Questions · last reviewed September 1, 2026
Document extraction questions, answered
Ten pages written for the moment you know the problem but not the category name. Each one opens with the answer, hands you something you can use without buying anything — a schema, a checklist, a decision table — and names the term the industry uses so you can search for more.
5 pages
covered by one credit
50 pages
self-serve ceiling per document
25 credits
free every month, no card
Getting the data out
Turning documents into fields your systems can accept, without building a template per sender.
How do I stop manually entering invoices?
Declare the ten fields your ledger needs, check the arithmetic in code, and review exceptions instead of typing every invoice.
Schema + outputHow do I turn PDFs into JSON my code can validate?
Write the JSON Schema before you see the file, then validate shape and business rules as separate, mechanical gates.
Schema + outputHow do I process documents whose layouts are all different?
Model the fields, not the page. One schema per document class survives new senders; one template per vendor does not.
Decision table
Difficult documents
Scans, faxes, handwriting and tables that run past the bottom of the page.
How do I extract tables from bad scans?
Fix the image before blaming the parser, then ask for a typed array of rows so every row can be checked.
ChecklistHow do I extract line items when they span multiple pages?
Send the document as one unit, label summary rows, and reconcile row counts and sums against the totals printed on the page.
Schema + outputHow do I pull data from handwritten or faxed forms?
Constrain the form, fix the transport, and never auto-accept a hand-filled identifier or amount.
Decision table
Trust and operations
Catching errors before they cost money, deciding what a human reviews, and running the thing at volume.
OCR gets the numbers wrong — how do I catch it before it hits accounting?
Arithmetic, not eyeballs: line sums, tax math, duplicates and range checks catch digit errors before they reach the ledger.
Decision tableWhich extractions need a human to look at them?
Route by what a wrong value costs, use deterministic triggers alongside confidence, and track the review rate weekly.
Decision tableHow do I send documents with SSNs or PHI to an extraction API?
Ask what is stored, for how long and who can read it; then reduce exposure at the source, because OCR reads the whole page.
ChecklistHow do I process thousands of documents overnight?
Use a bounded worker pool, make each job idempotent, and do the credit and rate-limit arithmetic before the run, not during it.
Decision table
Try the answer on your own document
Define a schema, send one file, and read the JSON. 25 credits a month on the free plan, no card required.