How extraction works
Between a document arriving and data appearing, Merchaint does three things in order. It works out what the document is, picks a schema to read it with, then reads the fields. Knowing that order answers most "why did it read that?" questions.
- A walkthrough video, then why the order matters
- The three steps, and which schema wins when there is a choice
- How to tell which step went wrong from what you see
- FAQ
01 · Video
Feature walkthrough
02 · Why
Why does this matter?
If you know the order Merchaint works in, you can tell which step went wrong. A blank field on a page that clearly shows a value is a different problem from a document read as the wrong type — and they need different fixes.
Why doesn't it ask me what each document is?
Because for most documents the answer is obvious from the document itself, and asking every time would remove the point. Merchaint decides, and you can override it afterwards.
03 · What it means for you
The three steps
| Step | What happens | What you see |
|---|---|---|
| 1. Work out what it is | Merchaint decides the document type | The type next to the status — Invoice detected |
| 2. Pick a schema | It chooses the schema for that type | The schema decides which fields are looked for at all |
| 3. Read the fields | It pulls values into those fields | What you see under Extracted data |
Which schema wins?
Merchaint uses the schema for the type it detected. If that's wrong, you decide instead. Reprocess lets you leave it on Auto-detect Schema or pick one yourself. A schema you pick always beats Merchaint's guess.
Which step went wrong?
| What you see | Which step | What to do |
|---|---|---|
| Right fields, wrong values | Reading | Reprocess and re-read it |
| Fields you didn't expect, or missing ones | Schema | The wrong schema was used — reprocess with the right one |
| The document type is plainly wrong | Working out what it is | Reprocess and pick the schema yourself |
| Nothing read at all | It couldn't be read | See Fix a document that was read wrong |
04 · FAQ
Why did it read the wrong number?
Usually because the wrong schema was applied, so it looked in the wrong place. Check the document type first, before assuming the reading failed.
Can I make it always use one schema for a supplier?
Not from the document itself. You can reprocess documents one at a time. If a whole supplier is consistently wrong, that's a schema question for your admin.
Common questions
Only if you turn on Force re-extract. Left off, it reuses what was already read.