Analyse a document with AI
Use the one-click Analyse with AI button to OCR and extract fields together, then review.
When a template has an AI service attached, you don't run OCR and field extraction as two separate steps — one button does both. This page shows you how to read a document with Analyse with AI, wire each index field to the value it should pull from, and review what the AI filled in before it exports.
Analyse with AI is the AI counterpart to plain OCR. If your template has no AI service, the primary button stays OCR · Current page and you capture fields from OCR zones instead — see Run OCR.
Run Analyse with AI
The primary button in the Viewer's right sidebar changes label depending on what the active template does. When an AI service is in play, it reads Analyse with AI and runs OCR followed by the AI service in a single pass.
Open the document in the Viewer
Import or scan your pages so they land in the Viewer. The primary action buttons sit at the top of the right sidebar.
Confirm an AI service is in play
The button reads Analyse with AI only when the active template has an AI service configured, or — for a document with no template — when you've picked a module from the AI dropdown above the button. Its tooltip reads Run OCR + AI extraction on this document. If you instead see OCR · Current page, no AI service is attached and the button runs plain OCR only.
Press Analyse with AI
Click it once. Scanix reads the page text with OCR, then sends it to the AI service, which fills your index fields. The button shows a busy state while it works; when it finishes, the captured values appear in the Fields panel below.
Screenshot
The Viewer right sidebar with the primary button labelled Analyse with AI above the green Process button. — shot ai-capture-analyse-with-ai-01
Analyse with AI always re-reads the page
Because the AI service needs fresh, full-page text, Analyse with AI runs OCR on the page even if it was read before. You don't need to press OCR first.
An AI provider has to be configured
AI extraction only runs once a provider is set up. If none is configured, configure one under Settings → AI Providers first — add a cloud API key or install an on-device model. Until then there's nothing for Analyse with AI to call.
Tell the AI which field is which: Mapped Fields
Extraction is only useful if each value lands in the right index field. Mapped Fields is where you wire that up — for each of your template's index fields, you choose where its value comes from. Open the Mapped Fields tab in the Template Designer; the editor is headed Index fields → source.
Each field row offers one source from its dropdown:
- Use OCR zone (unmapped) — the default. The field keeps its OCR-zone extraction; the AI doesn't fill it. Any field you leave unmapped falls back to OCR zone extraction.
- AI: <module> — pull the value from one of the AI service's extracted fields. The options are grouped per module (for example, AI: Invoices), and nested values appear as their own entries so you can map, say, an amount on its own rather than a whole object.
- From: <template> — pull the value from a field on a sibling document. This source only appears for Job Groups, where related documents are processed together; the value is read from the matching field on another member document in the same batch.
Screenshot
The Mapped Fields tab headed Index fields → source, with one row's dropdown open showing Use OCR zone (unmapped) and an AI: Invoices option group. — shot ai-capture-analyse-with-ai-02
A counter at the top of the editor reads N / N mapped so you can see how many fields are wired. When you're done, press Save mappings.
If the tab shows Pick an AI module first, the template has no AI service yet. Choose one in the template's AI Services tab, then come back to map its fields. To browse what each module extracts, see the AI Services catalogue.
Review and correct the AI-filled fields
AI extraction is fast, but you stay in control of what exports. After Analyse with AI runs, the Fields panel shows each captured value with a status, so you can scan for anything that needs a human eye:
- A green tick marks a valid field.
- An amber triangle marks a field flagged for review — typically because its confidence is low.
- A red marker means a field is invalid or missing and needs your attention.
Click any field to focus it and edit the value inline. If you fixed the underlying OCR text or want a fresh pass, the panel header offers Re-run AI to call the AI service again on the document, and Re-extract to re-pull values from the template's zones.
Screenshot
The Fields panel after Analyse with AI, showing valid (green), review (amber) and missing (red) field statuses, with Re-run AI and Re-extract in the header. — shot ai-capture-analyse-with-ai-03
The auto-accept threshold and the Verification Queue
How much review you do by hand depends on one setting: the auto-accept confidence bar. Under Settings → AI Services, the Automation card carries a single slider — Auto-accept above N% confidence — which defaults to 97%.
When a document's extracted fields all clear that bar, it exports automatically (straight-through processing). Anything below the bar is held for a person to check in the Verification Queue — a dedicated list of documents waiting on review, where you can filter by state (Review, Rejected, On Hold), sort, and open each one back in the Viewer to fix and clear it.
Screenshot
The Automation card in Settings → AI Services with the slider set to Auto-accept above 97% confidence. — shot ai-capture-analyse-with-ai-04
Raise the threshold to send more documents to review and catch more by hand; lower it to let more export untouched. The default of 97% mirrors the industry norm of roughly 97–98%.
Troubleshooting
Next steps
AI Services catalogue
Browse the prebuilt AI modules and the fields each one extracts.
Per-page field capture
How captured fields are tracked page by page across a document.
Learning from your corrections
How Scanix uses the edits you make to improve future extraction.
Related
- AI Services catalogue — the modules you map fields to and run with Analyse with AI.
- Per-page field capture — how field values are recorded for each page.
- Learning from your corrections — turn your review edits into better future captures.