excel.

Blue Return vs. White Return: Design Your Column Set by Filing Type

The choice between ‮青色申告‬ (Blue Return) and ‮白色申告‬ (White Return) doesn't just affect your tax bill — it determines which columns you need in your extraction spreadsheet. Getting this right at the column-design stage means your output maps directly to the forms you're filing, with no intermediate reformatting.

Blue Return offers a special deduction of up to ¥650,000 for filers who submit electronically via e-Tax and maintain double-entry bookkeeping (‮複式簿記‬). Starting from 2027, the maximum deduction increases to ¥750,000 for those who combine e-Tax with "excellent electronic records," while paper-only filers will be limited to ¥100,000. For a sole proprietor with ¥4 million in business income, the ¥650,000 deduction at the marginal tax rate of 20% (the bracket for income between ¥3.3M and ¥6.95M) translates to roughly ¥130,000 in income tax savings, plus roughly ¥65,000 in reduced residence tax (approximately 10% of the deducted amount). Real money for a solo operation — but it requires clean, granular records.

The practical implication for batch extraction: Blue Return filers need more columns because the NTA expects more granular reporting. White Return filers can use a simpler column set because the ‮収支内訳書‬ (Statement of Earnings and Expenses) doesn't demand the same level of detail as the ‮青色申告決算書‬ (Blue Return Financial Statement).

An expense category column can be set up as an inferred column: instead of typing the category manually for each of 150 receipts, you define a column name like "Category (Options: ‮旅費交通費‬/‮接待交際費‬/‮消耗品費‬/‮通信費‬/‮地代家賃‬/‮その他‬)" and the AI reads the ‮但‬し書き description to determine which standard expense category applies — filling it in automatically. A receipt from a taxi gets categorized as ‮旅費交通費‬, a restaurant receipt for a client meeting as ‮接待交際費‬. The extraction and the classification happen in one pass, not as separate steps.

A Two-Step Workflow: Upload Once, Get Your Spreadsheet

The workflow for batch-processing a year of Japanese receipts breaks into two core steps. Everything else is preparation and validation.

Step 1: Digitize Your Stack

For paper receipts, use your phone camera. Place each receipt flat on a dark surface, ensure the full receipt is visible from the ‮発行者‬ (issuer) name through the total amount, and avoid hand shadows. For thermal paper receipts — the kind issued by most Japanese convenience stores and restaurants — capture the image immediately upon receipt if possible. Thermal paper begins to degrade within 6-12 months, and the text on a one-year-old ‮コンビニ‬ slip may already be noticeably fainter than when issued. The Ministry of Economy, Trade and Industry (METI) recommends scanning thermal receipts within three months of issue.

For receipts received electronically — Amazon Japan order confirmations, utility PDFs, emailed invoices from service providers — save them to a single folder before upload. A consistent file naming convention like YYYY-MM-DD_Vendor (e.g., 2025-08-15_Yoshinoya.pdf) makes each file individually findable. If you have too many to rename manually, the extraction tool preserves the original filename in the output spreadsheet, giving you a traceable link from each data row back to its source file.

Step 2: Define Columns and Upload as One Batch

Name the columns you need — based on your filing type (Blue or White return) and the fields discussed above. Once defined, these columns work across every receipt format: a thermal slip from 7-Eleven, a handwritten ‮領収書‬ from a small restaurant, an email PDF from Amazon Japan. Upload all 150 receipts at once. The AI processes each one, locates the values matching your defined columns, and populates a single spreadsheet.

Step 3: Validate the Output Before Filing

No extraction tool achieves 100% accuracy on every receipt. The goal of batch extraction is to shift the human effort from "type every field" to "spot-check outliers." A targeted validation routine takes about 30 minutes for a 150-receipt batch:

1. Sort by Amount (descending) and visually scan the top 10 rows. These are your largest expenses and carry the highest audit risk. Verify each against its original receipt. A ¥45,000 expense misclassified as office supplies instead of equipment purchase could trigger a different depreciation treatment under NTA rules.
2. Scan the Category column for any "‮その他‬" (Other) classifications — the catch-all the AI uses when it can't confidently determine the category. Reclassify these manually.
3. Sort by Date and verify no entries fall outside the current tax year (January 1 through December 31). A January 2026 receipt in your 2025 filing is a correction waiting to happen.
4. For Blue Return filers: spot-check the tax subtotal sums. The 10% subtotal plus the 8% subtotal should roughly equal the total amount (minor rounding differences are normal). A discrepancy of more than a few hundred yen flags a potential extraction error on that receipt.

