AI-powered document data extraction
Company Documents includes an AI Data Extraction tool that reads a document and pulls out key details such as vendor, date, and amount, along with a short summary. It's powered by Mistral.
Running AI Extract
- In the file list, tick the checkbox next to a single document. A floating action bar appears at the bottom of the screen once at least one file is selected.
- Click AI Extract in that action bar. The button is greyed out (disabled) unless exactly one file is selected, since it analyses one document at a time.
- A toast confirms "Running AI Deep Analysis..." while it works, and the button label updates to show progress.
- When it finishes, the results open automatically in the AI Data Extraction modal.
What gets extracted, and how accurately
- Image files (JPG, PNG, GIF, WEBP, SVG, and similar): Accupe runs genuine text recognition (OCR) on the image in your browser, with a live progress indicator, so the vendor, date, and amount details come from the actual image content.
- PDFs and other non-image files: this covers most invoices and statements, but Accupe does not yet read the full text of these documents. Instead it builds a placeholder description from the file name alone, for example guessing the vendor from whatever text comes before the first underscore in the file name, so treat the results as a rough starting point and always check the vendor, date, and amount against the source document before relying on them.
Reading the results
The AI Data Extraction modal shows three summary cards. Vendor and Date each fall back to "Unknown" when no value is detected, while Amount is shown with its currency prefix and defaults to "0.00" if nothing was found. Below the cards, an Intelligent Summary card describes the document in plain language, falling back to "No summary available." if the AI didn't return one. Click Close Results when you're done.
The results are also saved to the document's record on the server, but this page doesn't currently offer a way to redisplay a past extraction. If you close the modal and want to see it again, run AI Extract on that file a second time.
If something goes wrong while analysing, for example a network problem or an OCR failure, you'll typically see a "Deep Analysis Failed" toast and can try again. If instead the AI Extract button simply returns to normal without the results modal opening, the analysis didn't complete either, try running it again.
Note: AI Data Extraction is most reliable for scanned images and photos of documents. For PDFs, treat the summary as a quick starting point and verify the vendor, date, and amount manually before using them.