Drop your invoices. Get a clean Excel in seconds.
Multi-tax, line items, batch — all extracted. Chinese 增值税 / Japanese 適格請求書 / EU VAT all first-class. No signup, no template.
Built on enterprise-grade OCR + AI calibration trusted by Fortune 100 AP teams. Free for everyone.
Drag & drop invoice PDFs or photos here, or click to select
Up to 20 files · 25 MB each · pdf / jpg / png / heic / scanned
B2B invoice OCR that captures every line item
Most free invoice-to-Excel tools strip line items and emit only a header. ExcelMaster.ai's pipeline runs Aspose.PDF native text extraction first (digital invoices) with OCR fallback (scans + photos), then a structured AI calibration pass that pulls out the supplier + buyer parties, every line item with 8 fields (description / SKU / quantity / unit / unit_price / discount / tax_rate / line_total), and a multi-tax summary in EN 16931 categories. Server-side line-totals matrix self-check catches LLM number hallucinations before they reach your Excel.
Header + line items + tax summary, structured
Invoice header — 25 fields
Number, date, due date, PO, currency, supplier (name / address / VAT ID / tax ID / IBAN / email / phone), buyer (same shape), payment terms.
Line items — 8 fields each
Description, SKU, quantity, unit, unit price, discount, tax rate (fractional), line total. Source order preserved.
Multi-tax EN 16931 summary
Lines auto-grouped by tax rate into standard / reduced / zero / exempt / reverse_charge categories. Each summary row has rate, base, tax.
Line-totals matrix self-check
Server recomputes qty × unit_price - discount and uses the computed value when the LLM disagrees by more than 0.02. Industry-blank — competitors trust the LLM blindly.
Master sheet with SUMIF
Batch mode emits a Master sheet with all line items + auto-generated SUMIF formulas per tax rate. Drop a folder in, get tax-ready totals out.
Regional first-class support
Chinese 增值税专用 / 普通 / 电子发票 (CNY + 13/9/6/0% rates), Japanese 適格請求書 (T+13 registration), Korean 세금계산서 (사업자등록번호) all captured into supplier.tax_id.
How to convert invoices to Excel
- 1Upload invoices — drag & drop up to 20 PDFs or photos (digital + scanned both work).
- 2Pick layout — per-sheet, master sheet (with SUMIF tax summary), or both.
- 3Extract — native PDF text + OCR fallback + AI calibration. 5-15s per invoice.
- 4Download — verified line-totals + tax categorisation, ready for your AP workflow.
Why ExcelMaster outperforms alternatives
Line items — every field
Most free tools emit only the invoice header. We extract every line with 8 fields, in source order, with discount support.
Multi-tax EN 16931 native
Standard / reduced / zero / exempt / reverse_charge categorisation per EU norm. EU AP teams' baseline expectation.
Self-check matrix kills hallucinations
qty × unit_price - discount = line_total cross-check. We don't trust the LLM blindly; we audit it.
Chinese fapiao + Japanese 適格請求書 first-class
Asian regional invoice formats are blank for most Western SaaS. We treat them as first-class with locale-aware prompts.
Native PDF + OCR fallback
Digital PDFs use Aspose.PDF native text (fast + accurate). Scans and photos route through Aspose.OCR with the same downstream pipeline.
Master sheet auto-SUMIF
20 invoices → one master sheet with SUMIF tax-rate roll-ups already in place. No manual pivoting.
Tax season starts now.
ExcelMaster Add-in: drop a folder of invoices, get a tax-filing pack — categorised, summed, PEPPOL UBL exportable. Free with your account.
Auto-categorise for tax seasonFAQ
Is this really free?
Yes — no signup, no payment for the beta. Up to 20 invoices per batch. We run the same backend used by ExcelMaster's paid AI add-in.
What file types and formats?
PDF (digital + scanned), JPG, PNG, HEIC/HEIF, WebP, TIFF, BMP. Up to 25 MB per file, up to 20 invoices in one batch.
How does the line-totals matrix self-check work?
After the LLM extracts each line item, the server recomputes line_total = quantity × unit_price - discount. If the LLM-extracted line_total differs by more than 0.02, the computed value replaces it and a warning lands in the Validation footer of the workbook.
Do you support Chinese 增值税发票 (fapiao)?
Yes, first-class. Currency locked to CNY. 纳税人识别号 captured into supplier.tax_id. Tax rates 13% / 9% / 6% / 0% mapped to standard / reduced / reduced (modern services) / zero categories. Locale hint zh-CN routes the LLM call to kimi-k2.6 for stronger Chinese OCR text understanding.
What about Japanese 適格請求書?
Yes — the T+13-digit registration number is captured into supplier.tax_id. JCT 8% / 10% mapped to reduced / standard. ja-JP locale routes the LLM call to claude-sonnet-4.6 (lowest hallucination on Japanese).
Per-sheet or master sheet — which should I pick?
Per-sheet (default) gives you one sheet per invoice with full header + lines + tax summary — best for archival. Master sheet flattens all lines from all invoices into one tabular sheet with auto-SUMIF tax-rate summary — best for monthly tax filing or AP roll-up. Pick 'both' if unsure — costs nothing extra.
How accurate is the extraction?
Industry-standard OCR + AI calibration. Digital PDFs (text-native) score 98%+ on header fields and 95%+ on line items in our internal benchmark. Scans depend on resolution; 300 DPI scans land around 92%+. We never advertise 100% — every output ships with a Validation footer that shows where we corrected the LLM.
Do you store my invoices?
Original files auto-delete from our servers within 24 hours. Extracted JSON held only for the active session. We do not train AI models on user invoices. We are SOC 2 Type II compliant in process.
Can I validate VAT IDs against EU VIES?
Coming in Sprint B.5. Today the workbook records each supplier's vat_id; live VIES round-trip lookup ships next.
Will it work in ExcelMaster Add-in?
Yes. The Add-in adds 'AI auto-categorise all invoices for tax season' — drop a folder, get a multi-tab tax filing pack with PEPPOL UBL XML export. One-click import on the result page.
What if the AI calibration fails?
The pipeline always emits a downloadable Excel. If the LLM is unavailable or returns non-JSON, a regex scaffold (invoice number + date) gives you a usable starting point. The result flags 'degraded' mode.
Is my data sent overseas?
Tools-api runs on EU infrastructure. AI calibration calls our LLM proxy which routes to Anthropic / OpenAI-compatible providers per locale. We do not claim 'data never leaves your computer'; the pipeline does upload to our servers (TLS 1.3 in transit, AES-256 at rest, 24h auto-delete).