The output spreadsheet includes a source-file reference for each row, so you can click back to the original receipt image. This traceability satisfies the ‮電子帳簿保存法‬ searchability requirement — every data point is linkable to its source document.

Japanese-Specific Challenges That Compound at Scale

Several characteristics of Japanese receipts create friction in per-receipt processing. At scale — 100+ receipts — these friction points amplify from minor annoyances into hours of manual work. Understanding how each one behaves in a batch workflow determines whether your extraction output needs heavy post-processing or comes out tax-ready.

Era Dates: Converting ‮令和‬/‮平成‬/‮昭和‬ to Western Calendar

Japanese receipts use the era calendar: ‮令和‬ (Reiwa, began May 2019), ‮平成‬ (Heisei, 1989-2019), and occasionally ‮昭和‬ (Showa, 1926-1989) for older archived receipts. A standard extraction column outputs the date as written on the document — "‮令和7年8月15日‬" — which most accounting software can't interpret directly. Manually converting each date, at 30 seconds per receipt for 150 receipts, burns 75 minutes on a single field. A computed column — a feature that performs calculations during extraction rather than requiring post-processing — solves this: define a column with a conversion rule (add 2018 for Reiwa years, 1988 for Heisei, 1925 for Showa) and the AI outputs "2025-08-15" automatically. No separate conversion step in Excel.

Dual Tax Rates: When the Receipt Doesn't Split What You Need

Since October 2019, Japan applies a standard 10% rate to most goods and services and a reduced 8% rate to food items (excluding alcohol and dining out) and newspaper subscriptions. On a single ‮領収書‬ from a convenience store, you might see three subtotals: 10%‮対象‬, 8%‮対象‬, and 0%‮対象‬ (non-taxable items like stamps). Modern chain-store receipts from 2023 onward — required under the Qualified Invoice System (‮インボイス制度‬) — clearly label these breakdowns. Older receipts or small-vendor handwritten ‮領収書‬ often show only a single total.

When the receipt clearly splits the rates, AI extraction handles it reliably — two columns capture the two subtotals. When a receipt shows only a single total without a tax breakdown, the AI can't split what isn't printed. In that case, you'll need to calculate the split manually based on the contents: food items at 8%, other items at 10%. The key workload difference: in a batch workflow, only the receipts missing the breakdown require manual attention — not all 150. The clearly printed ones flow through automatically.

‮但‬し書き (Description): The Field Audits Are Built On

The ‮但‬し書き is the legally required description of what a payment was for — and it's the field most likely to be questioned during an NTA audit. A receipt that says "‮品代‬" (merchandise) is considered too vague. Acceptable descriptions include "‮お食事代‬" (meal expenses), "‮セミナー参加費‬" (seminar participation fee), or "‮消耗品費‬" (supplies). General-purpose OCR trained on English receipts often skips this field entirely — but a vision-language model that reads the document holistically can extract it because it's looking for the field by meaning, not by position. The ‮但‬し書き also feeds the inferred column that auto-classifies expenses into the correct ‮勘定科目‬ — a two-for-one workflow that removes the most manual step in receipt processing.

Handwritten Receipts and Faded Thermal Paper: Know the Limits

Handwritten ‮領収書‬ remain common in Japan for services like repairs, consulting, or freelance gigs. Modern vision-language models can read printed Japanese with high accuracy, but handwritten kanji — particularly cursive or stylized brush writing — is the hardest test case for any OCR system. Clear, block-style handwritten fields are generally extractable; highly stylized writing should be expected to need manual verification.

