Documents and AI
Upload documents, what the AI reads from them and how to review what it understood.
Uploading documents
- Uploading your first documentStart hereWhat happens when you drop it in, and what to decide next.
- Photos taken on a phone at workStart hereThe delivery note, the meter, the damage, the plate: when a photo is a document and when it is a memento.
- Documents that arrive by messagingStart hereThe delivery note someone sends you on WhatsApp exists, and you must decide what to do with it.
- Which formats and what sizeGoing furtherThe real limits on an upload and how to work around whichever one you hit.
- When a new version arrivesGoing furtherReplace without losing the previous one, and know which was valid when.
- Scanning a document properlyGoing furtherFive minutes of method that avoid half the reading problems.
- Deleting a document, or notGoing furtherWhat can be deleted, what should be, and what must not.
- Uploading many documents at onceGoing furtherWhat to decide before dragging in two hundred files, because undoing it afterwards costs more.
- Documents that bring more personal data than neededGoing furtherYou ask for one thing and it arrives inside a document carrying five more details you never needed.
Automatic reading
- Why it sometimes reads wrongGoing furtherWhat makes automatic reading fail, and what you can change.
- Correcting a misread valueGoing furtherHow to correct it, what happens to the original, and why it is worth doing.
- Very long documentsGoing furtherTwo hundred pages of which four values matter. How to approach it.
- Documents in another languageGoing furtherA certificate in German, an invoice in Chinese: what can be done and what should be translated.
- Handwritten documentsGoing furtherJob sheets, delivery notes and pen annotations: what can be extracted and what is better typed.
- The same value appearing twice differentlyGoing furtherThe invoice says one thing and the delivery note another. Which wins and what to do with the gap.
- Documents with tablesGoing furtherA single field almost always reads fine. A table fails elsewhere: not on the number, but on the row.
- Documents that carry no dateGoing furtherThey arrive with no date, or with three different ones. What gets recorded there drives every expiry that follows.
- Which values are worth extractingAdvancedExtracting too much creates review work and returns nothing.
- They're asking for a translationGoing furtherAsk first what kind of translation they need: the gap between the two is measured in days and money.
- How the AI reads your documents, and when to trust itWhat it extracts, what the confidence level means and who has the final word.
- Uploading documents and taking photos that read wellFormats, size limit and the four mistakes that make a photo unreadable.