Feature
AI Processing
Turn PDFs and photos into clean, structured data you can trust. Clerko extracts the basics fast—and routes edge cases to review instead of guessing.
Fields extracted automatically
Anything the model isn't confident about gets flagged as Needs review—not auto-filled and silently wrong.
1 invoice · 5 fields extracted
Extract what's clear. Flag what isn't.
A simple pipeline designed to be trustworthy: good extractions go through, uncertain ones stop for review.
Detect invoice documents
Find invoices from your inbox, uploads, folders, and Telegram/WhatsApp bots—without relying on perfect naming or folder structure.
Extract key fields
Pull vendor, invoice number, dates, VAT, currency, and totals into structured data—no retyping.
Route exceptions to review
When something looks off, the invoice is flagged as "Needs review" instead of auto‑posting with a wrong value.
Hand off to reconciliation
Once fields are structured, matching invoices to bank transactions is straightforward—amounts and dates already parsed.
What gets extracted
In practice it's more than "vendor + total". Clerko extracts header fields, parties, tax breakdowns, and line items—then flags anything ambiguous for review.
Clerko marks the invoice as Needs review and highlights the specific field instead of guessing and moving on.
- Type invoice / receipt
- Invoice # number / reference
- Issue & due dates when present
- Service period start / end
- Seller & buyer names + addresses
- Tax IDs when present
- Bank details account + bank name
- Contact info email + phone
- Subtotal & total net + gross totals
- Currency for matching
- Tax breakdown rate + amount, multiple lines
- Special cases reverse charge / tax-exempt
- Line items description, qty, unit prices, totals
- Item tax rate + amount per line
- Payment terms method, terms, paid/unpaid
- Additional info notes captured from the document
AI processing FAQ
How do you handle accuracy?
Clear extractions are auto-filled. Anything ambiguous is flagged as "Needs review" so you stay in control.
What counts as an invoice?
PDFs or images that look like invoices (vendor, totals, dates). If something is borderline, it's surfaced for review.
Do you support multi-page PDFs and photos?
Yes. Multi-page PDFs and common image formats are supported. If a page is unclear, it's flagged.
What about duplicates?
If the same invoice is sent twice, it's designed to be detected and surfaced so you can confirm before processing.
Does AI processing use credits?
Yes—invoice processing uses credits. Inbox scans help you find invoices; processing extracts and structures the data.
Ready to process your first invoices?
Sign up and upload a PDF (or connect an inbox). You'll get structured fields and clear statuses for review.