Extraction Guide

Upload your invoices, say what you need in plain words, and download the spreadsheet.

Animated example prompts, from a simple field list to detailed business rules. For example: “Extract invoice number, invoice date, vendor name, net amount, tax, total”.

How it works

The whole workflow, in three steps

  1. Upload

    PDFs and images, exactly as you received them. One PDF holding fifty invoices is fine, and invoices, credit notes and statements can go in together. Tell our AI what to ignore, email cover sheets or remittance advice for instance, and it will.

  2. Say what you need

    A list of fields, the job you are doing, or nothing at all. Where your instructions or the documents leave something genuinely open, our AI can stop and ask you rather than decide on its own.

  3. Download, and know what to verify

    Every row carries the file and page it came from. A value or a row the panel of AI agents cannot agree on is flagged as Review Needed, with what to check, and in the results viewer one click opens that row beside the page it came from.

What comes back

The spreadsheet, and how to check it

Excel, CSV or JSON

In Excel, numbers come through as numbers and dates as dates, so your formulas and pivot tables work straight away. CSV and JSON carry every value as text.

Two columns after your data

  • Source File: the file and page each row came from.
  • Review Needed: what to check, explained below.

If you need the file exactly as you asked for it, either column can be left out in Preferences.

Failed pages

A page that could not be processed is reported, never skipped in silence: your dashboard shows which pages, and you are not charged for them.

Review Needed

  • A value or a row the panel of AI agents cannot agree on is flagged as Review Needed, with what to check.
  • Shown in your dashboard when the extraction finishes, and in the Review Needed column of your spreadsheet, with the warning cells and the data cells they refer to highlighted in Excel.
  • In Preferences you can leave the column out of your downloads, change or remove the highlighting, and choose the language the warnings are written in.

Example Review Needed warning

Affected field

Quantity

Warning

The extracted Quantity is 9, matching the printed value, but a handwritten value of 10 appears beside it. Verify whether the extracted Quantity should use the printed value or the handwritten value.

Source File

[Page 1] supplier-invoice.pdf

The results viewer

Check results, on any completed task in your dashboard, opens your spreadsheet beside the documents it was read from. Click the file name on any row and the page it came from opens beside it, ready to compare.

  • Spot-check a big job quickly

    Take a sample of rows and work through it. Each row and its page are already matched up, so there is no hunting for the right document.

  • A Review Needed value is one click from its invoice

    Click the file name on its row, compare the value with the page beside it, confirm or correct it, and move on.

  • 48 hours for the documents, 90 days for the results

    Source documents can be viewed for 48 hours after upload, when they are deleted. Your results stay available for 90 days unless you delete them earlier.

AI questions

Our AI can stop and ask you

With AI questions on, our AI reads your documents before extracting and, where your prompt or the documents leave something genuinely open, stops and asks a brief question in the task's conversation, then continues with your answer. If you step away, the task waits and emails you. When planning completes, your answers are written into a prompt you can save to your Prompt Library, so similar tasks do not need to ask again. AI questions are off for a new account and switched on in a click, on the task or as your default in Preferences.

With AI questions off: Prompt Notes

With AI questions off, what our AI found unclear becomes Prompt Notes instead: it makes the conservative, accounting-friendly call, writes down what it assumed, and gives you a line to copy into your prompt to settle it next time. They are shown in your dashboard when the extraction finishes; none means no assumptions were reported.

Prompts

Say what you need, in plain words

Start with a one-liner. Add detail when you want exact control. A prompt describes the spreadsheet you want, not the documents you have: the columns, the formats, what each row is, what to leave out; our AI reads the documents for itself, whatever their layout. Where your prompt leaves something open, it asks or makes the conservative call and tells you, as above.

Suggest prompt Not sure where to start? Add your files, click Suggest prompt, and let our AI write the prompt from what it finds in them. Adjust it, save it, run it.

A list of fields

Your prompt
Extract invoice number, invoice date, vendor name, net amount, tax, total

You get

Invoice #Invoice dateVendorNetTaxTotal

Exactly those columns, one row per invoice.

Your goal, then the fields

Your prompt
I'm processing invoices for payment. Extract invoice number, date, vendor, amount due, payment terms

You get

Invoice #DateVendorAmount duePayment terms

The fields you listed, with the edge cases and anything you left unsaid handled the way paying invoices calls for.

Line items

Your prompt
Extract line items: description, quantity, unit price, line total

You get

Invoice #DescriptionQuantityUnit priceLine total

One row per line item, with the invoice's details repeated on each row.

Name the job

Name the job the data is for, and our AI handles the edge cases in your documents, and anything you left unsaid, the way that job calls for: which total counts, how a credit note is shown, which page of a bundle to read. Any of these, in your own words:

  • I'm processing these invoices for payment approval
  • I need this for our quarterly VAT return
  • I'm doing a line-item spend analysis across vendors
  • I'm reconciling bank statement transactions against our invoices
  • I'm extracting monthly utility charges across our sites
  • I'm preparing payroll data for our monthly pay run

Exact rules, for a repeatable, auditable workflow

Name the fields, formats and rules once, save the prompt to your Prompt Library, and every batch comes back the same way. More of what a prompt can carry is in the reference below.

