When Extraction Gets Something Wrong

AI extraction is highly capable but not infallible, and Orlo is built around the assumption that you'll review what it produces rather than trust it blindly. Here's how to fix the common issues.

The document type is wrong

Open the document and use the "Document Type" dropdown on the review screen to reclassify it. The fields shown below the dropdown update automatically to match whatever type you pick, and any values you'd already entered are kept where the field names match.

A field is missing or incorrect

Every extracted field is editable before you confirm. Dates, amounts, provider names — correct anything that looks wrong. This matters especially for provider names and dates, since those drive reminder timing and renewal price tracking.

Extraction failed completely

This usually happens with heavily damaged scans, handwritten documents, or very low-resolution photos. You'll be notified if it happens. Try re-uploading a clearer version — a flat, well-lit photo or a proper scan works far better than a photo taken at an angle with a shadow across it. PDFs with selectable text (rather than a scanned image saved as a PDF) tend to give the most accurate results.

What Orlo accepts

JPEG, PNG, WebP, HEIC/HEIF images, and PDF files, up to 20MB each. There's no support for Word documents, spreadsheets, or plain text files — if you have a document in one of those formats, exporting it to PDF first will let Orlo read it.

Suggested reminders look wrong

Suggested reminders are generated from the dates and fields on the document, so if a date was misread, the reminder will be too. Correct the underlying field and re-confirm, or just dismiss the suggested reminder and add the correct one manually — see how reminder timing works for the logic behind each type.

If something consistently extracts badly, let us know — we read every message and use them to improve extraction accuracy for that document type.