Thermal paper receipts from convenience stores and restaurants begin to fade within 6-12 months. Once faded to the point that a human eye can't read the text, an AI can't recover it either. The practical rule: digitize as soon as you receive, or at minimum before the year ends. A receipt that was legible in June may be a faint gray rectangle by the following February.

Frequently Asked Questions

Can I really upload 150 receipts in one batch?

Yes. Unlike some accounting software platforms that cap batch OCR uploads at 50 files per session, a batch extraction tool designed for this workflow accepts all your receipt images in a single upload — JPG, PNG, or PDF — and processes them into one output spreadsheet with one row per receipt. The processing time scales with volume: a 150-receipt batch completes in minutes, not hours.

How accurately does it split the 8% and 10% consumption tax?

When the receipt clearly labels the 10% and 8% subtotals separately — as most chain-store receipts from 2023 onward do under the Qualified Invoice System — AI extraction handles the split reliably. When a receipt shows only a single total without any tax breakdown (common on older or small-vendor handwritten ‮領収書‬), the AI can't extract a split that isn't printed. In these cases, you'll need to calculate the breakdown manually based on whether the items qualify for the reduced rate — but this applies only to the minority of receipts that lack the printed breakdown.

Does it work with handwritten ‮領収書‬?

Partially. AI vision models handle clear, block-style handwriting on receipts better than traditional template OCR because they process the document holistically — inferring text from context, not just pixel contrast. Cursive or stylized handwritten kanji, particularly on ‮領収書‬ written with a brush pen, will have lower extraction accuracy. If handwritten receipts represent more than 20-30% of your yearly volume, test a sample batch before committing to a workflow and budget extra time for manual verification on those receipts.

Can I import the output into freee, Yayoi, or MoneyForward?

Yes. All three major Japanese accounting platforms support CSV import for transaction data. Export your batch extraction spreadsheet as CSV, match the column order to the import format required by your software, and import one file instead of entering 150 individual receipts. freee's import template expects columns for date, amount, account item (‮勘定科目‬), description (‮摘要‬), and counterparty (‮取引先‬). MoneyForward and Yayoi follow similar column conventions. The batch extraction output maps directly to these fields.

Does an extracted spreadsheet satisfy NTA audit requirements?

The NTA does not require paper originals of receipts received electronically — the ‮電子帳簿保存法‬ mandates electronic storage. For paper receipts you digitize by scanning, the scanned copies serve as supporting evidence, and the NTA may request originals during an audit. A batch extraction spreadsheet with source-file references — each row linking back to the original receipt image — combined with organized digital folders satisfies the requirement that records be searchable by date, amount, and counterparty. The key is traceability: every expense claim on your tax return should be linkable to a source document you can retrieve on demand. The NTA expects you to maintain records for 7 years (Blue Return) or 5 years (White Return) — a well-kept spreadsheet and digital archive make retrieval a search, not a hunt.

A February Afternoon Instead of a February Weekend

The most common question Japanese freelancers ask each other in January isn't about deduction strategies or filing deadlines — it's "how many receipts do you still need to enter?" The answer almost always produces a grimace. Not because any single ‮領収書‬ is difficult to process, but because processing them one at a time turns a modest volume problem into a time sink that feels disproportionate to the actual complexity of the task.

Batch extraction doesn't eliminate the need to verify your data — that's a step every responsible filer performs regardless of workflow. It eliminates the creation of data from scratch, the part where you type "2025-08-15" for the 143rd time, squint at a faded ‮但‬し書き field to determine whether it says "‮お食事代‬" or "‮お菓子代‬", and recalculate the 10% vs 8% split because the thermal paper receipt from July is now barely legible. At 150 receipts, the math is straightforward: 3-5 minutes of column setup, a few minutes of processing, and about 30 minutes of targeted validation. Under an hour total versus most of a workday.

Test the workflow on your own receipts. See if a stack that would cost you an afternoon becomes a spreadsheet you can validate over coffee — and if the validation time feels like checking work rather than creating it from empty cells.