Your prompt

I'm preparing AP data for our month-end close.

Extract:

  • Invoice Number (alphanumeric)
  • Invoice Date (YYYY-MM-DD)
  • Vendor Legal Name (prefer extracting from footer)
  • Net Amount (pre-tax invoice total)
  • VAT Rate (if no VAT is listed use 0, use Excel type percentage)
  • VAT Amount (if no VAT is present use 0)
  • Total Amount (invoice final total)
  • Document Type (classify as Invoice or Credit Note)
  • - For credit notes prefix Invoice Number with 'CR-' and show amounts as negative.
  • - One row for each invoice or credit note.
  • - Skip any pages that are email cover sheets or summary pages.

You get

One row per invoice or credit note, your date and number formats applied, cover sheets and summary pages skipped, credit notes negative: the same spreadsheet every month.

Prompt Library

One prompt per kind of spreadsheet you want, not per vendor or layout: our AI adapts to the documents. Save as many prompts as you have jobs, one per client or per document type, and apply the right one with a click. Keep separate prompts where what you want out differs:

  • Different jobs, such as VAT reporting and expense tracking
  • Different document types, such as invoices and bank statements
  • A client's own output format
  • A vendor that needs rules of its own

That is the whole workflow. What follows is the detail, for exact control.

Uploads

What you can upload

PDF files

  • Regular and scanned PDFs
  • Up to 150 MB per file
  • Up to 5,000 pages per file
  • One file can hold many invoices

Images

  • JPG and PNG, scans and phone photos included
  • Up to 5 MB per file
  • One invoice per image

Batches

  • Up to 6,000 files in one task, and several tasks can run at once
  • Invoices, credit notes and statements can go in together, with different rules for each
  • Tell our AI what to ignore, email cover sheets or remittance advice for instance, and it will

Need more than 6,000 files in one task? Message us.

Prepare files with PDF Tools

PDF Tools, in your dashboard: remove pages, split files too big to upload, rotate pages, merge files, keep only the odd or the even pages. Free, and nothing there uses credits. Files are processed in your browser and never leave your device.

Structured Prompt

Exact headers, in exact order

An alternative way to write your prompt: name each column yourself. Your spreadsheet uses exactly these headers, in this order, and any column can carry its own instruction, so a file can match the layout another system imports, your accounting software's for instance. With the free-text Prompt, column names and order are usually consistent, but our AI may adjust them unless you say otherwise.

PromptStructured Prompt

Choose Structured Prompt from the switcher above the prompt box.

Column headers

Each header becomes the exact column name in your spreadsheet. Prefer names that say what the value is.

Good

  • Invoice Number
  • Supplier Name
  • Total Amount

Avoid

  • Column 1
  • Field A
  • Data

An instruction per column (Optional)

Which value, from where, in what format, for that column alone.

  • The date the invoice was issued, not the payment due date
  • No currency symbol, 2 decimal places
  • Extract from the table beneath 'Description'

Additional instructions (Optional)

For the whole extraction: your goal, what each row represents, formats and handling rules.

  • I'm preparing AP data for month-end close
  • One row per invoice
  • Ignore email cover pages and remittance advice
  • For credit notes, show amounts as negative
  • Format all dates as YYYY-MM-DD

Prompt instructions

What a prompt can say

Examples of the instructions a prompt can carry, in your own words, on their own or together.

Fields and columns

  • Extract 'Invoice Number', 'Invoice Date' and 'Total'
  • Use the column header 'Supplier_Name'
  • Order the columns as: Date, Vendor, Invoice Number, Total
  • One row per line item, repeat 'Invoice Number' on each row
  • Join all line item descriptions into one cell, separated by semicolons

Business rules

  • Hint'Product Code' is in the 'Description' column and begins with 'SKU-'
  • DefaultIf 'Tax Amount' is missing, set it to 0
  • FallbackFind 'PO Number' in the header; if it is not there, use 'Reference'
  • ConditionalIf 'Currency' is 'USD', use 'State Tax'; if 'EUR', use 'VAT'

Documents and pages

  • From statements of account, extract each invoice listed as a separate row
  • Extract from 'Tax Invoice' pages only, not the attached delivery notes
  • If a value is crossed out and handwritten, use the handwritten value

Formats and classification

  • Ensure all currency fields have 2 decimal places
  • Store 'Invoice Date' as text, not as an Excel date
  • Add an 'Expense Category' column: classify as 'Office Supplies', 'Software', 'Travel' or 'Utilities'
  • Add a 'Payment Priority' column: 'Urgent' if overdue or due within 7 days, else 'Standard'

Excel shows numbers and dates according to your own Excel settings.

Try it on your own invoices

50 free pages every month. Sign up with an email address: no subscription, no credit card.

Start extracting for free