Advanced · Work and documents · module 107 of 155
Data from documents into a table
Have data from invoices, forms or emails transferred into a fixed schema and check it systematically.
After this lesson
You design a schema, leave missing fields as unknown and check the result with rules and spot checks.
In view · Step by step
From document to reliable table
Fields, format and what happens with an empty or unreadable field.
Check amounts, dates and numbers, and compare a few rows with the original.
A missing or unusual field is flagged, not filled in.
Part 1 of 3 · Fixed schema
Your preference for less movement is on. Choose each part yourself.
Define the schema first
Decide in advance which fields you need, with name, meaning and format. For example supplier, invoice number, amount excluding VAT and due date in the form YYYY-MM-DD. Say what should happen if a field is missing or illegible: write unknown. Without that agreement, a model sometimes fills in a likely value, such as a usual payment term of thirty days. A fixed schema makes the result comparable and checkable.
Check with rules and spot checks
Do not let the AI confirm by itself that everything is correct. Use simple rules: is every amount a number, is the date in the right format, does an invoice number appear only once, does the VAT amount match the total? Then compare a sample with the original, for example every tenth row and every row with a striking value. Watch out for typical reading errors: comma and full stop in amounts, 1 and 7, or a shifted column. An empty cell is not zero.
Make the step repeatable
Keep the schema, the prompt, the check rules and a few sample documents with the correct result. That way, after a change to the prompt or the tool, you can test again whether the result is still correct. Work with documents you are allowed to process in the chosen tool; invoices and forms may contain personal or business data. Never let an extracted table automatically start payments or bookkeeping entries without approval and a check of the amounts.
New practice example
From explanation to a usable result.
You extract data from documents using a fixed schema. What is missing stays unknown; rules and spot checks catch reading errors.
Fictional practice situationFictional registration by email: “I’m Mira and I’d like to sign up for the course on Tuesday. My phone number will follow.” Schema: first name, course day, phone, paid (yes or no).
View a possible solution
This is how it can be done
First name: Mira. Course day: Tuesday. Phone: unknown, will follow. Paid: unknown, not mentioned. An extraction that fills in “paid: no” turns missing information into a fact.
Why this works
Unknown and no are different values. A fixed schema with an explicit value for unknown prevents the table from saying more than the source.
Make it slightly more difficult
Think of two validation rules for this schema, for example about the format of the phone number. Which error does each rule catch?
Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.
Your workplace · learning by doing
Make something. See what changes.
Checking the AI connection… You can already view the written example.
Present your assignment to Ende.
Read the answer, change one thing, and compare.
Previously held
New version
You remain in control
What do you want to improve in your next attempt?
On this device only. AI answers remain only in this page memory. This is self-review, not a grade.
Why this approach works
Unknown and no are different values. A fixed schema with an explicit value for unknown prevents the table from saying more than the source.
A step further
Think of two validation rules for this schema, for example about the format of the phone number. Which error does each rule catch?
Try & discover · no obligation
Place source and concept side by side
Adjust the values in the concept. See what difference you would give back to the writer.
Fictional invoice: Atelier Noord, invoice F-2026-118, € 1.240,00 excluding VAT. There is no due date on the invoice. Compare with the values an extracted table might contain.
| Part | Source | Concept | Check |
|---|---|---|---|
| Supplier | Atelier Noord | Atelier Noord | Same |
| Amount excluding VAT | € 1.240,00 | € 1,24 | Difference |
| Due date | Unknown | 30 days | Difference |
Targeted feedback you could provide
- Amount excluding VAT: This amount differs from the invoice. Look at the full stop and the comma: a reading error changes the amount by a factor of a thousand. Ask to retain the source value "€ 1.240,00", unless a change is confirmed.
- Due date: This value is not known in the source. Do not fill in the empty field yourself. Ask to leave this as unknown.
You are only changing three fixed fields of a fictitious concept. This is not a full document analysis or a check of a real upload.
This trial works locally with fictitious data and does not count towards your lesson completion. It starts anew with a new lesson or reload.
A prompt to get you started
Extract these fields from the fictional invoice below: supplier, invoice number, invoice date (YYYY-MM-DD), amount excluding VAT, VAT amount and due date. Only use what is literally in the text. If a field is missing or illegible, write “unknown” and do not fill anything in. Give a table with an extra remarks column. Invoice: Atelier Noord, invoice F-2026-118 of 14 October 2026, total excluding VAT € 1.240,00, VAT 21% € 260,40.
Apply it
An extraction of twenty fictional invoices shows an amount of 1,24 euros instead of 1.240 euros for three invoices.
Which approach best fits this situation?
- Correct the three amounts and assume the rest is correct.
- Ask the AI whether the amounts are correct and accept its confirmation.
- Add a check rule for amounts, compare the deviating rows with the original and test the extraction again.
With JavaScript, you can check your approach and receive immediate feedback.
Test your insight
An invoice has no due date. The table shows “30 days”. What do you do?
- Leave it: thirty days is usual.
- Reset the field to unknown and ask about the due date or look it up in the agreements.
- Have the document processed again until a date appears.
Extra practice
Run the extraction on two fictional invoices. Check each field against the original and note which check rule could have caught an error.
Supplier: Atelier Noord. Invoice number: F-2026-118. Invoice date: 2026-10-14. Amount excluding VAT: 1.240,00. VAT: 260,40. Due date: unknown, not stated. Check rule: 21% of 1.240,00 is 260,40, so the VAT amount matches the amount. I would reset a filled-in due date to unknown.
